cciss: Dynamically allocate struct device for each logical drive as needed.
[linux-2.6.git] / 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);
174 static int deregister_disk(ctlr_info_t *h, int drv_index,
175                            int clear_all);
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
204 #ifdef CONFIG_PROC_FS
205 static void cciss_procinit(int i);
206 #else
207 static void cciss_procinit(int i)
208 {
209 }
210 #endif                          /* CONFIG_PROC_FS */
211
212 #ifdef CONFIG_COMPAT
213 static int cciss_compat_ioctl(struct block_device *, fmode_t,
214                               unsigned, unsigned long);
215 #endif
216
217 static const struct block_device_operations cciss_fops = {
218         .owner = THIS_MODULE,
219         .open = cciss_open,
220         .release = cciss_release,
221         .locked_ioctl = cciss_ioctl,
222         .getgeo = cciss_getgeo,
223 #ifdef CONFIG_COMPAT
224         .compat_ioctl = cciss_compat_ioctl,
225 #endif
226         .revalidate_disk = cciss_revalidate,
227 };
228
229 /*
230  * Enqueuing and dequeuing functions for cmdlists.
231  */
232 static inline void addQ(struct hlist_head *list, CommandList_struct *c)
233 {
234         hlist_add_head(&c->list, list);
235 }
236
237 static inline void removeQ(CommandList_struct *c)
238 {
239         /*
240          * After kexec/dump some commands might still
241          * be in flight, which the firmware will try
242          * to complete. Resetting the firmware doesn't work
243          * with old fw revisions, so we have to mark
244          * them off as 'stale' to prevent the driver from
245          * falling over.
246          */
247         if (WARN_ON(hlist_unhashed(&c->list))) {
248                 c->cmd_type = CMD_MSG_STALE;
249                 return;
250         }
251
252         hlist_del_init(&c->list);
253 }
254
255 #include "cciss_scsi.c"         /* For SCSI tape support */
256
257 #define RAID_UNKNOWN 6
258
259 #ifdef CONFIG_PROC_FS
260
261 /*
262  * Report information about this controller.
263  */
264 #define ENG_GIG 1000000000
265 #define ENG_GIG_FACTOR (ENG_GIG/512)
266 #define ENGAGE_SCSI     "engage scsi"
267 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
268         "UNKNOWN"
269 };
270
271 static struct proc_dir_entry *proc_cciss;
272
273 static void cciss_seq_show_header(struct seq_file *seq)
274 {
275         ctlr_info_t *h = seq->private;
276
277         seq_printf(seq, "%s: HP %s Controller\n"
278                 "Board ID: 0x%08lx\n"
279                 "Firmware Version: %c%c%c%c\n"
280                 "IRQ: %d\n"
281                 "Logical drives: %d\n"
282                 "Current Q depth: %d\n"
283                 "Current # commands on controller: %d\n"
284                 "Max Q depth since init: %d\n"
285                 "Max # commands on controller since init: %d\n"
286                 "Max SG entries since init: %d\n",
287                 h->devname,
288                 h->product_name,
289                 (unsigned long)h->board_id,
290                 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
291                 h->firm_ver[3], (unsigned int)h->intr[SIMPLE_MODE_INT],
292                 h->num_luns,
293                 h->Qdepth, h->commands_outstanding,
294                 h->maxQsinceinit, h->max_outstanding, h->maxSG);
295
296 #ifdef CONFIG_CISS_SCSI_TAPE
297         cciss_seq_tape_report(seq, h->ctlr);
298 #endif /* CONFIG_CISS_SCSI_TAPE */
299 }
300
301 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
302 {
303         ctlr_info_t *h = seq->private;
304         unsigned ctlr = h->ctlr;
305         unsigned long flags;
306
307         /* prevent displaying bogus info during configuration
308          * or deconfiguration of a logical volume
309          */
310         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
311         if (h->busy_configuring) {
312                 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
313                 return ERR_PTR(-EBUSY);
314         }
315         h->busy_configuring = 1;
316         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
317
318         if (*pos == 0)
319                 cciss_seq_show_header(seq);
320
321         return pos;
322 }
323
324 static int cciss_seq_show(struct seq_file *seq, void *v)
325 {
326         sector_t vol_sz, vol_sz_frac;
327         ctlr_info_t *h = seq->private;
328         unsigned ctlr = h->ctlr;
329         loff_t *pos = v;
330         drive_info_struct *drv = &h->drv[*pos];
331
332         if (*pos > h->highest_lun)
333                 return 0;
334
335         if (drv->heads == 0)
336                 return 0;
337
338         vol_sz = drv->nr_blocks;
339         vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
340         vol_sz_frac *= 100;
341         sector_div(vol_sz_frac, ENG_GIG_FACTOR);
342
343         if (drv->raid_level > 5)
344                 drv->raid_level = RAID_UNKNOWN;
345         seq_printf(seq, "cciss/c%dd%d:"
346                         "\t%4u.%02uGB\tRAID %s\n",
347                         ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
348                         raid_label[drv->raid_level]);
349         return 0;
350 }
351
352 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
353 {
354         ctlr_info_t *h = seq->private;
355
356         if (*pos > h->highest_lun)
357                 return NULL;
358         *pos += 1;
359
360         return pos;
361 }
362
363 static void cciss_seq_stop(struct seq_file *seq, void *v)
364 {
365         ctlr_info_t *h = seq->private;
366
367         /* Only reset h->busy_configuring if we succeeded in setting
368          * it during cciss_seq_start. */
369         if (v == ERR_PTR(-EBUSY))
370                 return;
371
372         h->busy_configuring = 0;
373 }
374
375 static const struct seq_operations cciss_seq_ops = {
376         .start = cciss_seq_start,
377         .show  = cciss_seq_show,
378         .next  = cciss_seq_next,
379         .stop  = cciss_seq_stop,
380 };
381
382 static int cciss_seq_open(struct inode *inode, struct file *file)
383 {
384         int ret = seq_open(file, &cciss_seq_ops);
385         struct seq_file *seq = file->private_data;
386
387         if (!ret)
388                 seq->private = PDE(inode)->data;
389
390         return ret;
391 }
392
393 static ssize_t
394 cciss_proc_write(struct file *file, const char __user *buf,
395                  size_t length, loff_t *ppos)
396 {
397         int err;
398         char *buffer;
399
400 #ifndef CONFIG_CISS_SCSI_TAPE
401         return -EINVAL;
402 #endif
403
404         if (!buf || length > PAGE_SIZE - 1)
405                 return -EINVAL;
406
407         buffer = (char *)__get_free_page(GFP_KERNEL);
408         if (!buffer)
409                 return -ENOMEM;
410
411         err = -EFAULT;
412         if (copy_from_user(buffer, buf, length))
413                 goto out;
414         buffer[length] = '\0';
415
416 #ifdef CONFIG_CISS_SCSI_TAPE
417         if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
418                 struct seq_file *seq = file->private_data;
419                 ctlr_info_t *h = seq->private;
420                 int rc;
421
422                 rc = cciss_engage_scsi(h->ctlr);
423                 if (rc != 0)
424                         err = -rc;
425                 else
426                         err = length;
427         } else
428 #endif /* CONFIG_CISS_SCSI_TAPE */
429                 err = -EINVAL;
430         /* might be nice to have "disengage" too, but it's not
431            safely possible. (only 1 module use count, lock issues.) */
432
433 out:
434         free_page((unsigned long)buffer);
435         return err;
436 }
437
438 static struct file_operations cciss_proc_fops = {
439         .owner   = THIS_MODULE,
440         .open    = cciss_seq_open,
441         .read    = seq_read,
442         .llseek  = seq_lseek,
443         .release = seq_release,
444         .write   = cciss_proc_write,
445 };
446
447 static void __devinit cciss_procinit(int i)
448 {
449         struct proc_dir_entry *pde;
450
451         if (proc_cciss == NULL)
452                 proc_cciss = proc_mkdir("driver/cciss", NULL);
453         if (!proc_cciss)
454                 return;
455         pde = proc_create_data(hba[i]->devname, S_IWUSR | S_IRUSR | S_IRGRP |
456                                         S_IROTH, proc_cciss,
457                                         &cciss_proc_fops, hba[i]);
458 }
459 #endif                          /* CONFIG_PROC_FS */
460
461 #define MAX_PRODUCT_NAME_LEN 19
462
463 #define to_hba(n) container_of(n, struct ctlr_info, dev)
464
465 static ssize_t host_store_rescan(struct device *dev,
466                                  struct device_attribute *attr,
467                                  const char *buf, size_t count)
468 {
469         struct ctlr_info *h = to_hba(dev);
470
471         add_to_scan_list(h);
472         wake_up_process(cciss_scan_thread);
473         wait_for_completion_interruptible(&h->scan_wait);
474
475         return count;
476 }
477 DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
478
479 static ssize_t dev_show_unique_id(struct device *dev,
480                                  struct device_attribute *attr,
481                                  char *buf)
482 {
483         drive_info_struct *drv = dev_get_drvdata(dev);
484         struct ctlr_info *h = to_hba(drv->dev->parent);
485         __u8 sn[16];
486         unsigned long flags;
487         int ret = 0;
488
489         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
490         if (h->busy_configuring)
491                 ret = -EBUSY;
492         else
493                 memcpy(sn, drv->serial_no, sizeof(sn));
494         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
495
496         if (ret)
497                 return ret;
498         else
499                 return snprintf(buf, 16 * 2 + 2,
500                                 "%02X%02X%02X%02X%02X%02X%02X%02X"
501                                 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
502                                 sn[0], sn[1], sn[2], sn[3],
503                                 sn[4], sn[5], sn[6], sn[7],
504                                 sn[8], sn[9], sn[10], sn[11],
505                                 sn[12], sn[13], sn[14], sn[15]);
506 }
507 DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
508
509 static ssize_t dev_show_vendor(struct device *dev,
510                                struct device_attribute *attr,
511                                char *buf)
512 {
513         drive_info_struct *drv = dev_get_drvdata(dev);
514         struct ctlr_info *h = to_hba(drv->dev->parent);
515         char vendor[VENDOR_LEN + 1];
516         unsigned long flags;
517         int ret = 0;
518
519         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
520         if (h->busy_configuring)
521                 ret = -EBUSY;
522         else
523                 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
524         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
525
526         if (ret)
527                 return ret;
528         else
529                 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
530 }
531 DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
532
533 static ssize_t dev_show_model(struct device *dev,
534                               struct device_attribute *attr,
535                               char *buf)
536 {
537         drive_info_struct *drv = dev_get_drvdata(dev);
538         struct ctlr_info *h = to_hba(drv->dev->parent);
539         char model[MODEL_LEN + 1];
540         unsigned long flags;
541         int ret = 0;
542
543         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
544         if (h->busy_configuring)
545                 ret = -EBUSY;
546         else
547                 memcpy(model, drv->model, MODEL_LEN + 1);
548         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
549
550         if (ret)
551                 return ret;
552         else
553                 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
554 }
555 DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
556
557 static ssize_t dev_show_rev(struct device *dev,
558                             struct device_attribute *attr,
559                             char *buf)
560 {
561         drive_info_struct *drv = dev_get_drvdata(dev);
562         struct ctlr_info *h = to_hba(drv->dev->parent);
563         char rev[REV_LEN + 1];
564         unsigned long flags;
565         int ret = 0;
566
567         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
568         if (h->busy_configuring)
569                 ret = -EBUSY;
570         else
571                 memcpy(rev, drv->rev, REV_LEN + 1);
572         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
573
574         if (ret)
575                 return ret;
576         else
577                 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
578 }
579 DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
580
581 static struct attribute *cciss_host_attrs[] = {
582         &dev_attr_rescan.attr,
583         NULL
584 };
585
586 static struct attribute_group cciss_host_attr_group = {
587         .attrs = cciss_host_attrs,
588 };
589
590 static struct attribute_group *cciss_host_attr_groups[] = {
591         &cciss_host_attr_group,
592         NULL
593 };
594
595 static struct device_type cciss_host_type = {
596         .name           = "cciss_host",
597         .groups         = cciss_host_attr_groups,
598         .release        = cciss_hba_release,
599 };
600
601 static struct attribute *cciss_dev_attrs[] = {
602         &dev_attr_unique_id.attr,
603         &dev_attr_model.attr,
604         &dev_attr_vendor.attr,
605         &dev_attr_rev.attr,
606         NULL
607 };
608
609 static struct attribute_group cciss_dev_attr_group = {
610         .attrs = cciss_dev_attrs,
611 };
612
613 static const struct attribute_group *cciss_dev_attr_groups[] = {
614         &cciss_dev_attr_group,
615         NULL
616 };
617
618 static struct device_type cciss_dev_type = {
619         .name           = "cciss_device",
620         .groups         = cciss_dev_attr_groups,
621         .release        = cciss_device_release,
622 };
623
624 static struct bus_type cciss_bus_type = {
625         .name           = "cciss",
626 };
627
628 /*
629  * cciss_hba_release is called when the reference count
630  * of h->dev goes to zero.
631  */
632 static void cciss_hba_release(struct device *dev)
633 {
634         /*
635          * nothing to do, but need this to avoid a warning
636          * about not having a release handler from lib/kref.c.
637          */
638 }
639
640 /*
641  * Initialize sysfs entry for each controller.  This sets up and registers
642  * the 'cciss#' directory for each individual controller under
643  * /sys/bus/pci/devices/<dev>/.
644  */
645 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
646 {
647         device_initialize(&h->dev);
648         h->dev.type = &cciss_host_type;
649         h->dev.bus = &cciss_bus_type;
650         dev_set_name(&h->dev, "%s", h->devname);
651         h->dev.parent = &h->pdev->dev;
652
653         return device_add(&h->dev);
654 }
655
656 /*
657  * Remove sysfs entries for an hba.
658  */
659 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
660 {
661         device_del(&h->dev);
662         put_device(&h->dev); /* final put. */
663 }
664
665 /* cciss_device_release is called when the reference count
666  * of h->drv[x].dev goes to zero.
667  */
668 static void cciss_device_release(struct device *dev)
669 {
670         kfree(dev);
671 }
672
673 /*
674  * Initialize sysfs for each logical drive.  This sets up and registers
675  * the 'c#d#' directory for each individual logical drive under
676  * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
677  * /sys/block/cciss!c#d# to this entry.
678  */
679 static long cciss_create_ld_sysfs_entry(struct ctlr_info *h,
680                                        int drv_index)
681 {
682         struct device *dev;
683
684         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
685         if (!dev)
686                 return -ENOMEM;
687         device_initialize(dev);
688         dev->type = &cciss_dev_type;
689         dev->bus = &cciss_bus_type;
690         dev_set_name(dev, "c%dd%d", h->ctlr, drv_index);
691         dev->parent = &h->dev;
692         h->drv[drv_index].dev = dev;
693         dev_set_drvdata(dev, &h->drv[drv_index]);
694         return device_add(dev);
695 }
696
697 /*
698  * Remove sysfs entries for a logical drive.
699  */
700 static void cciss_destroy_ld_sysfs_entry(struct ctlr_info *h, int drv_index)
701 {
702         struct device *dev = h->drv[drv_index].dev;
703         device_del(dev);
704         put_device(dev); /* the "final" put. */
705         h->drv[drv_index].dev = NULL;
706 }
707
708 /*
709  * For operations that cannot sleep, a command block is allocated at init,
710  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
711  * which ones are free or in use.  For operations that can wait for kmalloc
712  * to possible sleep, this routine can be called with get_from_pool set to 0.
713  * cmd_free() MUST be called with a got_from_pool set to 0 if cmd_alloc was.
714  */
715 static CommandList_struct *cmd_alloc(ctlr_info_t *h, int get_from_pool)
716 {
717         CommandList_struct *c;
718         int i;
719         u64bit temp64;
720         dma_addr_t cmd_dma_handle, err_dma_handle;
721
722         if (!get_from_pool) {
723                 c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
724                         sizeof(CommandList_struct), &cmd_dma_handle);
725                 if (c == NULL)
726                         return NULL;
727                 memset(c, 0, sizeof(CommandList_struct));
728
729                 c->cmdindex = -1;
730
731                 c->err_info = (ErrorInfo_struct *)
732                     pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
733                             &err_dma_handle);
734
735                 if (c->err_info == NULL) {
736                         pci_free_consistent(h->pdev,
737                                 sizeof(CommandList_struct), c, cmd_dma_handle);
738                         return NULL;
739                 }
740                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
741         } else {                /* get it out of the controllers pool */
742
743                 do {
744                         i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
745                         if (i == h->nr_cmds)
746                                 return NULL;
747                 } while (test_and_set_bit
748                          (i & (BITS_PER_LONG - 1),
749                           h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
750 #ifdef CCISS_DEBUG
751                 printk(KERN_DEBUG "cciss: using command buffer %d\n", i);
752 #endif
753                 c = h->cmd_pool + i;
754                 memset(c, 0, sizeof(CommandList_struct));
755                 cmd_dma_handle = h->cmd_pool_dhandle
756                     + i * sizeof(CommandList_struct);
757                 c->err_info = h->errinfo_pool + i;
758                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
759                 err_dma_handle = h->errinfo_pool_dhandle
760                     + i * sizeof(ErrorInfo_struct);
761                 h->nr_allocs++;
762
763                 c->cmdindex = i;
764         }
765
766         INIT_HLIST_NODE(&c->list);
767         c->busaddr = (__u32) cmd_dma_handle;
768         temp64.val = (__u64) err_dma_handle;
769         c->ErrDesc.Addr.lower = temp64.val32.lower;
770         c->ErrDesc.Addr.upper = temp64.val32.upper;
771         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
772
773         c->ctlr = h->ctlr;
774         return c;
775 }
776
777 /*
778  * Frees a command block that was previously allocated with cmd_alloc().
779  */
780 static void cmd_free(ctlr_info_t *h, CommandList_struct *c, int got_from_pool)
781 {
782         int i;
783         u64bit temp64;
784
785         if (!got_from_pool) {
786                 temp64.val32.lower = c->ErrDesc.Addr.lower;
787                 temp64.val32.upper = c->ErrDesc.Addr.upper;
788                 pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
789                                     c->err_info, (dma_addr_t) temp64.val);
790                 pci_free_consistent(h->pdev, sizeof(CommandList_struct),
791                                     c, (dma_addr_t) c->busaddr);
792         } else {
793                 i = c - h->cmd_pool;
794                 clear_bit(i & (BITS_PER_LONG - 1),
795                           h->cmd_pool_bits + (i / BITS_PER_LONG));
796                 h->nr_frees++;
797         }
798 }
799
800 static inline ctlr_info_t *get_host(struct gendisk *disk)
801 {
802         return disk->queue->queuedata;
803 }
804
805 static inline drive_info_struct *get_drv(struct gendisk *disk)
806 {
807         return disk->private_data;
808 }
809
810 /*
811  * Open.  Make sure the device is really there.
812  */
813 static int cciss_open(struct block_device *bdev, fmode_t mode)
814 {
815         ctlr_info_t *host = get_host(bdev->bd_disk);
816         drive_info_struct *drv = get_drv(bdev->bd_disk);
817
818 #ifdef CCISS_DEBUG
819         printk(KERN_DEBUG "cciss_open %s\n", bdev->bd_disk->disk_name);
820 #endif                          /* CCISS_DEBUG */
821
822         if (host->busy_initializing || drv->busy_configuring)
823                 return -EBUSY;
824         /*
825          * Root is allowed to open raw volume zero even if it's not configured
826          * so array config can still work. Root is also allowed to open any
827          * volume that has a LUN ID, so it can issue IOCTL to reread the
828          * disk information.  I don't think I really like this
829          * but I'm already using way to many device nodes to claim another one
830          * for "raw controller".
831          */
832         if (drv->heads == 0) {
833                 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
834                         /* if not node 0 make sure it is a partition = 0 */
835                         if (MINOR(bdev->bd_dev) & 0x0f) {
836                                 return -ENXIO;
837                                 /* if it is, make sure we have a LUN ID */
838                         } else if (drv->LunID == 0) {
839                                 return -ENXIO;
840                         }
841                 }
842                 if (!capable(CAP_SYS_ADMIN))
843                         return -EPERM;
844         }
845         drv->usage_count++;
846         host->usage_count++;
847         return 0;
848 }
849
850 /*
851  * Close.  Sync first.
852  */
853 static int cciss_release(struct gendisk *disk, fmode_t mode)
854 {
855         ctlr_info_t *host = get_host(disk);
856         drive_info_struct *drv = get_drv(disk);
857
858 #ifdef CCISS_DEBUG
859         printk(KERN_DEBUG "cciss_release %s\n", disk->disk_name);
860 #endif                          /* CCISS_DEBUG */
861
862         drv->usage_count--;
863         host->usage_count--;
864         return 0;
865 }
866
867 #ifdef CONFIG_COMPAT
868
869 static int do_ioctl(struct block_device *bdev, fmode_t mode,
870                     unsigned cmd, unsigned long arg)
871 {
872         int ret;
873         lock_kernel();
874         ret = cciss_ioctl(bdev, mode, cmd, arg);
875         unlock_kernel();
876         return ret;
877 }
878
879 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
880                                   unsigned cmd, unsigned long arg);
881 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
882                                       unsigned cmd, unsigned long arg);
883
884 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
885                               unsigned cmd, unsigned long arg)
886 {
887         switch (cmd) {
888         case CCISS_GETPCIINFO:
889         case CCISS_GETINTINFO:
890         case CCISS_SETINTINFO:
891         case CCISS_GETNODENAME:
892         case CCISS_SETNODENAME:
893         case CCISS_GETHEARTBEAT:
894         case CCISS_GETBUSTYPES:
895         case CCISS_GETFIRMVER:
896         case CCISS_GETDRIVVER:
897         case CCISS_REVALIDVOLS:
898         case CCISS_DEREGDISK:
899         case CCISS_REGNEWDISK:
900         case CCISS_REGNEWD:
901         case CCISS_RESCANDISK:
902         case CCISS_GETLUNINFO:
903                 return do_ioctl(bdev, mode, cmd, arg);
904
905         case CCISS_PASSTHRU32:
906                 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
907         case CCISS_BIG_PASSTHRU32:
908                 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
909
910         default:
911                 return -ENOIOCTLCMD;
912         }
913 }
914
915 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
916                                   unsigned cmd, unsigned long arg)
917 {
918         IOCTL32_Command_struct __user *arg32 =
919             (IOCTL32_Command_struct __user *) arg;
920         IOCTL_Command_struct arg64;
921         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
922         int err;
923         u32 cp;
924
925         err = 0;
926         err |=
927             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
928                            sizeof(arg64.LUN_info));
929         err |=
930             copy_from_user(&arg64.Request, &arg32->Request,
931                            sizeof(arg64.Request));
932         err |=
933             copy_from_user(&arg64.error_info, &arg32->error_info,
934                            sizeof(arg64.error_info));
935         err |= get_user(arg64.buf_size, &arg32->buf_size);
936         err |= get_user(cp, &arg32->buf);
937         arg64.buf = compat_ptr(cp);
938         err |= copy_to_user(p, &arg64, sizeof(arg64));
939
940         if (err)
941                 return -EFAULT;
942
943         err = do_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
944         if (err)
945                 return err;
946         err |=
947             copy_in_user(&arg32->error_info, &p->error_info,
948                          sizeof(arg32->error_info));
949         if (err)
950                 return -EFAULT;
951         return err;
952 }
953
954 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
955                                       unsigned cmd, unsigned long arg)
956 {
957         BIG_IOCTL32_Command_struct __user *arg32 =
958             (BIG_IOCTL32_Command_struct __user *) arg;
959         BIG_IOCTL_Command_struct arg64;
960         BIG_IOCTL_Command_struct __user *p =
961             compat_alloc_user_space(sizeof(arg64));
962         int err;
963         u32 cp;
964
965         err = 0;
966         err |=
967             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
968                            sizeof(arg64.LUN_info));
969         err |=
970             copy_from_user(&arg64.Request, &arg32->Request,
971                            sizeof(arg64.Request));
972         err |=
973             copy_from_user(&arg64.error_info, &arg32->error_info,
974                            sizeof(arg64.error_info));
975         err |= get_user(arg64.buf_size, &arg32->buf_size);
976         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
977         err |= get_user(cp, &arg32->buf);
978         arg64.buf = compat_ptr(cp);
979         err |= copy_to_user(p, &arg64, sizeof(arg64));
980
981         if (err)
982                 return -EFAULT;
983
984         err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
985         if (err)
986                 return err;
987         err |=
988             copy_in_user(&arg32->error_info, &p->error_info,
989                          sizeof(arg32->error_info));
990         if (err)
991                 return -EFAULT;
992         return err;
993 }
994 #endif
995
996 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
997 {
998         drive_info_struct *drv = get_drv(bdev->bd_disk);
999
1000         if (!drv->cylinders)
1001                 return -ENXIO;
1002
1003         geo->heads = drv->heads;
1004         geo->sectors = drv->sectors;
1005         geo->cylinders = drv->cylinders;
1006         return 0;
1007 }
1008
1009 static void check_ioctl_unit_attention(ctlr_info_t *host, CommandList_struct *c)
1010 {
1011         if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1012                         c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1013                 (void)check_for_unit_attention(host, c);
1014 }
1015 /*
1016  * ioctl
1017  */
1018 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1019                        unsigned int cmd, unsigned long arg)
1020 {
1021         struct gendisk *disk = bdev->bd_disk;
1022         ctlr_info_t *host = get_host(disk);
1023         drive_info_struct *drv = get_drv(disk);
1024         int ctlr = host->ctlr;
1025         void __user *argp = (void __user *)arg;
1026
1027 #ifdef CCISS_DEBUG
1028         printk(KERN_DEBUG "cciss_ioctl: Called with cmd=%x %lx\n", cmd, arg);
1029 #endif                          /* CCISS_DEBUG */
1030
1031         switch (cmd) {
1032         case CCISS_GETPCIINFO:
1033                 {
1034                         cciss_pci_info_struct pciinfo;
1035
1036                         if (!arg)
1037                                 return -EINVAL;
1038                         pciinfo.domain = pci_domain_nr(host->pdev->bus);
1039                         pciinfo.bus = host->pdev->bus->number;
1040                         pciinfo.dev_fn = host->pdev->devfn;
1041                         pciinfo.board_id = host->board_id;
1042                         if (copy_to_user
1043                             (argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1044                                 return -EFAULT;
1045                         return 0;
1046                 }
1047         case CCISS_GETINTINFO:
1048                 {
1049                         cciss_coalint_struct intinfo;
1050                         if (!arg)
1051                                 return -EINVAL;
1052                         intinfo.delay =
1053                             readl(&host->cfgtable->HostWrite.CoalIntDelay);
1054                         intinfo.count =
1055                             readl(&host->cfgtable->HostWrite.CoalIntCount);
1056                         if (copy_to_user
1057                             (argp, &intinfo, sizeof(cciss_coalint_struct)))
1058                                 return -EFAULT;
1059                         return 0;
1060                 }
1061         case CCISS_SETINTINFO:
1062                 {
1063                         cciss_coalint_struct intinfo;
1064                         unsigned long flags;
1065                         int i;
1066
1067                         if (!arg)
1068                                 return -EINVAL;
1069                         if (!capable(CAP_SYS_ADMIN))
1070                                 return -EPERM;
1071                         if (copy_from_user
1072                             (&intinfo, argp, sizeof(cciss_coalint_struct)))
1073                                 return -EFAULT;
1074                         if ((intinfo.delay == 0) && (intinfo.count == 0))
1075                         {
1076 //                      printk("cciss_ioctl: delay and count cannot be 0\n");
1077                                 return -EINVAL;
1078                         }
1079                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1080                         /* Update the field, and then ring the doorbell */
1081                         writel(intinfo.delay,
1082                                &(host->cfgtable->HostWrite.CoalIntDelay));
1083                         writel(intinfo.count,
1084                                &(host->cfgtable->HostWrite.CoalIntCount));
1085                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1086
1087                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1088                                 if (!(readl(host->vaddr + SA5_DOORBELL)
1089                                       & CFGTBL_ChangeReq))
1090                                         break;
1091                                 /* delay and try again */
1092                                 udelay(1000);
1093                         }
1094                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1095                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1096                                 return -EAGAIN;
1097                         return 0;
1098                 }
1099         case CCISS_GETNODENAME:
1100                 {
1101                         NodeName_type NodeName;
1102                         int i;
1103
1104                         if (!arg)
1105                                 return -EINVAL;
1106                         for (i = 0; i < 16; i++)
1107                                 NodeName[i] =
1108                                     readb(&host->cfgtable->ServerName[i]);
1109                         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1110                                 return -EFAULT;
1111                         return 0;
1112                 }
1113         case CCISS_SETNODENAME:
1114                 {
1115                         NodeName_type NodeName;
1116                         unsigned long flags;
1117                         int i;
1118
1119                         if (!arg)
1120                                 return -EINVAL;
1121                         if (!capable(CAP_SYS_ADMIN))
1122                                 return -EPERM;
1123
1124                         if (copy_from_user
1125                             (NodeName, argp, sizeof(NodeName_type)))
1126                                 return -EFAULT;
1127
1128                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1129
1130                         /* Update the field, and then ring the doorbell */
1131                         for (i = 0; i < 16; i++)
1132                                 writeb(NodeName[i],
1133                                        &host->cfgtable->ServerName[i]);
1134
1135                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1136
1137                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1138                                 if (!(readl(host->vaddr + SA5_DOORBELL)
1139                                       & CFGTBL_ChangeReq))
1140                                         break;
1141                                 /* delay and try again */
1142                                 udelay(1000);
1143                         }
1144                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1145                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1146                                 return -EAGAIN;
1147                         return 0;
1148                 }
1149
1150         case CCISS_GETHEARTBEAT:
1151                 {
1152                         Heartbeat_type heartbeat;
1153
1154                         if (!arg)
1155                                 return -EINVAL;
1156                         heartbeat = readl(&host->cfgtable->HeartBeat);
1157                         if (copy_to_user
1158                             (argp, &heartbeat, sizeof(Heartbeat_type)))
1159                                 return -EFAULT;
1160                         return 0;
1161                 }
1162         case CCISS_GETBUSTYPES:
1163                 {
1164                         BusTypes_type BusTypes;
1165
1166                         if (!arg)
1167                                 return -EINVAL;
1168                         BusTypes = readl(&host->cfgtable->BusTypes);
1169                         if (copy_to_user
1170                             (argp, &BusTypes, sizeof(BusTypes_type)))
1171                                 return -EFAULT;
1172                         return 0;
1173                 }
1174         case CCISS_GETFIRMVER:
1175                 {
1176                         FirmwareVer_type firmware;
1177
1178                         if (!arg)
1179                                 return -EINVAL;
1180                         memcpy(firmware, host->firm_ver, 4);
1181
1182                         if (copy_to_user
1183                             (argp, firmware, sizeof(FirmwareVer_type)))
1184                                 return -EFAULT;
1185                         return 0;
1186                 }
1187         case CCISS_GETDRIVVER:
1188                 {
1189                         DriverVer_type DriverVer = DRIVER_VERSION;
1190
1191                         if (!arg)
1192                                 return -EINVAL;
1193
1194                         if (copy_to_user
1195                             (argp, &DriverVer, sizeof(DriverVer_type)))
1196                                 return -EFAULT;
1197                         return 0;
1198                 }
1199
1200         case CCISS_DEREGDISK:
1201         case CCISS_REGNEWD:
1202         case CCISS_REVALIDVOLS:
1203                 return rebuild_lun_table(host, 0);
1204
1205         case CCISS_GETLUNINFO:{
1206                         LogvolInfo_struct luninfo;
1207
1208                         luninfo.LunID = drv->LunID;
1209                         luninfo.num_opens = drv->usage_count;
1210                         luninfo.num_parts = 0;
1211                         if (copy_to_user(argp, &luninfo,
1212                                          sizeof(LogvolInfo_struct)))
1213                                 return -EFAULT;
1214                         return 0;
1215                 }
1216         case CCISS_PASSTHRU:
1217                 {
1218                         IOCTL_Command_struct iocommand;
1219                         CommandList_struct *c;
1220                         char *buff = NULL;
1221                         u64bit temp64;
1222                         unsigned long flags;
1223                         DECLARE_COMPLETION_ONSTACK(wait);
1224
1225                         if (!arg)
1226                                 return -EINVAL;
1227
1228                         if (!capable(CAP_SYS_RAWIO))
1229                                 return -EPERM;
1230
1231                         if (copy_from_user
1232                             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1233                                 return -EFAULT;
1234                         if ((iocommand.buf_size < 1) &&
1235                             (iocommand.Request.Type.Direction != XFER_NONE)) {
1236                                 return -EINVAL;
1237                         }
1238 #if 0                           /* 'buf_size' member is 16-bits, and always smaller than kmalloc limit */
1239                         /* Check kmalloc limits */
1240                         if (iocommand.buf_size > 128000)
1241                                 return -EINVAL;
1242 #endif
1243                         if (iocommand.buf_size > 0) {
1244                                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1245                                 if (buff == NULL)
1246                                         return -EFAULT;
1247                         }
1248                         if (iocommand.Request.Type.Direction == XFER_WRITE) {
1249                                 /* Copy the data into the buffer we created */
1250                                 if (copy_from_user
1251                                     (buff, iocommand.buf, iocommand.buf_size)) {
1252                                         kfree(buff);
1253                                         return -EFAULT;
1254                                 }
1255                         } else {
1256                                 memset(buff, 0, iocommand.buf_size);
1257                         }
1258                         if ((c = cmd_alloc(host, 0)) == NULL) {
1259                                 kfree(buff);
1260                                 return -ENOMEM;
1261                         }
1262                         // Fill in the command type
1263                         c->cmd_type = CMD_IOCTL_PEND;
1264                         // Fill in Command Header
1265                         c->Header.ReplyQueue = 0;       // unused in simple mode
1266                         if (iocommand.buf_size > 0)     // buffer to fill
1267                         {
1268                                 c->Header.SGList = 1;
1269                                 c->Header.SGTotal = 1;
1270                         } else  // no buffers to fill
1271                         {
1272                                 c->Header.SGList = 0;
1273                                 c->Header.SGTotal = 0;
1274                         }
1275                         c->Header.LUN = iocommand.LUN_info;
1276                         c->Header.Tag.lower = c->busaddr;       // use the kernel address the cmd block for tag
1277
1278                         // Fill in Request block
1279                         c->Request = iocommand.Request;
1280
1281                         // Fill in the scatter gather information
1282                         if (iocommand.buf_size > 0) {
1283                                 temp64.val = pci_map_single(host->pdev, buff,
1284                                         iocommand.buf_size,
1285                                         PCI_DMA_BIDIRECTIONAL);
1286                                 c->SG[0].Addr.lower = temp64.val32.lower;
1287                                 c->SG[0].Addr.upper = temp64.val32.upper;
1288                                 c->SG[0].Len = iocommand.buf_size;
1289                                 c->SG[0].Ext = 0;       // we are not chaining
1290                         }
1291                         c->waiting = &wait;
1292
1293                         /* Put the request on the tail of the request queue */
1294                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1295                         addQ(&host->reqQ, c);
1296                         host->Qdepth++;
1297                         start_io(host);
1298                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1299
1300                         wait_for_completion(&wait);
1301
1302                         /* unlock the buffers from DMA */
1303                         temp64.val32.lower = c->SG[0].Addr.lower;
1304                         temp64.val32.upper = c->SG[0].Addr.upper;
1305                         pci_unmap_single(host->pdev, (dma_addr_t) temp64.val,
1306                                          iocommand.buf_size,
1307                                          PCI_DMA_BIDIRECTIONAL);
1308
1309                         check_ioctl_unit_attention(host, c);
1310
1311                         /* Copy the error information out */
1312                         iocommand.error_info = *(c->err_info);
1313                         if (copy_to_user
1314                             (argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1315                                 kfree(buff);
1316                                 cmd_free(host, c, 0);
1317                                 return -EFAULT;
1318                         }
1319
1320                         if (iocommand.Request.Type.Direction == XFER_READ) {
1321                                 /* Copy the data out of the buffer we created */
1322                                 if (copy_to_user
1323                                     (iocommand.buf, buff, iocommand.buf_size)) {
1324                                         kfree(buff);
1325                                         cmd_free(host, c, 0);
1326                                         return -EFAULT;
1327                                 }
1328                         }
1329                         kfree(buff);
1330                         cmd_free(host, c, 0);
1331                         return 0;
1332                 }
1333         case CCISS_BIG_PASSTHRU:{
1334                         BIG_IOCTL_Command_struct *ioc;
1335                         CommandList_struct *c;
1336                         unsigned char **buff = NULL;
1337                         int *buff_size = NULL;
1338                         u64bit temp64;
1339                         unsigned long flags;
1340                         BYTE sg_used = 0;
1341                         int status = 0;
1342                         int i;
1343                         DECLARE_COMPLETION_ONSTACK(wait);
1344                         __u32 left;
1345                         __u32 sz;
1346                         BYTE __user *data_ptr;
1347
1348                         if (!arg)
1349                                 return -EINVAL;
1350                         if (!capable(CAP_SYS_RAWIO))
1351                                 return -EPERM;
1352                         ioc = (BIG_IOCTL_Command_struct *)
1353                             kmalloc(sizeof(*ioc), GFP_KERNEL);
1354                         if (!ioc) {
1355                                 status = -ENOMEM;
1356                                 goto cleanup1;
1357                         }
1358                         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1359                                 status = -EFAULT;
1360                                 goto cleanup1;
1361                         }
1362                         if ((ioc->buf_size < 1) &&
1363                             (ioc->Request.Type.Direction != XFER_NONE)) {
1364                                 status = -EINVAL;
1365                                 goto cleanup1;
1366                         }
1367                         /* Check kmalloc limits  using all SGs */
1368                         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1369                                 status = -EINVAL;
1370                                 goto cleanup1;
1371                         }
1372                         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1373                                 status = -EINVAL;
1374                                 goto cleanup1;
1375                         }
1376                         buff =
1377                             kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1378                         if (!buff) {
1379                                 status = -ENOMEM;
1380                                 goto cleanup1;
1381                         }
1382                         buff_size = kmalloc(MAXSGENTRIES * sizeof(int),
1383                                                    GFP_KERNEL);
1384                         if (!buff_size) {
1385                                 status = -ENOMEM;
1386                                 goto cleanup1;
1387                         }
1388                         left = ioc->buf_size;
1389                         data_ptr = ioc->buf;
1390                         while (left) {
1391                                 sz = (left >
1392                                       ioc->malloc_size) ? ioc->
1393                                     malloc_size : left;
1394                                 buff_size[sg_used] = sz;
1395                                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1396                                 if (buff[sg_used] == NULL) {
1397                                         status = -ENOMEM;
1398                                         goto cleanup1;
1399                                 }
1400                                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1401                                         if (copy_from_user
1402                                             (buff[sg_used], data_ptr, sz)) {
1403                                                 status = -EFAULT;
1404                                                 goto cleanup1;
1405                                         }
1406                                 } else {
1407                                         memset(buff[sg_used], 0, sz);
1408                                 }
1409                                 left -= sz;
1410                                 data_ptr += sz;
1411                                 sg_used++;
1412                         }
1413                         if ((c = cmd_alloc(host, 0)) == NULL) {
1414                                 status = -ENOMEM;
1415                                 goto cleanup1;
1416                         }
1417                         c->cmd_type = CMD_IOCTL_PEND;
1418                         c->Header.ReplyQueue = 0;
1419
1420                         if (ioc->buf_size > 0) {
1421                                 c->Header.SGList = sg_used;
1422                                 c->Header.SGTotal = sg_used;
1423                         } else {
1424                                 c->Header.SGList = 0;
1425                                 c->Header.SGTotal = 0;
1426                         }
1427                         c->Header.LUN = ioc->LUN_info;
1428                         c->Header.Tag.lower = c->busaddr;
1429
1430                         c->Request = ioc->Request;
1431                         if (ioc->buf_size > 0) {
1432                                 int i;
1433                                 for (i = 0; i < sg_used; i++) {
1434                                         temp64.val =
1435                                             pci_map_single(host->pdev, buff[i],
1436                                                     buff_size[i],
1437                                                     PCI_DMA_BIDIRECTIONAL);
1438                                         c->SG[i].Addr.lower =
1439                                             temp64.val32.lower;
1440                                         c->SG[i].Addr.upper =
1441                                             temp64.val32.upper;
1442                                         c->SG[i].Len = buff_size[i];
1443                                         c->SG[i].Ext = 0;       /* we are not chaining */
1444                                 }
1445                         }
1446                         c->waiting = &wait;
1447                         /* Put the request on the tail of the request queue */
1448                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1449                         addQ(&host->reqQ, c);
1450                         host->Qdepth++;
1451                         start_io(host);
1452                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1453                         wait_for_completion(&wait);
1454                         /* unlock the buffers from DMA */
1455                         for (i = 0; i < sg_used; i++) {
1456                                 temp64.val32.lower = c->SG[i].Addr.lower;
1457                                 temp64.val32.upper = c->SG[i].Addr.upper;
1458                                 pci_unmap_single(host->pdev,
1459                                         (dma_addr_t) temp64.val, buff_size[i],
1460                                         PCI_DMA_BIDIRECTIONAL);
1461                         }
1462                         check_ioctl_unit_attention(host, c);
1463                         /* Copy the error information out */
1464                         ioc->error_info = *(c->err_info);
1465                         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1466                                 cmd_free(host, c, 0);
1467                                 status = -EFAULT;
1468                                 goto cleanup1;
1469                         }
1470                         if (ioc->Request.Type.Direction == XFER_READ) {
1471                                 /* Copy the data out of the buffer we created */
1472                                 BYTE __user *ptr = ioc->buf;
1473                                 for (i = 0; i < sg_used; i++) {
1474                                         if (copy_to_user
1475                                             (ptr, buff[i], buff_size[i])) {
1476                                                 cmd_free(host, c, 0);
1477                                                 status = -EFAULT;
1478                                                 goto cleanup1;
1479                                         }
1480                                         ptr += buff_size[i];
1481                                 }
1482                         }
1483                         cmd_free(host, c, 0);
1484                         status = 0;
1485                       cleanup1:
1486                         if (buff) {
1487                                 for (i = 0; i < sg_used; i++)
1488                                         kfree(buff[i]);
1489                                 kfree(buff);
1490                         }
1491                         kfree(buff_size);
1492                         kfree(ioc);
1493                         return status;
1494                 }
1495
1496         /* scsi_cmd_ioctl handles these, below, though some are not */
1497         /* very meaningful for cciss.  SG_IO is the main one people want. */
1498
1499         case SG_GET_VERSION_NUM:
1500         case SG_SET_TIMEOUT:
1501         case SG_GET_TIMEOUT:
1502         case SG_GET_RESERVED_SIZE:
1503         case SG_SET_RESERVED_SIZE:
1504         case SG_EMULATED_HOST:
1505         case SG_IO:
1506         case SCSI_IOCTL_SEND_COMMAND:
1507                 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
1508
1509         /* scsi_cmd_ioctl would normally handle these, below, but */
1510         /* they aren't a good fit for cciss, as CD-ROMs are */
1511         /* not supported, and we don't have any bus/target/lun */
1512         /* which we present to the kernel. */
1513
1514         case CDROM_SEND_PACKET:
1515         case CDROMCLOSETRAY:
1516         case CDROMEJECT:
1517         case SCSI_IOCTL_GET_IDLUN:
1518         case SCSI_IOCTL_GET_BUS_NUMBER:
1519         default:
1520                 return -ENOTTY;
1521         }
1522 }
1523
1524 static void cciss_check_queues(ctlr_info_t *h)
1525 {
1526         int start_queue = h->next_to_run;
1527         int i;
1528
1529         /* check to see if we have maxed out the number of commands that can
1530          * be placed on the queue.  If so then exit.  We do this check here
1531          * in case the interrupt we serviced was from an ioctl and did not
1532          * free any new commands.
1533          */
1534         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1535                 return;
1536
1537         /* We have room on the queue for more commands.  Now we need to queue
1538          * them up.  We will also keep track of the next queue to run so
1539          * that every queue gets a chance to be started first.
1540          */
1541         for (i = 0; i < h->highest_lun + 1; i++) {
1542                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1543                 /* make sure the disk has been added and the drive is real
1544                  * because this can be called from the middle of init_one.
1545                  */
1546                 if (!(h->drv[curr_queue].queue) || !(h->drv[curr_queue].heads))
1547                         continue;
1548                 blk_start_queue(h->gendisk[curr_queue]->queue);
1549
1550                 /* check to see if we have maxed out the number of commands
1551                  * that can be placed on the queue.
1552                  */
1553                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1554                         if (curr_queue == start_queue) {
1555                                 h->next_to_run =
1556                                     (start_queue + 1) % (h->highest_lun + 1);
1557                                 break;
1558                         } else {
1559                                 h->next_to_run = curr_queue;
1560                                 break;
1561                         }
1562                 }
1563         }
1564 }
1565
1566 static void cciss_softirq_done(struct request *rq)
1567 {
1568         CommandList_struct *cmd = rq->completion_data;
1569         ctlr_info_t *h = hba[cmd->ctlr];
1570         unsigned long flags;
1571         u64bit temp64;
1572         int i, ddir;
1573
1574         if (cmd->Request.Type.Direction == XFER_READ)
1575                 ddir = PCI_DMA_FROMDEVICE;
1576         else
1577                 ddir = PCI_DMA_TODEVICE;
1578
1579         /* command did not need to be retried */
1580         /* unmap the DMA mapping for all the scatter gather elements */
1581         for (i = 0; i < cmd->Header.SGList; i++) {
1582                 temp64.val32.lower = cmd->SG[i].Addr.lower;
1583                 temp64.val32.upper = cmd->SG[i].Addr.upper;
1584                 pci_unmap_page(h->pdev, temp64.val, cmd->SG[i].Len, ddir);
1585         }
1586
1587 #ifdef CCISS_DEBUG
1588         printk("Done with %p\n", rq);
1589 #endif                          /* CCISS_DEBUG */
1590
1591         /* set the residual count for pc requests */
1592         if (blk_pc_request(rq))
1593                 rq->resid_len = cmd->err_info->ResidualCnt;
1594
1595         blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1596
1597         spin_lock_irqsave(&h->lock, flags);
1598         cmd_free(h, cmd, 1);
1599         cciss_check_queues(h);
1600         spin_unlock_irqrestore(&h->lock, flags);
1601 }
1602
1603 static void log_unit_to_scsi3addr(ctlr_info_t *h, unsigned char scsi3addr[],
1604         uint32_t log_unit)
1605 {
1606         log_unit = h->drv[log_unit].LunID & 0x03fff;
1607         memset(&scsi3addr[4], 0, 4);
1608         memcpy(&scsi3addr[0], &log_unit, 4);
1609         scsi3addr[3] |= 0x40;
1610 }
1611
1612 /* This function gets the SCSI vendor, model, and revision of a logical drive
1613  * via the inquiry page 0.  Model, vendor, and rev are set to empty strings if
1614  * they cannot be read.
1615  */
1616 static void cciss_get_device_descr(int ctlr, int logvol, int withirq,
1617                                    char *vendor, char *model, char *rev)
1618 {
1619         int rc;
1620         InquiryData_struct *inq_buf;
1621         unsigned char scsi3addr[8];
1622
1623         *vendor = '\0';
1624         *model = '\0';
1625         *rev = '\0';
1626
1627         inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1628         if (!inq_buf)
1629                 return;
1630
1631         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1632         if (withirq)
1633                 rc = sendcmd_withirq(CISS_INQUIRY, ctlr, inq_buf,
1634                              sizeof(InquiryData_struct), 0,
1635                                 scsi3addr, TYPE_CMD);
1636         else
1637                 rc = sendcmd(CISS_INQUIRY, ctlr, inq_buf,
1638                              sizeof(InquiryData_struct), 0,
1639                                 scsi3addr, TYPE_CMD);
1640         if (rc == IO_OK) {
1641                 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1642                 vendor[VENDOR_LEN] = '\0';
1643                 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1644                 model[MODEL_LEN] = '\0';
1645                 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1646                 rev[REV_LEN] = '\0';
1647         }
1648
1649         kfree(inq_buf);
1650         return;
1651 }
1652
1653 /* This function gets the serial number of a logical drive via
1654  * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1655  * number cannot be had, for whatever reason, 16 bytes of 0xff
1656  * are returned instead.
1657  */
1658 static void cciss_get_serial_no(int ctlr, int logvol, int withirq,
1659                                 unsigned char *serial_no, int buflen)
1660 {
1661 #define PAGE_83_INQ_BYTES 64
1662         int rc;
1663         unsigned char *buf;
1664         unsigned char scsi3addr[8];
1665
1666         if (buflen > 16)
1667                 buflen = 16;
1668         memset(serial_no, 0xff, buflen);
1669         buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1670         if (!buf)
1671                 return;
1672         memset(serial_no, 0, buflen);
1673         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1674         if (withirq)
1675                 rc = sendcmd_withirq(CISS_INQUIRY, ctlr, buf,
1676                         PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1677         else
1678                 rc = sendcmd(CISS_INQUIRY, ctlr, buf,
1679                         PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1680         if (rc == IO_OK)
1681                 memcpy(serial_no, &buf[8], buflen);
1682         kfree(buf);
1683         return;
1684 }
1685
1686 /*
1687  * cciss_add_disk sets up the block device queue for a logical drive
1688  */
1689 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1690                                 int drv_index)
1691 {
1692         disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1693         sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1694         disk->major = h->major;
1695         disk->first_minor = drv_index << NWD_SHIFT;
1696         disk->fops = &cciss_fops;
1697         if (h->drv[drv_index].dev == NULL) {
1698                 if (cciss_create_ld_sysfs_entry(h, drv_index))
1699                         goto cleanup_queue;
1700         }
1701         disk->private_data = &h->drv[drv_index];
1702         disk->driverfs_dev = h->drv[drv_index].dev;
1703
1704         /* Set up queue information */
1705         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1706
1707         /* This is a hardware imposed limit. */
1708         blk_queue_max_hw_segments(disk->queue, MAXSGENTRIES);
1709
1710         /* This is a limit in the driver and could be eliminated. */
1711         blk_queue_max_phys_segments(disk->queue, MAXSGENTRIES);
1712
1713         blk_queue_max_sectors(disk->queue, h->cciss_max_sectors);
1714
1715         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1716
1717         disk->queue->queuedata = h;
1718
1719         blk_queue_logical_block_size(disk->queue,
1720                                      h->drv[drv_index].block_size);
1721
1722         /* Make sure all queue data is written out before */
1723         /* setting h->drv[drv_index].queue, as setting this */
1724         /* allows the interrupt handler to start the queue */
1725         wmb();
1726         h->drv[drv_index].queue = disk->queue;
1727         add_disk(disk);
1728         return 0;
1729
1730 cleanup_queue:
1731         blk_cleanup_queue(disk->queue);
1732         disk->queue = NULL;
1733         return -1;
1734 }
1735
1736 /* This function will check the usage_count of the drive to be updated/added.
1737  * If the usage_count is zero and it is a heretofore unknown drive, or,
1738  * the drive's capacity, geometry, or serial number has changed,
1739  * then the drive information will be updated and the disk will be
1740  * re-registered with the kernel.  If these conditions don't hold,
1741  * then it will be left alone for the next reboot.  The exception to this
1742  * is disk 0 which will always be left registered with the kernel since it
1743  * is also the controller node.  Any changes to disk 0 will show up on
1744  * the next reboot.
1745  */
1746 static void cciss_update_drive_info(int ctlr, int drv_index, int first_time)
1747 {
1748         ctlr_info_t *h = hba[ctlr];
1749         struct gendisk *disk;
1750         InquiryData_struct *inq_buff = NULL;
1751         unsigned int block_size;
1752         sector_t total_size;
1753         unsigned long flags = 0;
1754         int ret = 0;
1755         drive_info_struct *drvinfo;
1756
1757         /* Get information about the disk and modify the driver structure */
1758         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1759         drvinfo = kmalloc(sizeof(*drvinfo), GFP_KERNEL);
1760         if (inq_buff == NULL || drvinfo == NULL)
1761                 goto mem_msg;
1762
1763         /* testing to see if 16-byte CDBs are already being used */
1764         if (h->cciss_read == CCISS_READ_16) {
1765                 cciss_read_capacity_16(h->ctlr, drv_index, 1,
1766                         &total_size, &block_size);
1767
1768         } else {
1769                 cciss_read_capacity(ctlr, drv_index, 1,
1770                                     &total_size, &block_size);
1771
1772                 /* if read_capacity returns all F's this volume is >2TB */
1773                 /* in size so we switch to 16-byte CDB's for all */
1774                 /* read/write ops */
1775                 if (total_size == 0xFFFFFFFFULL) {
1776                         cciss_read_capacity_16(ctlr, drv_index, 1,
1777                         &total_size, &block_size);
1778                         h->cciss_read = CCISS_READ_16;
1779                         h->cciss_write = CCISS_WRITE_16;
1780                 } else {
1781                         h->cciss_read = CCISS_READ_10;
1782                         h->cciss_write = CCISS_WRITE_10;
1783                 }
1784         }
1785
1786         cciss_geometry_inquiry(ctlr, drv_index, 1, total_size, block_size,
1787                                inq_buff, drvinfo);
1788         drvinfo->block_size = block_size;
1789         drvinfo->nr_blocks = total_size + 1;
1790
1791         cciss_get_device_descr(ctlr, drv_index, 1, drvinfo->vendor,
1792                                 drvinfo->model, drvinfo->rev);
1793         cciss_get_serial_no(ctlr, drv_index, 1, drvinfo->serial_no,
1794                         sizeof(drvinfo->serial_no));
1795
1796         /* Is it the same disk we already know, and nothing's changed? */
1797         if (h->drv[drv_index].raid_level != -1 &&
1798                 ((memcmp(drvinfo->serial_no,
1799                                 h->drv[drv_index].serial_no, 16) == 0) &&
1800                 drvinfo->block_size == h->drv[drv_index].block_size &&
1801                 drvinfo->nr_blocks == h->drv[drv_index].nr_blocks &&
1802                 drvinfo->heads == h->drv[drv_index].heads &&
1803                 drvinfo->sectors == h->drv[drv_index].sectors &&
1804                 drvinfo->cylinders == h->drv[drv_index].cylinders))
1805                         /* The disk is unchanged, nothing to update */
1806                         goto freeret;
1807
1808         /* If we get here it's not the same disk, or something's changed,
1809          * so we need to * deregister it, and re-register it, if it's not
1810          * in use.
1811          * If the disk already exists then deregister it before proceeding
1812          * (unless it's the first disk (for the controller node).
1813          */
1814         if (h->drv[drv_index].raid_level != -1 && drv_index != 0) {
1815                 printk(KERN_WARNING "disk %d has changed.\n", drv_index);
1816                 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1817                 h->drv[drv_index].busy_configuring = 1;
1818                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1819
1820                 /* deregister_disk sets h->drv[drv_index].queue = NULL
1821                  * which keeps the interrupt handler from starting
1822                  * the queue.
1823                  */
1824                 ret = deregister_disk(h, drv_index, 0);
1825                 h->drv[drv_index].busy_configuring = 0;
1826         }
1827
1828         /* If the disk is in use return */
1829         if (ret)
1830                 goto freeret;
1831
1832         /* Save the new information from cciss_geometry_inquiry
1833          * and serial number inquiry.
1834          */
1835         h->drv[drv_index].block_size = drvinfo->block_size;
1836         h->drv[drv_index].nr_blocks = drvinfo->nr_blocks;
1837         h->drv[drv_index].heads = drvinfo->heads;
1838         h->drv[drv_index].sectors = drvinfo->sectors;
1839         h->drv[drv_index].cylinders = drvinfo->cylinders;
1840         h->drv[drv_index].raid_level = drvinfo->raid_level;
1841         memcpy(h->drv[drv_index].serial_no, drvinfo->serial_no, 16);
1842         memcpy(h->drv[drv_index].vendor, drvinfo->vendor, VENDOR_LEN + 1);
1843         memcpy(h->drv[drv_index].model, drvinfo->model, MODEL_LEN + 1);
1844         memcpy(h->drv[drv_index].rev, drvinfo->rev, REV_LEN + 1);
1845
1846         ++h->num_luns;
1847         disk = h->gendisk[drv_index];
1848         set_capacity(disk, h->drv[drv_index].nr_blocks);
1849
1850         /* If it's not disk 0 (drv_index != 0)
1851          * or if it was disk 0, but there was previously
1852          * no actual corresponding configured logical drive
1853          * (raid_leve == -1) then we want to update the
1854          * logical drive's information.
1855          */
1856         if (drv_index || first_time)
1857                 cciss_add_disk(h, disk, drv_index);
1858
1859 freeret:
1860         kfree(inq_buff);
1861         kfree(drvinfo);
1862         return;
1863 mem_msg:
1864         printk(KERN_ERR "cciss: out of memory\n");
1865         goto freeret;
1866 }
1867
1868 /* This function will find the first index of the controllers drive array
1869  * that has a -1 for the raid_level and will return that index.  This is
1870  * where new drives will be added.  If the index to be returned is greater
1871  * than the highest_lun index for the controller then highest_lun is set
1872  * to this new index.  If there are no available indexes then -1 is returned.
1873  * "controller_node" is used to know if this is a real logical drive, or just
1874  * the controller node, which determines if this counts towards highest_lun.
1875  */
1876 static int cciss_find_free_drive_index(int ctlr, int controller_node)
1877 {
1878         int i;
1879
1880         for (i = 0; i < CISS_MAX_LUN; i++) {
1881                 if (hba[ctlr]->drv[i].raid_level == -1) {
1882                         if (i > hba[ctlr]->highest_lun)
1883                                 if (!controller_node)
1884                                         hba[ctlr]->highest_lun = i;
1885                         return i;
1886                 }
1887         }
1888         return -1;
1889 }
1890
1891 /* cciss_add_gendisk finds a free hba[]->drv structure
1892  * and allocates a gendisk if needed, and sets the lunid
1893  * in the drvinfo structure.   It returns the index into
1894  * the ->drv[] array, or -1 if none are free.
1895  * is_controller_node indicates whether highest_lun should
1896  * count this disk, or if it's only being added to provide
1897  * a means to talk to the controller in case no logical
1898  * drives have yet been configured.
1899  */
1900 static int cciss_add_gendisk(ctlr_info_t *h, __u32 lunid, int controller_node)
1901 {
1902         int drv_index;
1903
1904         drv_index = cciss_find_free_drive_index(h->ctlr, controller_node);
1905         if (drv_index == -1)
1906                 return -1;
1907         /*Check if the gendisk needs to be allocated */
1908         if (!h->gendisk[drv_index]) {
1909                 h->gendisk[drv_index] =
1910                         alloc_disk(1 << NWD_SHIFT);
1911                 if (!h->gendisk[drv_index]) {
1912                         printk(KERN_ERR "cciss%d: could not "
1913                                 "allocate a new disk %d\n",
1914                                 h->ctlr, drv_index);
1915                         return -1;
1916                 }
1917         }
1918         h->drv[drv_index].LunID = lunid;
1919         if (cciss_create_ld_sysfs_entry(h, drv_index))
1920                 goto err_free_disk;
1921
1922         /* Don't need to mark this busy because nobody */
1923         /* else knows about this disk yet to contend */
1924         /* for access to it. */
1925         h->drv[drv_index].busy_configuring = 0;
1926         wmb();
1927         return drv_index;
1928
1929 err_free_disk:
1930         put_disk(h->gendisk[drv_index]);
1931         h->gendisk[drv_index] = NULL;
1932         return -1;
1933 }
1934
1935 /* This is for the special case of a controller which
1936  * has no logical drives.  In this case, we still need
1937  * to register a disk so the controller can be accessed
1938  * by the Array Config Utility.
1939  */
1940 static void cciss_add_controller_node(ctlr_info_t *h)
1941 {
1942         struct gendisk *disk;
1943         int drv_index;
1944
1945         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
1946                 return;
1947
1948         drv_index = cciss_add_gendisk(h, 0, 1);
1949         if (drv_index == -1) {
1950                 printk(KERN_WARNING "cciss%d: could not "
1951                         "add disk 0.\n", h->ctlr);
1952                 return;
1953         }
1954         h->drv[drv_index].block_size = 512;
1955         h->drv[drv_index].nr_blocks = 0;
1956         h->drv[drv_index].heads = 0;
1957         h->drv[drv_index].sectors = 0;
1958         h->drv[drv_index].cylinders = 0;
1959         h->drv[drv_index].raid_level = -1;
1960         memset(h->drv[drv_index].serial_no, 0, 16);
1961         disk = h->gendisk[drv_index];
1962         cciss_add_disk(h, disk, drv_index);
1963 }
1964
1965 /* This function will add and remove logical drives from the Logical
1966  * drive array of the controller and maintain persistency of ordering
1967  * so that mount points are preserved until the next reboot.  This allows
1968  * for the removal of logical drives in the middle of the drive array
1969  * without a re-ordering of those drives.
1970  * INPUT
1971  * h            = The controller to perform the operations on
1972  */
1973 static int rebuild_lun_table(ctlr_info_t *h, int first_time)
1974 {
1975         int ctlr = h->ctlr;
1976         int num_luns;
1977         ReportLunData_struct *ld_buff = NULL;
1978         int return_code;
1979         int listlength = 0;
1980         int i;
1981         int drv_found;
1982         int drv_index = 0;
1983         __u32 lunid = 0;
1984         unsigned long flags;
1985
1986         if (!capable(CAP_SYS_RAWIO))
1987                 return -EPERM;
1988
1989         /* Set busy_configuring flag for this operation */
1990         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1991         if (h->busy_configuring) {
1992                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1993                 return -EBUSY;
1994         }
1995         h->busy_configuring = 1;
1996         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1997
1998         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
1999         if (ld_buff == NULL)
2000                 goto mem_msg;
2001
2002         return_code = sendcmd_withirq(CISS_REPORT_LOG, ctlr, ld_buff,
2003                                       sizeof(ReportLunData_struct),
2004                                       0, CTLR_LUNID, TYPE_CMD);
2005
2006         if (return_code == IO_OK)
2007                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2008         else {  /* reading number of logical volumes failed */
2009                 printk(KERN_WARNING "cciss: report logical volume"
2010                        " command failed\n");
2011                 listlength = 0;
2012                 goto freeret;
2013         }
2014
2015         num_luns = listlength / 8;      /* 8 bytes per entry */
2016         if (num_luns > CISS_MAX_LUN) {
2017                 num_luns = CISS_MAX_LUN;
2018                 printk(KERN_WARNING "cciss: more luns configured"
2019                        " on controller than can be handled by"
2020                        " this driver.\n");
2021         }
2022
2023         if (num_luns == 0)
2024                 cciss_add_controller_node(h);
2025
2026         /* Compare controller drive array to driver's drive array
2027          * to see if any drives are missing on the controller due
2028          * to action of Array Config Utility (user deletes drive)
2029          * and deregister logical drives which have disappeared.
2030          */
2031         for (i = 0; i <= h->highest_lun; i++) {
2032                 int j;
2033                 drv_found = 0;
2034
2035                 /* skip holes in the array from already deleted drives */
2036                 if (h->drv[i].raid_level == -1)
2037                         continue;
2038
2039                 for (j = 0; j < num_luns; j++) {
2040                         memcpy(&lunid, &ld_buff->LUN[j][0], 4);
2041                         lunid = le32_to_cpu(lunid);
2042                         if (h->drv[i].LunID == lunid) {
2043                                 drv_found = 1;
2044                                 break;
2045                         }
2046                 }
2047                 if (!drv_found) {
2048                         /* Deregister it from the OS, it's gone. */
2049                         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2050                         h->drv[i].busy_configuring = 1;
2051                         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2052                         return_code = deregister_disk(h, i, 1);
2053                         h->drv[i].busy_configuring = 0;
2054                 }
2055         }
2056
2057         /* Compare controller drive array to driver's drive array.
2058          * Check for updates in the drive information and any new drives
2059          * on the controller due to ACU adding logical drives, or changing
2060          * a logical drive's size, etc.  Reregister any new/changed drives
2061          */
2062         for (i = 0; i < num_luns; i++) {
2063                 int j;
2064
2065                 drv_found = 0;
2066
2067                 memcpy(&lunid, &ld_buff->LUN[i][0], 4);
2068                 lunid = le32_to_cpu(lunid);
2069
2070                 /* Find if the LUN is already in the drive array
2071                  * of the driver.  If so then update its info
2072                  * if not in use.  If it does not exist then find
2073                  * the first free index and add it.
2074                  */
2075                 for (j = 0; j <= h->highest_lun; j++) {
2076                         if (h->drv[j].raid_level != -1 &&
2077                                 h->drv[j].LunID == lunid) {
2078                                 drv_index = j;
2079                                 drv_found = 1;
2080                                 break;
2081                         }
2082                 }
2083
2084                 /* check if the drive was found already in the array */
2085                 if (!drv_found) {
2086                         drv_index = cciss_add_gendisk(h, lunid, 0);
2087                         if (drv_index == -1)
2088                                 goto freeret;
2089                 }
2090                 cciss_update_drive_info(ctlr, drv_index, first_time);
2091         }               /* end for */
2092
2093 freeret:
2094         kfree(ld_buff);
2095         h->busy_configuring = 0;
2096         /* We return -1 here to tell the ACU that we have registered/updated
2097          * all of the drives that we can and to keep it from calling us
2098          * additional times.
2099          */
2100         return -1;
2101 mem_msg:
2102         printk(KERN_ERR "cciss: out of memory\n");
2103         h->busy_configuring = 0;
2104         goto freeret;
2105 }
2106
2107 /* This function will deregister the disk and it's queue from the
2108  * kernel.  It must be called with the controller lock held and the
2109  * drv structures busy_configuring flag set.  It's parameters are:
2110  *
2111  * disk = This is the disk to be deregistered
2112  * drv  = This is the drive_info_struct associated with the disk to be
2113  *        deregistered.  It contains information about the disk used
2114  *        by the driver.
2115  * clear_all = This flag determines whether or not the disk information
2116  *             is going to be completely cleared out and the highest_lun
2117  *             reset.  Sometimes we want to clear out information about
2118  *             the disk in preparation for re-adding it.  In this case
2119  *             the highest_lun should be left unchanged and the LunID
2120  *             should not be cleared.
2121 */
2122 static int deregister_disk(ctlr_info_t *h, int drv_index,
2123                            int clear_all)
2124 {
2125         int i;
2126         struct gendisk *disk;
2127         drive_info_struct *drv;
2128
2129         if (!capable(CAP_SYS_RAWIO))
2130                 return -EPERM;
2131
2132         drv = &h->drv[drv_index];
2133         disk = h->gendisk[drv_index];
2134
2135         /* make sure logical volume is NOT is use */
2136         if (clear_all || (h->gendisk[0] == disk)) {
2137                 if (drv->usage_count > 1)
2138                         return -EBUSY;
2139         } else if (drv->usage_count > 0)
2140                 return -EBUSY;
2141
2142         /* invalidate the devices and deregister the disk.  If it is disk
2143          * zero do not deregister it but just zero out it's values.  This
2144          * allows us to delete disk zero but keep the controller registered.
2145          */
2146         if (h->gendisk[0] != disk) {
2147                 struct request_queue *q = disk->queue;
2148                 if (disk->flags & GENHD_FL_UP)
2149                         del_gendisk(disk);
2150                 if (q) {
2151                         blk_cleanup_queue(q);
2152                         /* Set drv->queue to NULL so that we do not try
2153                          * to call blk_start_queue on this queue in the
2154                          * interrupt handler
2155                          */
2156                         drv->queue = NULL;
2157                 }
2158                 /* If clear_all is set then we are deleting the logical
2159                  * drive, not just refreshing its info.  For drives
2160                  * other than disk 0 we will call put_disk.  We do not
2161                  * do this for disk 0 as we need it to be able to
2162                  * configure the controller.
2163                  */
2164                 if (clear_all){
2165                         /* This isn't pretty, but we need to find the
2166                          * disk in our array and NULL our the pointer.
2167                          * This is so that we will call alloc_disk if
2168                          * this index is used again later.
2169                          */
2170                         for (i=0; i < CISS_MAX_LUN; i++){
2171                                 if (h->gendisk[i] == disk) {
2172                                         h->gendisk[i] = NULL;
2173                                         break;
2174                                 }
2175                         }
2176                         put_disk(disk);
2177                 }
2178         } else {
2179                 set_capacity(disk, 0);
2180         }
2181
2182         --h->num_luns;
2183         /* zero out the disk size info */
2184         drv->nr_blocks = 0;
2185         drv->block_size = 0;
2186         drv->heads = 0;
2187         drv->sectors = 0;
2188         drv->cylinders = 0;
2189         drv->raid_level = -1;   /* This can be used as a flag variable to
2190                                  * indicate that this element of the drive
2191                                  * array is free.
2192                                  */
2193         cciss_destroy_ld_sysfs_entry(h, drv_index);
2194
2195         if (clear_all) {
2196                 /* check to see if it was the last disk */
2197                 if (drv == h->drv + h->highest_lun) {
2198                         /* if so, find the new hightest lun */
2199                         int i, newhighest = -1;
2200                         for (i = 0; i <= h->highest_lun; i++) {
2201                                 /* if the disk has size > 0, it is available */
2202                                 if (h->drv[i].heads)
2203                                         newhighest = i;
2204                         }
2205                         h->highest_lun = newhighest;
2206                 }
2207
2208                 drv->LunID = 0;
2209         }
2210         return 0;
2211 }
2212
2213 static int fill_cmd(CommandList_struct *c, __u8 cmd, int ctlr, void *buff,
2214                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2215                 int cmd_type)
2216 {
2217         ctlr_info_t *h = hba[ctlr];
2218         u64bit buff_dma_handle;
2219         int status = IO_OK;
2220
2221         c->cmd_type = CMD_IOCTL_PEND;
2222         c->Header.ReplyQueue = 0;
2223         if (buff != NULL) {
2224                 c->Header.SGList = 1;
2225                 c->Header.SGTotal = 1;
2226         } else {
2227                 c->Header.SGList = 0;
2228                 c->Header.SGTotal = 0;
2229         }
2230         c->Header.Tag.lower = c->busaddr;
2231         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2232
2233         c->Request.Type.Type = cmd_type;
2234         if (cmd_type == TYPE_CMD) {
2235                 switch (cmd) {
2236                 case CISS_INQUIRY:
2237                         /* are we trying to read a vital product page */
2238                         if (page_code != 0) {
2239                                 c->Request.CDB[1] = 0x01;
2240                                 c->Request.CDB[2] = page_code;
2241                         }
2242                         c->Request.CDBLen = 6;
2243                         c->Request.Type.Attribute = ATTR_SIMPLE;
2244                         c->Request.Type.Direction = XFER_READ;
2245                         c->Request.Timeout = 0;
2246                         c->Request.CDB[0] = CISS_INQUIRY;
2247                         c->Request.CDB[4] = size & 0xFF;
2248                         break;
2249                 case CISS_REPORT_LOG:
2250                 case CISS_REPORT_PHYS:
2251                         /* Talking to controller so It's a physical command
2252                            mode = 00 target = 0.  Nothing to write.
2253                          */
2254                         c->Request.CDBLen = 12;
2255                         c->Request.Type.Attribute = ATTR_SIMPLE;
2256                         c->Request.Type.Direction = XFER_READ;
2257                         c->Request.Timeout = 0;
2258                         c->Request.CDB[0] = cmd;
2259                         c->Request.CDB[6] = (size >> 24) & 0xFF;        //MSB
2260                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2261                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2262                         c->Request.CDB[9] = size & 0xFF;
2263                         break;
2264
2265                 case CCISS_READ_CAPACITY:
2266                         c->Request.CDBLen = 10;
2267                         c->Request.Type.Attribute = ATTR_SIMPLE;
2268                         c->Request.Type.Direction = XFER_READ;
2269                         c->Request.Timeout = 0;
2270                         c->Request.CDB[0] = cmd;
2271                         break;
2272                 case CCISS_READ_CAPACITY_16:
2273                         c->Request.CDBLen = 16;
2274                         c->Request.Type.Attribute = ATTR_SIMPLE;
2275                         c->Request.Type.Direction = XFER_READ;
2276                         c->Request.Timeout = 0;
2277                         c->Request.CDB[0] = cmd;
2278                         c->Request.CDB[1] = 0x10;
2279                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2280                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2281                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2282                         c->Request.CDB[13] = size & 0xFF;
2283                         c->Request.Timeout = 0;
2284                         c->Request.CDB[0] = cmd;
2285                         break;
2286                 case CCISS_CACHE_FLUSH:
2287                         c->Request.CDBLen = 12;
2288                         c->Request.Type.Attribute = ATTR_SIMPLE;
2289                         c->Request.Type.Direction = XFER_WRITE;
2290                         c->Request.Timeout = 0;
2291                         c->Request.CDB[0] = BMIC_WRITE;
2292                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2293                         break;
2294                 case TEST_UNIT_READY:
2295                         c->Request.CDBLen = 6;
2296                         c->Request.Type.Attribute = ATTR_SIMPLE;
2297                         c->Request.Type.Direction = XFER_NONE;
2298                         c->Request.Timeout = 0;
2299                         break;
2300                 default:
2301                         printk(KERN_WARNING
2302                                "cciss%d:  Unknown Command 0x%c\n", ctlr, cmd);
2303                         return IO_ERROR;
2304                 }
2305         } else if (cmd_type == TYPE_MSG) {
2306                 switch (cmd) {
2307                 case 0: /* ABORT message */
2308                         c->Request.CDBLen = 12;
2309                         c->Request.Type.Attribute = ATTR_SIMPLE;
2310                         c->Request.Type.Direction = XFER_WRITE;
2311                         c->Request.Timeout = 0;
2312                         c->Request.CDB[0] = cmd;        /* abort */
2313                         c->Request.CDB[1] = 0;  /* abort a command */
2314                         /* buff contains the tag of the command to abort */
2315                         memcpy(&c->Request.CDB[4], buff, 8);
2316                         break;
2317                 case 1: /* RESET message */
2318                         c->Request.CDBLen = 16;
2319                         c->Request.Type.Attribute = ATTR_SIMPLE;
2320                         c->Request.Type.Direction = XFER_NONE;
2321                         c->Request.Timeout = 0;
2322                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2323                         c->Request.CDB[0] = cmd;        /* reset */
2324                         c->Request.CDB[1] = 0x03;       /* reset a target */
2325                         break;
2326                 case 3: /* No-Op message */
2327                         c->Request.CDBLen = 1;
2328                         c->Request.Type.Attribute = ATTR_SIMPLE;
2329                         c->Request.Type.Direction = XFER_WRITE;
2330                         c->Request.Timeout = 0;
2331                         c->Request.CDB[0] = cmd;
2332                         break;
2333                 default:
2334                         printk(KERN_WARNING
2335                                "cciss%d: unknown message type %d\n", ctlr, cmd);
2336                         return IO_ERROR;
2337                 }
2338         } else {
2339                 printk(KERN_WARNING
2340                        "cciss%d: unknown command type %d\n", ctlr, cmd_type);
2341                 return IO_ERROR;
2342         }
2343         /* Fill in the scatter gather information */
2344         if (size > 0) {
2345                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2346                                                              buff, size,
2347                                                              PCI_DMA_BIDIRECTIONAL);
2348                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2349                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2350                 c->SG[0].Len = size;
2351                 c->SG[0].Ext = 0;       /* we are not chaining */
2352         }
2353         return status;
2354 }
2355
2356 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2357 {
2358         switch (c->err_info->ScsiStatus) {
2359         case SAM_STAT_GOOD:
2360                 return IO_OK;
2361         case SAM_STAT_CHECK_CONDITION:
2362                 switch (0xf & c->err_info->SenseInfo[2]) {
2363                 case 0: return IO_OK; /* no sense */
2364                 case 1: return IO_OK; /* recovered error */
2365                 default:
2366                         printk(KERN_WARNING "cciss%d: cmd 0x%02x "
2367                                 "check condition, sense key = 0x%02x\n",
2368                                 h->ctlr, c->Request.CDB[0],
2369                                 c->err_info->SenseInfo[2]);
2370                 }
2371                 break;
2372         default:
2373                 printk(KERN_WARNING "cciss%d: cmd 0x%02x"
2374                         "scsi status = 0x%02x\n", h->ctlr,
2375                         c->Request.CDB[0], c->err_info->ScsiStatus);
2376                 break;
2377         }
2378         return IO_ERROR;
2379 }
2380
2381 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2382 {
2383         int return_status = IO_OK;
2384
2385         if (c->err_info->CommandStatus == CMD_SUCCESS)
2386                 return IO_OK;
2387
2388         switch (c->err_info->CommandStatus) {
2389         case CMD_TARGET_STATUS:
2390                 return_status = check_target_status(h, c);
2391                 break;
2392         case CMD_DATA_UNDERRUN:
2393         case CMD_DATA_OVERRUN:
2394                 /* expected for inquiry and report lun commands */
2395                 break;
2396         case CMD_INVALID:
2397                 printk(KERN_WARNING "cciss: cmd 0x%02x is "
2398                        "reported invalid\n", c->Request.CDB[0]);
2399                 return_status = IO_ERROR;
2400                 break;
2401         case CMD_PROTOCOL_ERR:
2402                 printk(KERN_WARNING "cciss: cmd 0x%02x has "
2403                        "protocol error \n", c->Request.CDB[0]);
2404                 return_status = IO_ERROR;
2405                 break;
2406         case CMD_HARDWARE_ERR:
2407                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2408                        " hardware error\n", c->Request.CDB[0]);
2409                 return_status = IO_ERROR;
2410                 break;
2411         case CMD_CONNECTION_LOST:
2412                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2413                        "connection lost\n", c->Request.CDB[0]);
2414                 return_status = IO_ERROR;
2415                 break;
2416         case CMD_ABORTED:
2417                 printk(KERN_WARNING "cciss: cmd 0x%02x was "
2418                        "aborted\n", c->Request.CDB[0]);
2419                 return_status = IO_ERROR;
2420                 break;
2421         case CMD_ABORT_FAILED:
2422                 printk(KERN_WARNING "cciss: cmd 0x%02x reports "
2423                        "abort failed\n", c->Request.CDB[0]);
2424                 return_status = IO_ERROR;
2425                 break;
2426         case CMD_UNSOLICITED_ABORT:
2427                 printk(KERN_WARNING
2428                        "cciss%d: unsolicited abort 0x%02x\n", h->ctlr,
2429                         c->Request.CDB[0]);
2430                 return_status = IO_NEEDS_RETRY;
2431                 break;
2432         default:
2433                 printk(KERN_WARNING "cciss: cmd 0x%02x returned "
2434                        "unknown status %x\n", c->Request.CDB[0],
2435                        c->err_info->CommandStatus);
2436                 return_status = IO_ERROR;
2437         }
2438         return return_status;
2439 }
2440
2441 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2442         int attempt_retry)
2443 {
2444         DECLARE_COMPLETION_ONSTACK(wait);
2445         u64bit buff_dma_handle;
2446         unsigned long flags;
2447         int return_status = IO_OK;
2448
2449 resend_cmd2:
2450         c->waiting = &wait;
2451         /* Put the request on the tail of the queue and send it */
2452         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2453         addQ(&h->reqQ, c);
2454         h->Qdepth++;
2455         start_io(h);
2456         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2457
2458         wait_for_completion(&wait);
2459
2460         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2461                 goto command_done;
2462
2463         return_status = process_sendcmd_error(h, c);
2464
2465         if (return_status == IO_NEEDS_RETRY &&
2466                 c->retry_count < MAX_CMD_RETRIES) {
2467                 printk(KERN_WARNING "cciss%d: retrying 0x%02x\n", h->ctlr,
2468                         c->Request.CDB[0]);
2469                 c->retry_count++;
2470                 /* erase the old error information */
2471                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2472                 return_status = IO_OK;
2473                 INIT_COMPLETION(wait);
2474                 goto resend_cmd2;
2475         }
2476
2477 command_done:
2478         /* unlock the buffers from DMA */
2479         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2480         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2481         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2482                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2483         return return_status;
2484 }
2485
2486 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
2487                            __u8 page_code, unsigned char scsi3addr[],
2488                         int cmd_type)
2489 {
2490         ctlr_info_t *h = hba[ctlr];
2491         CommandList_struct *c;
2492         int return_status;
2493
2494         c = cmd_alloc(h, 0);
2495         if (!c)
2496                 return -ENOMEM;
2497         return_status = fill_cmd(c, cmd, ctlr, buff, size, page_code,
2498                 scsi3addr, cmd_type);
2499         if (return_status == IO_OK)
2500                 return_status = sendcmd_withirq_core(h, c, 1);
2501
2502         cmd_free(h, c, 0);
2503         return return_status;
2504 }
2505
2506 static void cciss_geometry_inquiry(int ctlr, int logvol,
2507                                    int withirq, sector_t total_size,
2508                                    unsigned int block_size,
2509                                    InquiryData_struct *inq_buff,
2510                                    drive_info_struct *drv)
2511 {
2512         int return_code;
2513         unsigned long t;
2514         unsigned char scsi3addr[8];
2515
2516         memset(inq_buff, 0, sizeof(InquiryData_struct));
2517         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2518         if (withirq)
2519                 return_code = sendcmd_withirq(CISS_INQUIRY, ctlr,
2520                                               inq_buff, sizeof(*inq_buff),
2521                                               0xC1, scsi3addr, TYPE_CMD);
2522         else
2523                 return_code = sendcmd(CISS_INQUIRY, ctlr, inq_buff,
2524                                       sizeof(*inq_buff), 0xC1, scsi3addr,
2525                                       TYPE_CMD);
2526         if (return_code == IO_OK) {
2527                 if (inq_buff->data_byte[8] == 0xFF) {
2528                         printk(KERN_WARNING
2529                                "cciss: reading geometry failed, volume "
2530                                "does not support reading geometry\n");
2531                         drv->heads = 255;
2532                         drv->sectors = 32;      // Sectors per track
2533                         drv->cylinders = total_size + 1;
2534                         drv->raid_level = RAID_UNKNOWN;
2535                 } else {
2536                         drv->heads = inq_buff->data_byte[6];
2537                         drv->sectors = inq_buff->data_byte[7];
2538                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2539                         drv->cylinders += inq_buff->data_byte[5];
2540                         drv->raid_level = inq_buff->data_byte[8];
2541                 }
2542                 drv->block_size = block_size;
2543                 drv->nr_blocks = total_size + 1;
2544                 t = drv->heads * drv->sectors;
2545                 if (t > 1) {
2546                         sector_t real_size = total_size + 1;
2547                         unsigned long rem = sector_div(real_size, t);
2548                         if (rem)
2549                                 real_size++;
2550                         drv->cylinders = real_size;
2551                 }
2552         } else {                /* Get geometry failed */
2553                 printk(KERN_WARNING "cciss: reading geometry failed\n");
2554         }
2555         printk(KERN_INFO "      heads=%d, sectors=%d, cylinders=%d\n\n",
2556                drv->heads, drv->sectors, drv->cylinders);
2557 }
2558
2559 static void
2560 cciss_read_capacity(int ctlr, int logvol, int withirq, sector_t *total_size,
2561                     unsigned int *block_size)
2562 {
2563         ReadCapdata_struct *buf;
2564         int return_code;
2565         unsigned char scsi3addr[8];
2566
2567         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2568         if (!buf) {
2569                 printk(KERN_WARNING "cciss: out of memory\n");
2570                 return;
2571         }
2572
2573         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2574         if (withirq)
2575                 return_code = sendcmd_withirq(CCISS_READ_CAPACITY,
2576                                 ctlr, buf, sizeof(ReadCapdata_struct),
2577                                         0, scsi3addr, TYPE_CMD);
2578         else
2579                 return_code = sendcmd(CCISS_READ_CAPACITY,
2580                                 ctlr, buf, sizeof(ReadCapdata_struct),
2581                                         0, scsi3addr, TYPE_CMD);
2582         if (return_code == IO_OK) {
2583                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2584                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2585         } else {                /* read capacity command failed */
2586                 printk(KERN_WARNING "cciss: read capacity failed\n");
2587                 *total_size = 0;
2588                 *block_size = BLOCK_SIZE;
2589         }
2590         if (*total_size != 0)
2591                 printk(KERN_INFO "      blocks= %llu block_size= %d\n",
2592                 (unsigned long long)*total_size+1, *block_size);
2593         kfree(buf);
2594 }
2595
2596 static void
2597 cciss_read_capacity_16(int ctlr, int logvol, int withirq, sector_t *total_size,                                 unsigned int *block_size)
2598 {
2599         ReadCapdata_struct_16 *buf;
2600         int return_code;
2601         unsigned char scsi3addr[8];
2602
2603         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2604         if (!buf) {
2605                 printk(KERN_WARNING "cciss: out of memory\n");
2606                 return;
2607         }
2608
2609         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2610         if (withirq) {
2611                 return_code = sendcmd_withirq(CCISS_READ_CAPACITY_16,
2612                         ctlr, buf, sizeof(ReadCapdata_struct_16),
2613                                 0, scsi3addr, TYPE_CMD);
2614         }
2615         else {
2616                 return_code = sendcmd(CCISS_READ_CAPACITY_16,
2617                         ctlr, buf, sizeof(ReadCapdata_struct_16),
2618                                 0, scsi3addr, TYPE_CMD);
2619         }
2620         if (return_code == IO_OK) {
2621                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2622                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2623         } else {                /* read capacity command failed */
2624                 printk(KERN_WARNING "cciss: read capacity failed\n");
2625                 *total_size = 0;
2626                 *block_size = BLOCK_SIZE;
2627         }
2628         printk(KERN_INFO "      blocks= %llu block_size= %d\n",
2629                (unsigned long long)*total_size+1, *block_size);
2630         kfree(buf);
2631 }
2632
2633 static int cciss_revalidate(struct gendisk *disk)
2634 {
2635         ctlr_info_t *h = get_host(disk);
2636         drive_info_struct *drv = get_drv(disk);
2637         int logvol;
2638         int FOUND = 0;
2639         unsigned int block_size;
2640         sector_t total_size;
2641         InquiryData_struct *inq_buff = NULL;
2642
2643         for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
2644                 if (h->drv[logvol].LunID == drv->LunID) {
2645                         FOUND = 1;
2646                         break;
2647                 }
2648         }
2649
2650         if (!FOUND)
2651                 return 1;
2652
2653         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2654         if (inq_buff == NULL) {
2655                 printk(KERN_WARNING "cciss: out of memory\n");
2656                 return 1;
2657         }
2658         if (h->cciss_read == CCISS_READ_10) {
2659                 cciss_read_capacity(h->ctlr, logvol, 1,
2660                                         &total_size, &block_size);
2661         } else {
2662                 cciss_read_capacity_16(h->ctlr, logvol, 1,
2663                                         &total_size, &block_size);
2664         }
2665         cciss_geometry_inquiry(h->ctlr, logvol, 1, total_size, block_size,
2666                                inq_buff, drv);
2667
2668         blk_queue_logical_block_size(drv->queue, drv->block_size);
2669         set_capacity(disk, drv->nr_blocks);
2670
2671         kfree(inq_buff);
2672         return 0;
2673 }
2674
2675 /*
2676  *   Wait polling for a command to complete.
2677  *   The memory mapped FIFO is polled for the completion.
2678  *   Used only at init time, interrupts from the HBA are disabled.
2679  */
2680 static unsigned long pollcomplete(int ctlr)
2681 {
2682         unsigned long done;
2683         int i;
2684
2685         /* Wait (up to 20 seconds) for a command to complete */
2686
2687         for (i = 20 * HZ; i > 0; i--) {
2688                 done = hba[ctlr]->access.command_completed(hba[ctlr]);
2689                 if (done == FIFO_EMPTY)
2690                         schedule_timeout_uninterruptible(1);
2691                 else
2692                         return done;
2693         }
2694         /* Invalid address to tell caller we ran out of time */
2695         return 1;
2696 }
2697
2698 /* Send command c to controller h and poll for it to complete.
2699  * Turns interrupts off on the board.  Used at driver init time
2700  * and during SCSI error recovery.
2701  */
2702 static int sendcmd_core(ctlr_info_t *h, CommandList_struct *c)
2703 {
2704         int i;
2705         unsigned long complete;
2706         int status = IO_ERROR;
2707         u64bit buff_dma_handle;
2708
2709 resend_cmd1:
2710
2711         /* Disable interrupt on the board. */
2712         h->access.set_intr_mask(h, CCISS_INTR_OFF);
2713
2714         /* Make sure there is room in the command FIFO */
2715         /* Actually it should be completely empty at this time */
2716         /* unless we are in here doing error handling for the scsi */
2717         /* tape side of the driver. */
2718         for (i = 200000; i > 0; i--) {
2719                 /* if fifo isn't full go */
2720                 if (!(h->access.fifo_full(h)))
2721                         break;
2722                 udelay(10);
2723                 printk(KERN_WARNING "cciss cciss%d: SendCmd FIFO full,"
2724                        " waiting!\n", h->ctlr);
2725         }
2726         h->access.submit_command(h, c); /* Send the cmd */
2727         do {
2728                 complete = pollcomplete(h->ctlr);
2729
2730 #ifdef CCISS_DEBUG
2731                 printk(KERN_DEBUG "cciss: command completed\n");
2732 #endif                          /* CCISS_DEBUG */
2733
2734                 if (complete == 1) {
2735                         printk(KERN_WARNING
2736                                "cciss cciss%d: SendCmd Timeout out, "
2737                                "No command list address returned!\n", h->ctlr);
2738                         status = IO_ERROR;
2739                         break;
2740                 }
2741
2742                 /* Make sure it's the command we're expecting. */
2743                 if ((complete & ~CISS_ERROR_BIT) != c->busaddr) {
2744                         printk(KERN_WARNING "cciss%d: Unexpected command "
2745                                 "completion.\n", h->ctlr);
2746                         continue;
2747                 }
2748
2749                 /* It is our command.  If no error, we're done. */
2750                 if (!(complete & CISS_ERROR_BIT)) {
2751                         status = IO_OK;
2752                         break;
2753                 }
2754
2755                 /* There is an error... */
2756
2757                 /* if data overrun or underun on Report command ignore it */
2758                 if (((c->Request.CDB[0] == CISS_REPORT_LOG) ||
2759                      (c->Request.CDB[0] == CISS_REPORT_PHYS) ||
2760                      (c->Request.CDB[0] == CISS_INQUIRY)) &&
2761                         ((c->err_info->CommandStatus == CMD_DATA_OVERRUN) ||
2762                          (c->err_info->CommandStatus == CMD_DATA_UNDERRUN))) {
2763                         complete = c->busaddr;
2764                         status = IO_OK;
2765                         break;
2766                 }
2767
2768                 if (c->err_info->CommandStatus == CMD_UNSOLICITED_ABORT) {
2769                         printk(KERN_WARNING "cciss%d: unsolicited abort %p\n",
2770                                 h->ctlr, c);
2771                         if (c->retry_count < MAX_CMD_RETRIES) {
2772                                 printk(KERN_WARNING "cciss%d: retrying %p\n",
2773                                    h->ctlr, c);
2774                                 c->retry_count++;
2775                                 /* erase the old error information */
2776                                 memset(c->err_info, 0, sizeof(c->err_info));
2777                                 goto resend_cmd1;
2778                         }
2779                         printk(KERN_WARNING "cciss%d: retried %p too many "
2780                                 "times\n", h->ctlr, c);
2781                         status = IO_ERROR;
2782                         break;
2783                 }
2784
2785                 if (c->err_info->CommandStatus == CMD_UNABORTABLE) {
2786                         printk(KERN_WARNING "cciss%d: command could not be "
2787                                 "aborted.\n", h->ctlr);
2788                         status = IO_ERROR;
2789                         break;
2790                 }
2791
2792                 if (c->err_info->CommandStatus == CMD_TARGET_STATUS) {
2793                         status = check_target_status(h, c);
2794                         break;
2795                 }
2796
2797                 printk(KERN_WARNING "cciss%d: sendcmd error\n", h->ctlr);
2798                 printk(KERN_WARNING "cmd = 0x%02x, CommandStatus = 0x%02x\n",
2799                         c->Request.CDB[0], c->err_info->CommandStatus);
2800                 status = IO_ERROR;
2801                 break;
2802
2803         } while (1);
2804
2805         /* unlock the data buffer from DMA */
2806         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2807         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2808         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2809                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2810         return status;
2811 }
2812
2813 /*
2814  * Send a command to the controller, and wait for it to complete.
2815  * Used at init time, and during SCSI error recovery.
2816  */
2817 static int sendcmd(__u8 cmd, int ctlr, void *buff, size_t size,
2818         __u8 page_code, unsigned char *scsi3addr, int cmd_type)
2819 {
2820         CommandList_struct *c;
2821         int status;
2822
2823         c = cmd_alloc(hba[ctlr], 1);
2824         if (!c) {
2825                 printk(KERN_WARNING "cciss: unable to get memory");
2826                 return IO_ERROR;
2827         }
2828         status = fill_cmd(c, cmd, ctlr, buff, size, page_code,
2829                 scsi3addr, cmd_type);
2830         if (status == IO_OK)
2831                 status = sendcmd_core(hba[ctlr], c);
2832         cmd_free(hba[ctlr], c, 1);
2833         return status;
2834 }
2835
2836 /*
2837  * Map (physical) PCI mem into (virtual) kernel space
2838  */
2839 static void __iomem *remap_pci_mem(ulong base, ulong size)
2840 {
2841         ulong page_base = ((ulong) base) & PAGE_MASK;
2842         ulong page_offs = ((ulong) base) - page_base;
2843         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2844
2845         return page_remapped ? (page_remapped + page_offs) : NULL;
2846 }
2847
2848 /*
2849  * Takes jobs of the Q and sends them to the hardware, then puts it on
2850  * the Q to wait for completion.
2851  */
2852 static void start_io(ctlr_info_t *h)
2853 {
2854         CommandList_struct *c;
2855
2856         while (!hlist_empty(&h->reqQ)) {
2857                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
2858                 /* can't do anything if fifo is full */
2859                 if ((h->access.fifo_full(h))) {
2860                         printk(KERN_WARNING "cciss: fifo full\n");
2861                         break;
2862                 }
2863
2864                 /* Get the first entry from the Request Q */
2865                 removeQ(c);
2866                 h->Qdepth--;
2867
2868                 /* Tell the controller execute command */
2869                 h->access.submit_command(h, c);
2870
2871                 /* Put job onto the completed Q */
2872                 addQ(&h->cmpQ, c);
2873         }
2874 }
2875
2876 /* Assumes that CCISS_LOCK(h->ctlr) is held. */
2877 /* Zeros out the error record and then resends the command back */
2878 /* to the controller */
2879 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2880 {
2881         /* erase the old error information */
2882         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2883
2884         /* add it to software queue and then send it to the controller */
2885         addQ(&h->reqQ, c);
2886         h->Qdepth++;
2887         if (h->Qdepth > h->maxQsinceinit)
2888                 h->maxQsinceinit = h->Qdepth;
2889
2890         start_io(h);
2891 }
2892
2893 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
2894         unsigned int msg_byte, unsigned int host_byte,
2895         unsigned int driver_byte)
2896 {
2897         /* inverse of macros in scsi.h */
2898         return (scsi_status_byte & 0xff) |
2899                 ((msg_byte & 0xff) << 8) |
2900                 ((host_byte & 0xff) << 16) |
2901                 ((driver_byte & 0xff) << 24);
2902 }
2903
2904 static inline int evaluate_target_status(ctlr_info_t *h,
2905                         CommandList_struct *cmd, int *retry_cmd)
2906 {
2907         unsigned char sense_key;
2908         unsigned char status_byte, msg_byte, host_byte, driver_byte;
2909         int error_value;
2910
2911         *retry_cmd = 0;
2912         /* If we get in here, it means we got "target status", that is, scsi status */
2913         status_byte = cmd->err_info->ScsiStatus;
2914         driver_byte = DRIVER_OK;
2915         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
2916
2917         if (blk_pc_request(cmd->rq))
2918                 host_byte = DID_PASSTHROUGH;
2919         else
2920                 host_byte = DID_OK;
2921
2922         error_value = make_status_bytes(status_byte, msg_byte,
2923                 host_byte, driver_byte);
2924
2925         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
2926                 if (!blk_pc_request(cmd->rq))
2927                         printk(KERN_WARNING "cciss: cmd %p "
2928                                "has SCSI Status 0x%x\n",
2929                                cmd, cmd->err_info->ScsiStatus);
2930                 return error_value;
2931         }
2932
2933         /* check the sense key */
2934         sense_key = 0xf & cmd->err_info->SenseInfo[2];
2935         /* no status or recovered error */
2936         if (((sense_key == 0x0) || (sense_key == 0x1)) && !blk_pc_request(cmd->rq))
2937                 error_value = 0;
2938
2939         if (check_for_unit_attention(h, cmd)) {
2940                 *retry_cmd = !blk_pc_request(cmd->rq);
2941                 return 0;
2942         }
2943
2944         if (!blk_pc_request(cmd->rq)) { /* Not SG_IO or similar? */
2945                 if (error_value != 0)
2946                         printk(KERN_WARNING "cciss: cmd %p has CHECK CONDITION"
2947                                " sense key = 0x%x\n", cmd, sense_key);
2948                 return error_value;
2949         }
2950
2951         /* SG_IO or similar, copy sense data back */
2952         if (cmd->rq->sense) {
2953                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
2954                         cmd->rq->sense_len = cmd->err_info->SenseLen;
2955                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
2956                         cmd->rq->sense_len);
2957         } else
2958                 cmd->rq->sense_len = 0;
2959
2960         return error_value;
2961 }
2962
2963 /* checks the status of the job and calls complete buffers to mark all
2964  * buffers for the completed job. Note that this function does not need
2965  * to hold the hba/queue lock.
2966  */
2967 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
2968                                     int timeout)
2969 {
2970         int retry_cmd = 0;
2971         struct request *rq = cmd->rq;
2972
2973         rq->errors = 0;
2974
2975         if (timeout)
2976                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
2977
2978         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
2979                 goto after_error_processing;
2980
2981         switch (cmd->err_info->CommandStatus) {
2982         case CMD_TARGET_STATUS:
2983                 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
2984                 break;
2985         case CMD_DATA_UNDERRUN:
2986                 if (blk_fs_request(cmd->rq)) {
2987                         printk(KERN_WARNING "cciss: cmd %p has"
2988                                " completed with data underrun "
2989                                "reported\n", cmd);
2990                         cmd->rq->resid_len = cmd->err_info->ResidualCnt;
2991                 }
2992                 break;
2993         case CMD_DATA_OVERRUN:
2994                 if (blk_fs_request(cmd->rq))
2995                         printk(KERN_WARNING "cciss: cmd %p has"
2996                                " completed with data overrun "
2997                                "reported\n", cmd);
2998                 break;
2999         case CMD_INVALID:
3000                 printk(KERN_WARNING "cciss: cmd %p is "
3001                        "reported invalid\n", cmd);
3002                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3003                         cmd->err_info->CommandStatus, DRIVER_OK,
3004                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3005                 break;
3006         case CMD_PROTOCOL_ERR:
3007                 printk(KERN_WARNING "cciss: cmd %p has "
3008                        "protocol error \n", cmd);
3009                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3010                         cmd->err_info->CommandStatus, DRIVER_OK,
3011                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3012                 break;
3013         case CMD_HARDWARE_ERR:
3014                 printk(KERN_WARNING "cciss: cmd %p had "
3015                        " hardware error\n", cmd);
3016                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3017                         cmd->err_info->CommandStatus, DRIVER_OK,
3018                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3019                 break;
3020         case CMD_CONNECTION_LOST:
3021                 printk(KERN_WARNING "cciss: cmd %p had "
3022                        "connection lost\n", cmd);
3023                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3024                         cmd->err_info->CommandStatus, DRIVER_OK,
3025                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3026                 break;
3027         case CMD_ABORTED:
3028                 printk(KERN_WARNING "cciss: cmd %p was "
3029                        "aborted\n", cmd);
3030                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3031                         cmd->err_info->CommandStatus, DRIVER_OK,
3032                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
3033                 break;
3034         case CMD_ABORT_FAILED:
3035                 printk(KERN_WARNING "cciss: cmd %p reports "
3036                        "abort failed\n", cmd);
3037                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3038                         cmd->err_info->CommandStatus, DRIVER_OK,
3039                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3040                 break;
3041         case CMD_UNSOLICITED_ABORT:
3042                 printk(KERN_WARNING "cciss%d: unsolicited "
3043                        "abort %p\n", h->ctlr, cmd);
3044                 if (cmd->retry_count < MAX_CMD_RETRIES) {
3045                         retry_cmd = 1;
3046                         printk(KERN_WARNING
3047                                "cciss%d: retrying %p\n", h->ctlr, cmd);
3048                         cmd->retry_count++;
3049                 } else
3050                         printk(KERN_WARNING
3051                                "cciss%d: %p retried too "
3052                                "many times\n", h->ctlr, cmd);
3053                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3054                         cmd->err_info->CommandStatus, DRIVER_OK,
3055                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
3056                 break;
3057         case CMD_TIMEOUT:
3058                 printk(KERN_WARNING "cciss: cmd %p timedout\n", cmd);
3059                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3060                         cmd->err_info->CommandStatus, DRIVER_OK,
3061                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3062                 break;
3063         default:
3064                 printk(KERN_WARNING "cciss: cmd %p returned "
3065                        "unknown status %x\n", cmd,
3066                        cmd->err_info->CommandStatus);
3067                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3068                         cmd->err_info->CommandStatus, DRIVER_OK,
3069                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3070         }
3071
3072 after_error_processing:
3073
3074         /* We need to return this command */
3075         if (retry_cmd) {
3076                 resend_cciss_cmd(h, cmd);
3077                 return;
3078         }
3079         cmd->rq->completion_data = cmd;
3080         blk_complete_request(cmd->rq);
3081 }
3082
3083 /*
3084  * Get a request and submit it to the controller.
3085  */
3086 static void do_cciss_request(struct request_queue *q)
3087 {
3088         ctlr_info_t *h = q->queuedata;
3089         CommandList_struct *c;
3090         sector_t start_blk;
3091         int seg;
3092         struct request *creq;
3093         u64bit temp64;
3094         struct scatterlist tmp_sg[MAXSGENTRIES];
3095         drive_info_struct *drv;
3096         int i, dir;
3097
3098         /* We call start_io here in case there is a command waiting on the
3099          * queue that has not been sent.
3100          */
3101         if (blk_queue_plugged(q))
3102                 goto startio;
3103
3104       queue:
3105         creq = blk_peek_request(q);
3106         if (!creq)
3107                 goto startio;
3108
3109         BUG_ON(creq->nr_phys_segments > MAXSGENTRIES);
3110
3111         if ((c = cmd_alloc(h, 1)) == NULL)
3112                 goto full;
3113
3114         blk_start_request(creq);
3115
3116         spin_unlock_irq(q->queue_lock);
3117
3118         c->cmd_type = CMD_RWREQ;
3119         c->rq = creq;
3120
3121         /* fill in the request */
3122         drv = creq->rq_disk->private_data;
3123         c->Header.ReplyQueue = 0;       // unused in simple mode
3124         /* got command from pool, so use the command block index instead */
3125         /* for direct lookups. */
3126         /* The first 2 bits are reserved for controller error reporting. */
3127         c->Header.Tag.lower = (c->cmdindex << 3);
3128         c->Header.Tag.lower |= 0x04;    /* flag for direct lookup. */
3129         c->Header.LUN.LogDev.VolId = drv->LunID;
3130         c->Header.LUN.LogDev.Mode = 1;
3131         c->Request.CDBLen = 10; // 12 byte commands not in FW yet;
3132         c->Request.Type.Type = TYPE_CMD;        // It is a command.
3133         c->Request.Type.Attribute = ATTR_SIMPLE;
3134         c->Request.Type.Direction =
3135             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3136         c->Request.Timeout = 0; // Don't time out
3137         c->Request.CDB[0] =
3138             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3139         start_blk = blk_rq_pos(creq);
3140 #ifdef CCISS_DEBUG
3141         printk(KERN_DEBUG "ciss: sector =%d nr_sectors=%d\n",
3142                (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3143 #endif                          /* CCISS_DEBUG */
3144
3145         sg_init_table(tmp_sg, MAXSGENTRIES);
3146         seg = blk_rq_map_sg(q, creq, tmp_sg);
3147
3148         /* get the DMA records for the setup */
3149         if (c->Request.Type.Direction == XFER_READ)
3150                 dir = PCI_DMA_FROMDEVICE;
3151         else
3152                 dir = PCI_DMA_TODEVICE;
3153
3154         for (i = 0; i < seg; i++) {
3155                 c->SG[i].Len = tmp_sg[i].length;
3156                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3157                                                   tmp_sg[i].offset,
3158                                                   tmp_sg[i].length, dir);
3159                 c->SG[i].Addr.lower = temp64.val32.lower;
3160                 c->SG[i].Addr.upper = temp64.val32.upper;
3161                 c->SG[i].Ext = 0;       // we are not chaining
3162         }
3163         /* track how many SG entries we are using */
3164         if (seg > h->maxSG)
3165                 h->maxSG = seg;
3166
3167 #ifdef CCISS_DEBUG
3168         printk(KERN_DEBUG "cciss: Submitting %u sectors in %d segments\n",
3169                blk_rq_sectors(creq), seg);
3170 #endif                          /* CCISS_DEBUG */
3171
3172         c->Header.SGList = c->Header.SGTotal = seg;
3173         if (likely(blk_fs_request(creq))) {
3174                 if(h->cciss_read == CCISS_READ_10) {
3175                         c->Request.CDB[1] = 0;
3176                         c->Request.CDB[2] = (start_blk >> 24) & 0xff;   //MSB
3177                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3178                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3179                         c->Request.CDB[5] = start_blk & 0xff;
3180                         c->Request.CDB[6] = 0;  // (sect >> 24) & 0xff; MSB
3181                         c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3182                         c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3183                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3184                 } else {
3185                         u32 upper32 = upper_32_bits(start_blk);
3186
3187                         c->Request.CDBLen = 16;
3188                         c->Request.CDB[1]= 0;
3189                         c->Request.CDB[2]= (upper32 >> 24) & 0xff;      //MSB
3190                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3191                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3192                         c->Request.CDB[5]= upper32 & 0xff;
3193                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3194                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3195                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3196                         c->Request.CDB[9]= start_blk & 0xff;
3197                         c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3198                         c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3199                         c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3200                         c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3201                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
3202                 }
3203         } else if (blk_pc_request(creq)) {
3204                 c->Request.CDBLen = creq->cmd_len;
3205                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3206         } else {
3207                 printk(KERN_WARNING "cciss%d: bad request type %d\n", h->ctlr, creq->cmd_type);
3208                 BUG();
3209         }
3210
3211         spin_lock_irq(q->queue_lock);
3212
3213         addQ(&h->reqQ, c);
3214         h->Qdepth++;
3215         if (h->Qdepth > h->maxQsinceinit)
3216                 h->maxQsinceinit = h->Qdepth;
3217
3218         goto queue;
3219 full:
3220         blk_stop_queue(q);
3221 startio:
3222         /* We will already have the driver lock here so not need
3223          * to lock it.
3224          */
3225         start_io(h);
3226 }
3227
3228 static inline unsigned long get_next_completion(ctlr_info_t *h)
3229 {
3230         return h->access.command_completed(h);
3231 }
3232
3233 static inline int interrupt_pending(ctlr_info_t *h)
3234 {
3235         return h->access.intr_pending(h);
3236 }
3237
3238 static inline long interrupt_not_for_us(ctlr_info_t *h)
3239 {
3240         return (((h->access.intr_pending(h) == 0) ||
3241                  (h->interrupts_enabled == 0)));
3242 }
3243
3244 static irqreturn_t do_cciss_intr(int irq, void *dev_id)
3245 {
3246         ctlr_info_t *h = dev_id;
3247         CommandList_struct *c;
3248         unsigned long flags;
3249         __u32 a, a1, a2;
3250
3251         if (interrupt_not_for_us(h))
3252                 return IRQ_NONE;
3253         /*
3254          * If there are completed commands in the completion queue,
3255          * we had better do something about it.
3256          */
3257         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
3258         while (interrupt_pending(h)) {
3259                 while ((a = get_next_completion(h)) != FIFO_EMPTY) {
3260                         a1 = a;
3261                         if ((a & 0x04)) {
3262                                 a2 = (a >> 3);
3263                                 if (a2 >= h->nr_cmds) {
3264                                         printk(KERN_WARNING
3265                                                "cciss: controller cciss%d failed, stopping.\n",
3266                                                h->ctlr);
3267                                         fail_all_cmds(h->ctlr);
3268                                         return IRQ_HANDLED;
3269                                 }
3270
3271                                 c = h->cmd_pool + a2;
3272                                 a = c->busaddr;
3273
3274                         } else {
3275                                 struct hlist_node *tmp;
3276
3277                                 a &= ~3;
3278                                 c = NULL;
3279                                 hlist_for_each_entry(c, tmp, &h->cmpQ, list) {
3280                                         if (c->busaddr == a)
3281                                                 break;
3282                                 }
3283                         }
3284                         /*
3285                          * If we've found the command, take it off the
3286                          * completion Q and free it
3287                          */
3288                         if (c && c->busaddr == a) {
3289                                 removeQ(c);
3290                                 if (c->cmd_type == CMD_RWREQ) {
3291                                         complete_command(h, c, 0);
3292                                 } else if (c->cmd_type == CMD_IOCTL_PEND) {
3293                                         complete(c->waiting);
3294                                 }
3295 #                               ifdef CONFIG_CISS_SCSI_TAPE
3296                                 else if (c->cmd_type == CMD_SCSI)
3297                                         complete_scsi_command(c, 0, a1);
3298 #                               endif
3299                                 continue;
3300                         }
3301                 }
3302         }
3303
3304         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
3305         return IRQ_HANDLED;
3306 }
3307
3308 /**
3309  * add_to_scan_list() - add controller to rescan queue
3310  * @h:                Pointer to the controller.
3311  *
3312  * Adds the controller to the rescan queue if not already on the queue.
3313  *
3314  * returns 1 if added to the queue, 0 if skipped (could be on the
3315  * queue already, or the controller could be initializing or shutting
3316  * down).
3317  **/
3318 static int add_to_scan_list(struct ctlr_info *h)
3319 {
3320         struct ctlr_info *test_h;
3321         int found = 0;
3322         int ret = 0;
3323
3324         if (h->busy_initializing)
3325                 return 0;
3326
3327         if (!mutex_trylock(&h->busy_shutting_down))
3328                 return 0;
3329
3330         mutex_lock(&scan_mutex);
3331         list_for_each_entry(test_h, &scan_q, scan_list) {
3332                 if (test_h == h) {
3333                         found = 1;
3334                         break;
3335                 }
3336         }
3337         if (!found && !h->busy_scanning) {
3338                 INIT_COMPLETION(h->scan_wait);
3339                 list_add_tail(&h->scan_list, &scan_q);
3340                 ret = 1;
3341         }
3342         mutex_unlock(&scan_mutex);
3343         mutex_unlock(&h->busy_shutting_down);
3344
3345         return ret;
3346 }
3347
3348 /**
3349  * remove_from_scan_list() - remove controller from rescan queue
3350  * @h:                     Pointer to the controller.
3351  *
3352  * Removes the controller from the rescan queue if present. Blocks if
3353  * the controller is currently conducting a rescan.
3354  **/
3355 static void remove_from_scan_list(struct ctlr_info *h)
3356 {
3357         struct ctlr_info *test_h, *tmp_h;
3358         int scanning = 0;
3359
3360         mutex_lock(&scan_mutex);
3361         list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3362                 if (test_h == h) {
3363                         list_del(&h->scan_list);
3364                         complete_all(&h->scan_wait);
3365                         mutex_unlock(&scan_mutex);
3366                         return;
3367                 }
3368         }
3369         if (&h->busy_scanning)
3370                 scanning = 0;
3371         mutex_unlock(&scan_mutex);
3372
3373         if (scanning)
3374                 wait_for_completion(&h->scan_wait);
3375 }
3376
3377 /**
3378  * scan_thread() - kernel thread used to rescan controllers
3379  * @data:        Ignored.
3380  *
3381  * A kernel thread used scan for drive topology changes on
3382  * controllers. The thread processes only one controller at a time
3383  * using a queue.  Controllers are added to the queue using
3384  * add_to_scan_list() and removed from the queue either after done
3385  * processing or using remove_from_scan_list().
3386  *
3387  * returns 0.
3388  **/
3389 static int scan_thread(void *data)
3390 {
3391         struct ctlr_info *h;
3392
3393         while (1) {
3394                 set_current_state(TASK_INTERRUPTIBLE);
3395                 schedule();
3396                 if (kthread_should_stop())
3397                         break;
3398
3399                 while (1) {
3400                         mutex_lock(&scan_mutex);
3401                         if (list_empty(&scan_q)) {
3402                                 mutex_unlock(&scan_mutex);
3403                                 break;
3404                         }
3405
3406                         h = list_entry(scan_q.next,
3407                                        struct ctlr_info,
3408                                        scan_list);
3409                         list_del(&h->scan_list);
3410                         h->busy_scanning = 1;
3411                         mutex_unlock(&scan_mutex);
3412
3413                         if (h) {
3414                                 rebuild_lun_table(h, 0);
3415                                 complete_all(&h->scan_wait);
3416                                 mutex_lock(&scan_mutex);
3417                                 h->busy_scanning = 0;
3418                                 mutex_unlock(&scan_mutex);
3419                         }
3420                 }
3421         }
3422
3423         return 0;
3424 }
3425
3426 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3427 {
3428         if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3429                 return 0;
3430
3431         switch (c->err_info->SenseInfo[12]) {
3432         case STATE_CHANGED:
3433                 printk(KERN_WARNING "cciss%d: a state change "
3434                         "detected, command retried\n", h->ctlr);
3435                 return 1;
3436         break;
3437         case LUN_FAILED:
3438                 printk(KERN_WARNING "cciss%d: LUN failure "
3439                         "detected, action required\n", h->ctlr);
3440                 return 1;
3441         break;
3442         case REPORT_LUNS_CHANGED:
3443                 printk(KERN_WARNING "cciss%d: report LUN data "
3444                         "changed\n", h->ctlr);
3445                 add_to_scan_list(h);
3446                 wake_up_process(cciss_scan_thread);
3447                 return 1;
3448         break;
3449         case POWER_OR_RESET:
3450                 printk(KERN_WARNING "cciss%d: a power on "
3451                         "or device reset detected\n", h->ctlr);
3452                 return 1;
3453         break;
3454         case UNIT_ATTENTION_CLEARED:
3455                 printk(KERN_WARNING "cciss%d: unit attention "
3456                     "cleared by another initiator\n", h->ctlr);
3457                 return 1;
3458         break;
3459         default:
3460                 printk(KERN_WARNING "cciss%d: unknown "
3461                         "unit attention detected\n", h->ctlr);
3462                                 return 1;
3463         }
3464 }
3465
3466 /*
3467  *  We cannot read the structure directly, for portability we must use
3468  *   the io functions.
3469  *   This is for debug only.
3470  */
3471 #ifdef CCISS_DEBUG
3472 static void print_cfg_table(CfgTable_struct *tb)
3473 {
3474         int i;
3475         char temp_name[17];
3476
3477         printk("Controller Configuration information\n");
3478         printk("------------------------------------\n");
3479         for (i = 0; i < 4; i++)
3480                 temp_name[i] = readb(&(tb->Signature[i]));
3481         temp_name[4] = '\0';
3482         printk("   Signature = %s\n", temp_name);
3483         printk("   Spec Number = %d\n", readl(&(tb->SpecValence)));
3484         printk("   Transport methods supported = 0x%x\n",
3485                readl(&(tb->TransportSupport)));
3486         printk("   Transport methods active = 0x%x\n",
3487                readl(&(tb->TransportActive)));
3488         printk("   Requested transport Method = 0x%x\n",
3489                readl(&(tb->HostWrite.TransportRequest)));
3490         printk("   Coalesce Interrupt Delay = 0x%x\n",
3491                readl(&(tb->HostWrite.CoalIntDelay)));
3492         printk("   Coalesce Interrupt Count = 0x%x\n",
3493                readl(&(tb->HostWrite.CoalIntCount)));
3494         printk("   Max outstanding commands = 0x%d\n",
3495                readl(&(tb->CmdsOutMax)));
3496         printk("   Bus Types = 0x%x\n", readl(&(tb->BusTypes)));
3497         for (i = 0; i < 16; i++)
3498                 temp_name[i] = readb(&(tb->ServerName[i]));
3499         temp_name[16] = '\0';
3500         printk("   Server Name = %s\n", temp_name);
3501         printk("   Heartbeat Counter = 0x%x\n\n\n", readl(&(tb->HeartBeat)));
3502 }
3503 #endif                          /* CCISS_DEBUG */
3504
3505 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3506 {
3507         int i, offset, mem_type, bar_type;
3508         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3509                 return 0;
3510         offset = 0;
3511         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3512                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3513                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3514                         offset += 4;
3515                 else {
3516                         mem_type = pci_resource_flags(pdev, i) &
3517                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3518                         switch (mem_type) {
3519                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
3520                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3521                                 offset += 4;    /* 32 bit */
3522                                 break;
3523                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
3524                                 offset += 8;
3525                                 break;
3526                         default:        /* reserved in PCI 2.2 */
3527                                 printk(KERN_WARNING
3528                                        "Base address is invalid\n");
3529                                 return -1;
3530                                 break;
3531                         }
3532                 }
3533                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3534                         return i + 1;
3535         }
3536         return -1;
3537 }
3538
3539 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
3540  * controllers that are capable. If not, we use IO-APIC mode.
3541  */
3542
3543 static void __devinit cciss_interrupt_mode(ctlr_info_t *c,
3544                                            struct pci_dev *pdev, __u32 board_id)
3545 {
3546 #ifdef CONFIG_PCI_MSI
3547         int err;
3548         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
3549         {0, 2}, {0, 3}
3550         };
3551
3552         /* Some boards advertise MSI but don't really support it */
3553         if ((board_id == 0x40700E11) ||
3554             (board_id == 0x40800E11) ||
3555             (board_id == 0x40820E11) || (board_id == 0x40830E11))
3556                 goto default_int_mode;
3557
3558         if (pci_find_capability(pdev, PCI_CAP_ID_MSIX)) {
3559                 err = pci_enable_msix(pdev, cciss_msix_entries, 4);
3560                 if (!err) {
3561                         c->intr[0] = cciss_msix_entries[0].vector;
3562                         c->intr[1] = cciss_msix_entries[1].vector;
3563                         c->intr[2] = cciss_msix_entries[2].vector;
3564                         c->intr[3] = cciss_msix_entries[3].vector;
3565                         c->msix_vector = 1;
3566                         return;
3567                 }
3568                 if (err > 0) {
3569                         printk(KERN_WARNING "cciss: only %d MSI-X vectors "
3570                                "available\n", err);
3571                         goto default_int_mode;
3572                 } else {
3573                         printk(KERN_WARNING "cciss: MSI-X init failed %d\n",
3574                                err);
3575                         goto default_int_mode;
3576                 }
3577         }
3578         if (pci_find_capability(pdev, PCI_CAP_ID_MSI)) {
3579                 if (!pci_enable_msi(pdev)) {
3580                         c->msi_vector = 1;
3581                 } else {
3582                         printk(KERN_WARNING "cciss: MSI init failed\n");
3583                 }
3584         }
3585 default_int_mode:
3586 #endif                          /* CONFIG_PCI_MSI */
3587         /* if we get here we're going to use the default interrupt mode */
3588         c->intr[SIMPLE_MODE_INT] = pdev->irq;
3589         return;
3590 }
3591
3592 static int __devinit cciss_pci_init(ctlr_info_t *c, struct pci_dev *pdev)
3593 {
3594         ushort subsystem_vendor_id, subsystem_device_id, command;
3595         __u32 board_id, scratchpad = 0;
3596         __u64 cfg_offset;
3597         __u32 cfg_base_addr;
3598         __u64 cfg_base_addr_index;
3599         int i, err;
3600
3601         /* check to see if controller has been disabled */
3602         /* BEFORE trying to enable it */
3603         (void)pci_read_config_word(pdev, PCI_COMMAND, &command);
3604         if (!(command & 0x02)) {
3605                 printk(KERN_WARNING
3606                        "cciss: controller appears to be disabled\n");
3607                 return -ENODEV;
3608         }
3609
3610         err = pci_enable_device(pdev);
3611         if (err) {
3612                 printk(KERN_ERR "cciss: Unable to Enable PCI device\n");
3613                 return err;
3614         }
3615
3616         err = pci_request_regions(pdev, "cciss");
3617         if (err) {
3618                 printk(KERN_ERR "cciss: Cannot obtain PCI resources, "
3619                        "aborting\n");
3620                 return err;
3621         }
3622
3623         subsystem_vendor_id = pdev->subsystem_vendor;
3624         subsystem_device_id = pdev->subsystem_device;
3625         board_id = (((__u32) (subsystem_device_id << 16) & 0xffff0000) |
3626                     subsystem_vendor_id);
3627
3628 #ifdef CCISS_DEBUG
3629         printk("command = %x\n", command);
3630         printk("irq = %x\n", pdev->irq);
3631         printk("board_id = %x\n", board_id);
3632 #endif                          /* CCISS_DEBUG */
3633
3634 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
3635  * else we use the IO-APIC interrupt assigned to us by system ROM.
3636  */
3637         cciss_interrupt_mode(c, pdev, board_id);
3638
3639         /* find the memory BAR */
3640         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3641                 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM)
3642                         break;
3643         }
3644         if (i == DEVICE_COUNT_RESOURCE) {
3645                 printk(KERN_WARNING "cciss: No memory BAR found\n");
3646                 err = -ENODEV;
3647                 goto err_out_free_res;
3648         }
3649
3650         c->paddr = pci_resource_start(pdev, i); /* addressing mode bits
3651                                                  * already removed
3652                                                  */
3653
3654 #ifdef CCISS_DEBUG
3655         printk("address 0 = %lx\n", c->paddr);
3656 #endif                          /* CCISS_DEBUG */
3657         c->vaddr = remap_pci_mem(c->paddr, 0x250);
3658
3659         /* Wait for the board to become ready.  (PCI hotplug needs this.)
3660          * We poll for up to 120 secs, once per 100ms. */
3661         for (i = 0; i < 1200; i++) {
3662                 scratchpad = readl(c->vaddr + SA5_SCRATCHPAD_OFFSET);
3663                 if (scratchpad == CCISS_FIRMWARE_READY)
3664                         break;
3665                 set_current_state(TASK_INTERRUPTIBLE);
3666                 schedule_timeout(msecs_to_jiffies(100));        /* wait 100ms */
3667         }
3668         if (scratchpad != CCISS_FIRMWARE_READY) {
3669                 printk(KERN_WARNING "cciss: Board not ready.  Timed out.\n");
3670                 err = -ENODEV;
3671                 goto err_out_free_res;
3672         }
3673
3674         /* get the address index number */
3675         cfg_base_addr = readl(c->vaddr + SA5_CTCFG_OFFSET);
3676         cfg_base_addr &= (__u32) 0x0000ffff;
3677 #ifdef CCISS_DEBUG
3678         printk("cfg base address = %x\n", cfg_base_addr);
3679 #endif                          /* CCISS_DEBUG */
3680         cfg_base_addr_index = find_PCI_BAR_index(pdev, cfg_base_addr);
3681 #ifdef CCISS_DEBUG
3682         printk("cfg base address index = %llx\n",
3683                 (unsigned long long)cfg_base_addr_index);
3684 #endif                          /* CCISS_DEBUG */
3685         if (cfg_base_addr_index == -1) {
3686                 printk(KERN_WARNING "cciss: Cannot find cfg_base_addr_index\n");
3687                 err = -ENODEV;
3688                 goto err_out_free_res;
3689         }
3690
3691         cfg_offset = readl(c->vaddr + SA5_CTMEM_OFFSET);
3692 #ifdef CCISS_DEBUG
3693         printk("cfg offset = %llx\n", (unsigned long long)cfg_offset);
3694 #endif                          /* CCISS_DEBUG */
3695         c->cfgtable = remap_pci_mem(pci_resource_start(pdev,
3696                                                        cfg_base_addr_index) +
3697                                     cfg_offset, sizeof(CfgTable_struct));
3698         c->board_id = board_id;
3699
3700 #ifdef CCISS_DEBUG
3701         print_cfg_table(c->cfgtable);
3702 #endif                          /* CCISS_DEBUG */
3703
3704         /* Some controllers support Zero Memory Raid (ZMR).
3705          * When configured in ZMR mode the number of supported
3706          * commands drops to 64. So instead of just setting an
3707          * arbitrary value we make the driver a little smarter.
3708          * We read the config table to tell us how many commands
3709          * are supported on the controller then subtract 4 to
3710          * leave a little room for ioctl calls.
3711          */
3712         c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
3713         for (i = 0; i < ARRAY_SIZE(products); i++) {
3714                 if (board_id == products[i].board_id) {
3715                         c->product_name = products[i].product_name;
3716                         c->access = *(products[i].access);
3717                         c->nr_cmds = c->max_commands - 4;
3718                         break;
3719                 }
3720         }
3721         if ((readb(&c->cfgtable->Signature[0]) != 'C') ||
3722             (readb(&c->cfgtable->Signature[1]) != 'I') ||
3723             (readb(&c->cfgtable->Signature[2]) != 'S') ||
3724             (readb(&c->cfgtable->Signature[3]) != 'S')) {
3725                 printk("Does not appear to be a valid CISS config table\n");
3726                 err = -ENODEV;
3727                 goto err_out_free_res;
3728         }
3729         /* We didn't find the controller in our list. We know the
3730          * signature is valid. If it's an HP device let's try to
3731          * bind to the device and fire it up. Otherwise we bail.
3732          */
3733         if (i == ARRAY_SIZE(products)) {
3734                 if (subsystem_vendor_id == PCI_VENDOR_ID_HP) {
3735                         c->product_name = products[i-1].product_name;
3736                         c->access = *(products[i-1].access);
3737                         c->nr_cmds = c->max_commands - 4;
3738                         printk(KERN_WARNING "cciss: This is an unknown "
3739                                 "Smart Array controller.\n"
3740                                 "cciss: Please update to the latest driver "
3741                                 "available from www.hp.com.\n");
3742                 } else {
3743                         printk(KERN_WARNING "cciss: Sorry, I don't know how"
3744                                 " to access the Smart Array controller %08lx\n"
3745                                         , (unsigned long)board_id);
3746                         err = -ENODEV;
3747                         goto err_out_free_res;
3748                 }
3749         }
3750 #ifdef CONFIG_X86
3751         {
3752                 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
3753                 __u32 prefetch;
3754                 prefetch = readl(&(c->cfgtable->SCSI_Prefetch));
3755                 prefetch |= 0x100;
3756                 writel(prefetch, &(c->cfgtable->SCSI_Prefetch));
3757         }
3758 #endif
3759
3760         /* Disabling DMA prefetch and refetch for the P600.
3761          * An ASIC bug may result in accesses to invalid memory addresses.
3762          * We've disabled prefetch for some time now. Testing with XEN
3763          * kernels revealed a bug in the refetch if dom0 resides on a P600.
3764          */
3765         if(board_id == 0x3225103C) {
3766                 __u32 dma_prefetch;
3767                 __u32 dma_refetch;
3768                 dma_prefetch = readl(c->vaddr + I2O_DMA1_CFG);
3769                 dma_prefetch |= 0x8000;
3770                 writel(dma_prefetch, c->vaddr + I2O_DMA1_CFG);
3771                 pci_read_config_dword(pdev, PCI_COMMAND_PARITY, &dma_refetch);
3772                 dma_refetch |= 0x1;
3773                 pci_write_config_dword(pdev, PCI_COMMAND_PARITY, dma_refetch);
3774         }
3775
3776 #ifdef CCISS_DEBUG
3777         printk("Trying to put board into Simple mode\n");
3778 #endif                          /* CCISS_DEBUG */
3779         c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
3780         /* Update the field, and then ring the doorbell */
3781         writel(CFGTBL_Trans_Simple, &(c->cfgtable->HostWrite.TransportRequest));
3782         writel(CFGTBL_ChangeReq, c->vaddr + SA5_DOORBELL);
3783
3784         /* under certain very rare conditions, this can take awhile.
3785          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3786          * as we enter this code.) */
3787         for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3788                 if (!(readl(c->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3789                         break;
3790                 /* delay and try again */
3791                 set_current_state(TASK_INTERRUPTIBLE);
3792                 schedule_timeout(msecs_to_jiffies(1));
3793         }
3794
3795 #ifdef CCISS_DEBUG
3796         printk(KERN_DEBUG "I counter got to %d %x\n", i,
3797                readl(c->vaddr + SA5_DOORBELL));
3798 #endif                          /* CCISS_DEBUG */
3799 #ifdef CCISS_DEBUG
3800         print_cfg_table(c->cfgtable);
3801 #endif                          /* CCISS_DEBUG */
3802
3803         if (!(readl(&(c->cfgtable->TransportActive)) & CFGTBL_Trans_Simple)) {
3804                 printk(KERN_WARNING "cciss: unable to get board into"
3805                        " simple mode\n");
3806                 err = -ENODEV;
3807                 goto err_out_free_res;
3808         }
3809         return 0;
3810
3811 err_out_free_res:
3812         /*
3813          * Deliberately omit pci_disable_device(): it does something nasty to
3814          * Smart Array controllers that pci_enable_device does not undo
3815          */
3816         pci_release_regions(pdev);
3817         return err;
3818 }
3819
3820 /* Function to find the first free pointer into our hba[] array
3821  * Returns -1 if no free entries are left.
3822  */
3823 static int alloc_cciss_hba(void)
3824 {
3825         int i;
3826
3827         for (i = 0; i < MAX_CTLR; i++) {
3828                 if (!hba[i]) {
3829                         ctlr_info_t *p;
3830
3831                         p = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
3832                         if (!p)
3833                                 goto Enomem;
3834                         hba[i] = p;
3835                         return i;
3836                 }
3837         }
3838         printk(KERN_WARNING "cciss: This driver supports a maximum"
3839                " of %d controllers.\n", MAX_CTLR);
3840         return -1;
3841 Enomem:
3842         printk(KERN_ERR "cciss: out of memory.\n");
3843         return -1;
3844 }
3845
3846 static void free_hba(int i)
3847 {
3848         ctlr_info_t *p = hba[i];
3849         int n;
3850
3851         hba[i] = NULL;
3852         for (n = 0; n < CISS_MAX_LUN; n++)
3853                 put_disk(p->gendisk[n]);
3854         kfree(p);
3855 }
3856
3857 /* Send a message CDB to the firmware. */
3858 static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
3859 {
3860         typedef struct {
3861                 CommandListHeader_struct CommandHeader;
3862                 RequestBlock_struct Request;
3863                 ErrDescriptor_struct ErrorDescriptor;
3864         } Command;
3865         static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
3866         Command *cmd;
3867         dma_addr_t paddr64;
3868         uint32_t paddr32, tag;
3869         void __iomem *vaddr;
3870         int i, err;
3871
3872         vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
3873         if (vaddr == NULL)
3874                 return -ENOMEM;
3875
3876         /* The Inbound Post Queue only accepts 32-bit physical addresses for the
3877            CCISS commands, so they must be allocated from the lower 4GiB of
3878            memory. */
3879         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
3880         if (err) {
3881                 iounmap(vaddr);
3882                 return -ENOMEM;
3883         }
3884
3885         cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
3886         if (cmd == NULL) {
3887                 iounmap(vaddr);
3888                 return -ENOMEM;
3889         }
3890
3891         /* This must fit, because of the 32-bit consistent DMA mask.  Also,
3892            although there's no guarantee, we assume that the address is at
3893            least 4-byte aligned (most likely, it's page-aligned). */
3894         paddr32 = paddr64;
3895
3896         cmd->CommandHeader.ReplyQueue = 0;
3897         cmd->CommandHeader.SGList = 0;
3898         cmd->CommandHeader.SGTotal = 0;
3899         cmd->CommandHeader.Tag.lower = paddr32;
3900         cmd->CommandHeader.Tag.upper = 0;
3901         memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
3902
3903         cmd->Request.CDBLen = 16;
3904         cmd->Request.Type.Type = TYPE_MSG;
3905         cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
3906         cmd->Request.Type.Direction = XFER_NONE;
3907         cmd->Request.Timeout = 0; /* Don't time out */
3908         cmd->Request.CDB[0] = opcode;
3909         cmd->Request.CDB[1] = type;
3910         memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
3911
3912         cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
3913         cmd->ErrorDescriptor.Addr.upper = 0;
3914         cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
3915
3916         writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
3917
3918         for (i = 0; i < 10; i++) {
3919                 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
3920                 if ((tag & ~3) == paddr32)
3921                         break;
3922                 schedule_timeout_uninterruptible(HZ);
3923         }
3924
3925         iounmap(vaddr);
3926
3927         /* we leak the DMA buffer here ... no choice since the controller could
3928            still complete the command. */
3929         if (i == 10) {
3930                 printk(KERN_ERR "cciss: controller message %02x:%02x timed out\n",
3931                         opcode, type);
3932                 return -ETIMEDOUT;
3933         }
3934
3935         pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
3936
3937         if (tag & 2) {
3938                 printk(KERN_ERR "cciss: controller message %02x:%02x failed\n",
3939                         opcode, type);
3940                 return -EIO;
3941         }
3942
3943         printk(KERN_INFO "cciss: controller message %02x:%02x succeeded\n",
3944                 opcode, type);
3945         return 0;
3946 }
3947
3948 #define cciss_soft_reset_controller(p) cciss_message(p, 1, 0)
3949 #define cciss_noop(p) cciss_message(p, 3, 0)
3950
3951 static __devinit int cciss_reset_msi(struct pci_dev *pdev)
3952 {
3953 /* the #defines are stolen from drivers/pci/msi.h. */
3954 #define msi_control_reg(base)           (base + PCI_MSI_FLAGS)
3955 #define PCI_MSIX_FLAGS_ENABLE           (1 << 15)
3956
3957         int pos;
3958         u16 control = 0;
3959
3960         pos = pci_find_capability(pdev, PCI_CAP_ID_MSI);
3961         if (pos) {
3962                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
3963                 if (control & PCI_MSI_FLAGS_ENABLE) {
3964                         printk(KERN_INFO "cciss: resetting MSI\n");
3965                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSI_FLAGS_ENABLE);
3966                 }
3967         }
3968
3969         pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
3970         if (pos) {
3971                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
3972                 if (control & PCI_MSIX_FLAGS_ENABLE) {
3973                         printk(KERN_INFO "cciss: resetting MSI-X\n");
3974                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSIX_FLAGS_ENABLE);
3975                 }
3976         }
3977
3978         return 0;
3979 }
3980
3981 /* This does a hard reset of the controller using PCI power management
3982  * states. */
3983 static __devinit int cciss_hard_reset_controller(struct pci_dev *pdev)
3984 {
3985         u16 pmcsr, saved_config_space[32];
3986         int i, pos;
3987
3988         printk(KERN_INFO "cciss: using PCI PM to reset controller\n");
3989
3990         /* This is very nearly the same thing as
3991
3992            pci_save_state(pci_dev);
3993            pci_set_power_state(pci_dev, PCI_D3hot);
3994            pci_set_power_state(pci_dev, PCI_D0);
3995            pci_restore_state(pci_dev);
3996
3997            but we can't use these nice canned kernel routines on
3998            kexec, because they also check the MSI/MSI-X state in PCI
3999            configuration space and do the wrong thing when it is
4000            set/cleared.  Also, the pci_save/restore_state functions
4001            violate the ordering requirements for restoring the
4002            configuration space from the CCISS document (see the
4003            comment below).  So we roll our own .... */
4004
4005         for (i = 0; i < 32; i++)
4006                 pci_read_config_word(pdev, 2*i, &saved_config_space[i]);
4007
4008         pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4009         if (pos == 0) {
4010                 printk(KERN_ERR "cciss_reset_controller: PCI PM not supported\n");
4011                 return -ENODEV;
4012         }
4013
4014         /* Quoting from the Open CISS Specification: "The Power
4015          * Management Control/Status Register (CSR) controls the power
4016          * state of the device.  The normal operating state is D0,
4017          * CSR=00h.  The software off state is D3, CSR=03h.  To reset
4018          * the controller, place the interface device in D3 then to
4019          * D0, this causes a secondary PCI reset which will reset the
4020          * controller." */
4021
4022         /* enter the D3hot power management state */
4023         pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4024         pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4025         pmcsr |= PCI_D3hot;
4026         pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4027
4028         schedule_timeout_uninterruptible(HZ >> 1);
4029
4030         /* enter the D0 power management state */
4031         pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4032         pmcsr |= PCI_D0;
4033         pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4034
4035         schedule_timeout_uninterruptible(HZ >> 1);
4036
4037         /* Restore the PCI configuration space.  The Open CISS
4038          * Specification says, "Restore the PCI Configuration
4039          * Registers, offsets 00h through 60h. It is important to
4040          * restore the command register, 16-bits at offset 04h,
4041          * last. Do not restore the configuration status register,
4042          * 16-bits at offset 06h."  Note that the offset is 2*i. */
4043         for (i = 0; i < 32; i++) {
4044                 if (i == 2 || i == 3)
4045                         continue;
4046                 pci_write_config_word(pdev, 2*i, saved_config_space[i]);
4047         }
4048         wmb();
4049         pci_write_config_word(pdev, 4, saved_config_space[2]);
4050
4051         return 0;
4052 }
4053
4054 /*
4055  *  This is it.  Find all the controllers and register them.  I really hate
4056  *  stealing all these major device numbers.
4057  *  returns the number of block devices registered.
4058  */
4059 static int __devinit cciss_init_one(struct pci_dev *pdev,
4060                                     const struct pci_device_id *ent)
4061 {
4062         int i;
4063         int j = 0;
4064         int rc;
4065         int dac, return_code;
4066         InquiryData_struct *inq_buff;
4067
4068         if (reset_devices) {
4069                 /* Reset the controller with a PCI power-cycle */
4070                 if (cciss_hard_reset_controller(pdev) || cciss_reset_msi(pdev))
4071                         return -ENODEV;
4072
4073                 /* Now try to get the controller to respond to a no-op. Some
4074                    devices (notably the HP Smart Array 5i Controller) need
4075                    up to 30 seconds to respond. */
4076                 for (i=0; i<30; i++) {
4077                         if (cciss_noop(pdev) == 0)
4078                                 break;
4079
4080                         schedule_timeout_uninterruptible(HZ);
4081                 }
4082                 if (i == 30) {
4083                         printk(KERN_ERR "cciss: controller seems dead\n");
4084                         return -EBUSY;
4085                 }
4086         }
4087
4088         i = alloc_cciss_hba();
4089         if (i < 0)
4090                 return -1;
4091
4092         hba[i]->busy_initializing = 1;
4093         INIT_HLIST_HEAD(&hba[i]->cmpQ);
4094         INIT_HLIST_HEAD(&hba[i]->reqQ);
4095         mutex_init(&hba[i]->busy_shutting_down);
4096
4097         if (cciss_pci_init(hba[i], pdev) != 0)
4098                 goto clean0;
4099
4100         sprintf(hba[i]->devname, "cciss%d", i);
4101         hba[i]->ctlr = i;
4102         hba[i]->pdev = pdev;
4103
4104         init_completion(&hba[i]->scan_wait);
4105
4106         if (cciss_create_hba_sysfs_entry(hba[i]))
4107                 goto clean0;
4108
4109         /* configure PCI DMA stuff */
4110         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
4111                 dac = 1;
4112         else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
4113                 dac = 0;
4114         else {
4115                 printk(KERN_ERR "cciss: no suitable DMA available\n");
4116                 goto clean1;
4117         }
4118
4119         /*
4120          * register with the major number, or get a dynamic major number
4121          * by passing 0 as argument.  This is done for greater than
4122          * 8 controller support.
4123          */
4124         if (i < MAX_CTLR_ORIG)
4125                 hba[i]->major = COMPAQ_CISS_MAJOR + i;
4126         rc = register_blkdev(hba[i]->major, hba[i]->devname);
4127         if (rc == -EBUSY || rc == -EINVAL) {
4128                 printk(KERN_ERR
4129                        "cciss:  Unable to get major number %d for %s "
4130                        "on hba %d\n", hba[i]->major, hba[i]->devname, i);
4131                 goto clean1;
4132         } else {
4133                 if (i >= MAX_CTLR_ORIG)
4134                         hba[i]->major = rc;
4135         }
4136
4137         /* make sure the board interrupts are off */
4138         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_OFF);
4139         if (request_irq(hba[i]->intr[SIMPLE_MODE_INT], do_cciss_intr,
4140                         IRQF_DISABLED | IRQF_SHARED, hba[i]->devname, hba[i])) {
4141                 printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
4142                        hba[i]->intr[SIMPLE_MODE_INT], hba[i]->devname);
4143                 goto clean2;
4144         }
4145
4146         printk(KERN_INFO "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
4147                hba[i]->devname, pdev->device, pci_name(pdev),
4148                hba[i]->intr[SIMPLE_MODE_INT], dac ? "" : " not");
4149
4150         hba[i]->cmd_pool_bits =
4151             kmalloc(DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4152                         * sizeof(unsigned long), GFP_KERNEL);
4153         hba[i]->cmd_pool = (CommandList_struct *)
4154             pci_alloc_consistent(hba[i]->pdev,
4155                     hba[i]->nr_cmds * sizeof(CommandList_struct),
4156                     &(hba[i]->cmd_pool_dhandle));
4157         hba[i]->errinfo_pool = (ErrorInfo_struct *)
4158             pci_alloc_consistent(hba[i]->pdev,
4159                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4160                     &(hba[i]->errinfo_pool_dhandle));
4161         if ((hba[i]->cmd_pool_bits == NULL)
4162             || (hba[i]->cmd_pool == NULL)
4163             || (hba[i]->errinfo_pool == NULL)) {
4164                 printk(KERN_ERR "cciss: out of memory");
4165                 goto clean4;
4166         }
4167         spin_lock_init(&hba[i]->lock);
4168
4169         /* Initialize the pdev driver private data.
4170            have it point to hba[i].  */
4171         pci_set_drvdata(pdev, hba[i]);
4172         /* command and error info recs zeroed out before
4173            they are used */
4174         memset(hba[i]->cmd_pool_bits, 0,
4175                DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4176                         * sizeof(unsigned long));
4177
4178         hba[i]->num_luns = 0;
4179         hba[i]->highest_lun = -1;
4180         for (j = 0; j < CISS_MAX_LUN; j++) {
4181                 hba[i]->drv[j].raid_level = -1;
4182                 hba[i]->drv[j].queue = NULL;
4183                 hba[i]->gendisk[j] = NULL;
4184         }
4185
4186         cciss_scsi_setup(i);
4187
4188         /* Turn the interrupts on so we can service requests */
4189         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_ON);
4190
4191         /* Get the firmware version */
4192         inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
4193         if (inq_buff == NULL) {
4194                 printk(KERN_ERR "cciss: out of memory\n");
4195                 goto clean4;
4196         }
4197
4198         return_code = sendcmd_withirq(CISS_INQUIRY, i, inq_buff,
4199                 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
4200         if (return_code == IO_OK) {
4201                 hba[i]->firm_ver[0] = inq_buff->data_byte[32];
4202                 hba[i]->firm_ver[1] = inq_buff->data_byte[33];
4203                 hba[i]->firm_ver[2] = inq_buff->data_byte[34];
4204                 hba[i]->firm_ver[3] = inq_buff->data_byte[35];
4205         } else {         /* send command failed */
4206                 printk(KERN_WARNING "cciss: unable to determine firmware"
4207                         " version of controller\n");
4208         }
4209         kfree(inq_buff);
4210
4211         cciss_procinit(i);
4212
4213         hba[i]->cciss_max_sectors = 2048;
4214
4215         rebuild_lun_table(hba[i], 1);
4216         hba[i]->busy_initializing = 0;
4217         return 1;
4218
4219 clean4:
4220         kfree(hba[i]->cmd_pool_bits);
4221         if (hba[i]->cmd_pool)
4222                 pci_free_consistent(hba[i]->pdev,
4223                                     hba[i]->nr_cmds * sizeof(CommandList_struct),
4224                                     hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4225         if (hba[i]->errinfo_pool)
4226                 pci_free_consistent(hba[i]->pdev,
4227                                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4228                                     hba[i]->errinfo_pool,
4229                                     hba[i]->errinfo_pool_dhandle);
4230         free_irq(hba[i]->intr[SIMPLE_MODE_INT], hba[i]);
4231 clean2:
4232         unregister_blkdev(hba[i]->major, hba[i]->devname);
4233 clean1:
4234         cciss_destroy_hba_sysfs_entry(hba[i]);
4235 clean0:
4236         hba[i]->busy_initializing = 0;
4237         /* cleanup any queues that may have been initialized */
4238         for (j=0; j <= hba[i]->highest_lun; j++){
4239                 drive_info_struct *drv = &(hba[i]->drv[j]);
4240                 if (drv->queue)
4241                         blk_cleanup_queue(drv->queue);
4242         }
4243         /*
4244          * Deliberately omit pci_disable_device(): it does something nasty to
4245          * Smart Array controllers that pci_enable_device does not undo
4246          */
4247         pci_release_regions(pdev);
4248         pci_set_drvdata(pdev, NULL);
4249         free_hba(i);
4250         return -1;
4251 }
4252
4253 static void cciss_shutdown(struct pci_dev *pdev)
4254 {
4255         ctlr_info_t *tmp_ptr;
4256         int i;
4257         char flush_buf[4];
4258         int return_code;
4259
4260         tmp_ptr = pci_get_drvdata(pdev);
4261         if (tmp_ptr == NULL)
4262                 return;
4263         i = tmp_ptr->ctlr;
4264         if (hba[i] == NULL)
4265                 return;
4266
4267         /* Turn board interrupts off  and send the flush cache command */
4268         /* sendcmd will turn off interrupt, and send the flush...
4269          * To write all data in the battery backed cache to disks */
4270         memset(flush_buf, 0, 4);
4271         return_code = sendcmd(CCISS_CACHE_FLUSH, i, flush_buf, 4, 0,
4272                 CTLR_LUNID, TYPE_CMD);
4273         if (return_code == IO_OK) {
4274                 printk(KERN_INFO "Completed flushing cache on controller %d\n", i);
4275         } else {
4276                 printk(KERN_WARNING "Error flushing cache on controller %d\n", i);
4277         }
4278         free_irq(hba[i]->intr[2], hba[i]);
4279 }
4280
4281 static void __devexit cciss_remove_one(struct pci_dev *pdev)
4282 {
4283         ctlr_info_t *tmp_ptr;
4284         int i, j;
4285
4286         if (pci_get_drvdata(pdev) == NULL) {
4287                 printk(KERN_ERR "cciss: Unable to remove device \n");
4288                 return;
4289         }
4290
4291         tmp_ptr = pci_get_drvdata(pdev);
4292         i = tmp_ptr->ctlr;
4293         if (hba[i] == NULL) {
4294                 printk(KERN_ERR "cciss: device appears to "
4295                        "already be removed \n");
4296                 return;
4297         }
4298
4299         mutex_lock(&hba[i]->busy_shutting_down);
4300
4301         remove_from_scan_list(hba[i]);
4302         remove_proc_entry(hba[i]->devname, proc_cciss);
4303         unregister_blkdev(hba[i]->major, hba[i]->devname);
4304
4305         /* remove it from the disk list */
4306         for (j = 0; j < CISS_MAX_LUN; j++) {
4307                 struct gendisk *disk = hba[i]->gendisk[j];
4308                 if (disk) {
4309                         struct request_queue *q = disk->queue;
4310
4311                         if (disk->flags & GENHD_FL_UP)
4312                                 del_gendisk(disk);
4313                         if (q)
4314                                 blk_cleanup_queue(q);
4315                 }
4316                 if (hba[i]->drv[j].dev != NULL &&
4317                         (j == 0 || hba[i]->drv[j].raid_level != -1))
4318                                 cciss_destroy_ld_sysfs_entry(hba[i], j);
4319
4320         }
4321
4322 #ifdef CONFIG_CISS_SCSI_TAPE
4323         cciss_unregister_scsi(i);       /* unhook from SCSI subsystem */
4324 #endif
4325
4326         cciss_shutdown(pdev);
4327
4328 #ifdef CONFIG_PCI_MSI
4329         if (hba[i]->msix_vector)
4330                 pci_disable_msix(hba[i]->pdev);
4331         else if (hba[i]->msi_vector)
4332                 pci_disable_msi(hba[i]->pdev);
4333 #endif                          /* CONFIG_PCI_MSI */
4334
4335         iounmap(hba[i]->vaddr);
4336
4337         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(CommandList_struct),
4338                             hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4339         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4340                             hba[i]->errinfo_pool, hba[i]->errinfo_pool_dhandle);
4341         kfree(hba[i]->cmd_pool_bits);
4342         /*
4343          * Deliberately omit pci_disable_device(): it does something nasty to
4344          * Smart Array controllers that pci_enable_device does not undo
4345          */
4346         pci_release_regions(pdev);
4347         pci_set_drvdata(pdev, NULL);
4348         cciss_destroy_hba_sysfs_entry(hba[i]);
4349         mutex_unlock(&hba[i]->busy_shutting_down);
4350         free_hba(i);
4351 }
4352
4353 static struct pci_driver cciss_pci_driver = {
4354         .name = "cciss",
4355         .probe = cciss_init_one,
4356         .remove = __devexit_p(cciss_remove_one),
4357         .id_table = cciss_pci_device_id,        /* id_table */
4358         .shutdown = cciss_shutdown,
4359 };
4360
4361 /*
4362  *  This is it.  Register the PCI driver information for the cards we control
4363  *  the OS will call our registered routines when it finds one of our cards.
4364  */
4365 static int __init cciss_init(void)
4366 {
4367         int err;
4368
4369         /*
4370          * The hardware requires that commands are aligned on a 64-bit
4371          * boundary. Given that we use pci_alloc_consistent() to allocate an
4372          * array of them, the size must be a multiple of 8 bytes.
4373          */
4374         BUILD_BUG_ON(sizeof(CommandList_struct) % 8);
4375
4376         printk(KERN_INFO DRIVER_NAME "\n");
4377
4378         err = bus_register(&cciss_bus_type);
4379         if (err)
4380                 return err;
4381
4382         /* Start the scan thread */
4383         cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
4384         if (IS_ERR(cciss_scan_thread)) {
4385                 err = PTR_ERR(cciss_scan_thread);
4386                 goto err_bus_unregister;
4387         }
4388
4389         /* Register for our PCI devices */
4390         err = pci_register_driver(&cciss_pci_driver);
4391         if (err)
4392                 goto err_thread_stop;
4393
4394         return err;
4395
4396 err_thread_stop:
4397         kthread_stop(cciss_scan_thread);
4398 err_bus_unregister:
4399         bus_unregister(&cciss_bus_type);
4400
4401         return err;
4402 }
4403
4404 static void __exit cciss_cleanup(void)
4405 {
4406         int i;
4407
4408         pci_unregister_driver(&cciss_pci_driver);
4409         /* double check that all controller entrys have been removed */
4410         for (i = 0; i < MAX_CTLR; i++) {
4411                 if (hba[i] != NULL) {
4412                         printk(KERN_WARNING "cciss: had to remove"
4413                                " controller %d\n", i);
4414                         cciss_remove_one(hba[i]->pdev);
4415                 }
4416         }
4417         kthread_stop(cciss_scan_thread);
4418         remove_proc_entry("driver/cciss", NULL);
4419         bus_unregister(&cciss_bus_type);
4420 }
4421
4422 static void fail_all_cmds(unsigned long ctlr)
4423 {
4424         /* If we get here, the board is apparently dead. */
4425         ctlr_info_t *h = hba[ctlr];
4426         CommandList_struct *c;
4427         unsigned long flags;
4428
4429         printk(KERN_WARNING "cciss%d: controller not responding.\n", h->ctlr);
4430         h->alive = 0;           /* the controller apparently died... */
4431
4432         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
4433
4434         pci_disable_device(h->pdev);    /* Make sure it is really dead. */
4435
4436         /* move everything off the request queue onto the completed queue */
4437         while (!hlist_empty(&h->reqQ)) {
4438                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
4439                 removeQ(c);
4440                 h->Qdepth--;
4441                 addQ(&h->cmpQ, c);
4442         }
4443
4444         /* Now, fail everything on the completed queue with a HW error */
4445         while (!hlist_empty(&h->cmpQ)) {
4446                 c = hlist_entry(h->cmpQ.first, CommandList_struct, list);
4447                 removeQ(c);
4448                 if (c->cmd_type != CMD_MSG_STALE)
4449                         c->err_info->CommandStatus = CMD_HARDWARE_ERR;
4450                 if (c->cmd_type == CMD_RWREQ) {
4451                         complete_command(h, c, 0);
4452                 } else if (c->cmd_type == CMD_IOCTL_PEND)
4453                         complete(c->waiting);
4454 #ifdef CONFIG_CISS_SCSI_TAPE
4455                 else if (c->cmd_type == CMD_SCSI)
4456                         complete_scsi_command(c, 0, 0);
4457 #endif
4458         }
4459         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
4460         return;
4461 }
4462
4463 module_init(cciss_init);
4464 module_exit(cciss_cleanup);