Merge branch 'next' of git://git.kernel.org/pub/scm/linux/kernel/git/benh/powerpc
[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/hdreg.h>
40 #include <linux/spinlock.h>
41 #include <linux/compat.h>
42 #include <asm/uaccess.h>
43 #include <asm/io.h>
44
45 #include <linux/dma-mapping.h>
46 #include <linux/blkdev.h>
47 #include <linux/genhd.h>
48 #include <linux/completion.h>
49 #include <scsi/scsi.h>
50 #include <scsi/sg.h>
51 #include <scsi/scsi_ioctl.h>
52 #include <linux/cdrom.h>
53 #include <linux/scatterlist.h>
54 #include <linux/kthread.h>
55
56 #define CCISS_DRIVER_VERSION(maj,min,submin) ((maj<<16)|(min<<8)|(submin))
57 #define DRIVER_NAME "HP CISS Driver (v 3.6.20)"
58 #define DRIVER_VERSION CCISS_DRIVER_VERSION(3, 6, 20)
59
60 /* Embedded module documentation macros - see modules.h */
61 MODULE_AUTHOR("Hewlett-Packard Company");
62 MODULE_DESCRIPTION("Driver for HP Smart Array Controllers");
63 MODULE_SUPPORTED_DEVICE("HP SA5i SA5i+ SA532 SA5300 SA5312 SA641 SA642 SA6400"
64                         " SA6i P600 P800 P400 P400i E200 E200i E500 P700m"
65                         " Smart Array G2 Series SAS/SATA Controllers");
66 MODULE_VERSION("3.6.20");
67 MODULE_LICENSE("GPL");
68
69 #include "cciss_cmd.h"
70 #include "cciss.h"
71 #include <linux/cciss_ioctl.h>
72
73 /* define the PCI info for the cards we can control */
74 static const struct pci_device_id cciss_pci_device_id[] = {
75         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISS,  0x0E11, 0x4070},
76         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4080},
77         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4082},
78         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4083},
79         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x4091},
80         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409A},
81         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409B},
82         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409C},
83         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409D},
84         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSA,     0x103C, 0x3225},
85         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3223},
86         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3234},
87         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3235},
88         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3211},
89         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3212},
90         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3213},
91         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3214},
92         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3215},
93         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3237},
94         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x323D},
95         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3241},
96         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3243},
97         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3245},
98         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3247},
99         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3249},
100         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x324A},
101         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x324B},
102         {PCI_VENDOR_ID_HP,     PCI_ANY_ID,      PCI_ANY_ID, PCI_ANY_ID,
103                 PCI_CLASS_STORAGE_RAID << 8, 0xffff << 8, 0},
104         {0,}
105 };
106
107 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
108
109 /*  board_id = Subsystem Device ID & Vendor ID
110  *  product = Marketing Name for the board
111  *  access = Address of the struct of function pointers
112  */
113 static struct board_type products[] = {
114         {0x40700E11, "Smart Array 5300", &SA5_access},
115         {0x40800E11, "Smart Array 5i", &SA5B_access},
116         {0x40820E11, "Smart Array 532", &SA5B_access},
117         {0x40830E11, "Smart Array 5312", &SA5B_access},
118         {0x409A0E11, "Smart Array 641", &SA5_access},
119         {0x409B0E11, "Smart Array 642", &SA5_access},
120         {0x409C0E11, "Smart Array 6400", &SA5_access},
121         {0x409D0E11, "Smart Array 6400 EM", &SA5_access},
122         {0x40910E11, "Smart Array 6i", &SA5_access},
123         {0x3225103C, "Smart Array P600", &SA5_access},
124         {0x3223103C, "Smart Array P800", &SA5_access},
125         {0x3234103C, "Smart Array P400", &SA5_access},
126         {0x3235103C, "Smart Array P400i", &SA5_access},
127         {0x3211103C, "Smart Array E200i", &SA5_access},
128         {0x3212103C, "Smart Array E200", &SA5_access},
129         {0x3213103C, "Smart Array E200i", &SA5_access},
130         {0x3214103C, "Smart Array E200i", &SA5_access},
131         {0x3215103C, "Smart Array E200i", &SA5_access},
132         {0x3237103C, "Smart Array E500", &SA5_access},
133         {0x323D103C, "Smart Array P700m", &SA5_access},
134         {0x3241103C, "Smart Array P212", &SA5_access},
135         {0x3243103C, "Smart Array P410", &SA5_access},
136         {0x3245103C, "Smart Array P410i", &SA5_access},
137         {0x3247103C, "Smart Array P411", &SA5_access},
138         {0x3249103C, "Smart Array P812", &SA5_access},
139         {0x324A103C, "Smart Array P712m", &SA5_access},
140         {0x324B103C, "Smart Array P711m", &SA5_access},
141         {0xFFFF103C, "Unknown Smart Array", &SA5_access},
142 };
143
144 /* How long to wait (in milliseconds) for board to go into simple mode */
145 #define MAX_CONFIG_WAIT 30000
146 #define MAX_IOCTL_CONFIG_WAIT 1000
147
148 /*define how many times we will try a command because of bus resets */
149 #define MAX_CMD_RETRIES 3
150
151 #define MAX_CTLR        32
152
153 /* Originally cciss driver only supports 8 major numbers */
154 #define MAX_CTLR_ORIG   8
155
156 static ctlr_info_t *hba[MAX_CTLR];
157
158 static void do_cciss_request(struct request_queue *q);
159 static irqreturn_t do_cciss_intr(int irq, void *dev_id);
160 static int cciss_open(struct block_device *bdev, fmode_t mode);
161 static int cciss_release(struct gendisk *disk, fmode_t mode);
162 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
163                        unsigned int cmd, unsigned long arg);
164 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
165
166 static int cciss_revalidate(struct gendisk *disk);
167 static int rebuild_lun_table(ctlr_info_t *h, int first_time);
168 static int deregister_disk(ctlr_info_t *h, int drv_index,
169                            int clear_all);
170
171 static void cciss_read_capacity(int ctlr, int logvol, int withirq,
172                         sector_t *total_size, unsigned int *block_size);
173 static void cciss_read_capacity_16(int ctlr, int logvol, int withirq,
174                         sector_t *total_size, unsigned int *block_size);
175 static void cciss_geometry_inquiry(int ctlr, int logvol,
176                         int withirq, sector_t total_size,
177                         unsigned int block_size, InquiryData_struct *inq_buff,
178                                    drive_info_struct *drv);
179 static void __devinit cciss_interrupt_mode(ctlr_info_t *, struct pci_dev *,
180                                            __u32);
181 static void start_io(ctlr_info_t *h);
182 static int sendcmd(__u8 cmd, int ctlr, void *buff, size_t size,
183                    __u8 page_code, unsigned char *scsi3addr, int cmd_type);
184 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
185                         __u8 page_code, unsigned char scsi3addr[],
186                         int cmd_type);
187 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
188         int attempt_retry);
189 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c);
190
191 static void fail_all_cmds(unsigned long ctlr);
192 static int scan_thread(void *data);
193 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c);
194
195 #ifdef CONFIG_PROC_FS
196 static void cciss_procinit(int i);
197 #else
198 static void cciss_procinit(int i)
199 {
200 }
201 #endif                          /* CONFIG_PROC_FS */
202
203 #ifdef CONFIG_COMPAT
204 static int cciss_compat_ioctl(struct block_device *, fmode_t,
205                               unsigned, unsigned long);
206 #endif
207
208 static struct block_device_operations cciss_fops = {
209         .owner = THIS_MODULE,
210         .open = cciss_open,
211         .release = cciss_release,
212         .locked_ioctl = cciss_ioctl,
213         .getgeo = cciss_getgeo,
214 #ifdef CONFIG_COMPAT
215         .compat_ioctl = cciss_compat_ioctl,
216 #endif
217         .revalidate_disk = cciss_revalidate,
218 };
219
220 /*
221  * Enqueuing and dequeuing functions for cmdlists.
222  */
223 static inline void addQ(struct hlist_head *list, CommandList_struct *c)
224 {
225         hlist_add_head(&c->list, list);
226 }
227
228 static inline void removeQ(CommandList_struct *c)
229 {
230         /*
231          * After kexec/dump some commands might still
232          * be in flight, which the firmware will try
233          * to complete. Resetting the firmware doesn't work
234          * with old fw revisions, so we have to mark
235          * them off as 'stale' to prevent the driver from
236          * falling over.
237          */
238         if (WARN_ON(hlist_unhashed(&c->list))) {
239                 c->cmd_type = CMD_MSG_STALE;
240                 return;
241         }
242
243         hlist_del_init(&c->list);
244 }
245
246 #include "cciss_scsi.c"         /* For SCSI tape support */
247
248 #define RAID_UNKNOWN 6
249
250 #ifdef CONFIG_PROC_FS
251
252 /*
253  * Report information about this controller.
254  */
255 #define ENG_GIG 1000000000
256 #define ENG_GIG_FACTOR (ENG_GIG/512)
257 #define ENGAGE_SCSI     "engage scsi"
258 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
259         "UNKNOWN"
260 };
261
262 static struct proc_dir_entry *proc_cciss;
263
264 static void cciss_seq_show_header(struct seq_file *seq)
265 {
266         ctlr_info_t *h = seq->private;
267
268         seq_printf(seq, "%s: HP %s Controller\n"
269                 "Board ID: 0x%08lx\n"
270                 "Firmware Version: %c%c%c%c\n"
271                 "IRQ: %d\n"
272                 "Logical drives: %d\n"
273                 "Current Q depth: %d\n"
274                 "Current # commands on controller: %d\n"
275                 "Max Q depth since init: %d\n"
276                 "Max # commands on controller since init: %d\n"
277                 "Max SG entries since init: %d\n",
278                 h->devname,
279                 h->product_name,
280                 (unsigned long)h->board_id,
281                 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
282                 h->firm_ver[3], (unsigned int)h->intr[SIMPLE_MODE_INT],
283                 h->num_luns,
284                 h->Qdepth, h->commands_outstanding,
285                 h->maxQsinceinit, h->max_outstanding, h->maxSG);
286
287 #ifdef CONFIG_CISS_SCSI_TAPE
288         cciss_seq_tape_report(seq, h->ctlr);
289 #endif /* CONFIG_CISS_SCSI_TAPE */
290 }
291
292 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
293 {
294         ctlr_info_t *h = seq->private;
295         unsigned ctlr = h->ctlr;
296         unsigned long flags;
297
298         /* prevent displaying bogus info during configuration
299          * or deconfiguration of a logical volume
300          */
301         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
302         if (h->busy_configuring) {
303                 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
304                 return ERR_PTR(-EBUSY);
305         }
306         h->busy_configuring = 1;
307         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
308
309         if (*pos == 0)
310                 cciss_seq_show_header(seq);
311
312         return pos;
313 }
314
315 static int cciss_seq_show(struct seq_file *seq, void *v)
316 {
317         sector_t vol_sz, vol_sz_frac;
318         ctlr_info_t *h = seq->private;
319         unsigned ctlr = h->ctlr;
320         loff_t *pos = v;
321         drive_info_struct *drv = &h->drv[*pos];
322
323         if (*pos > h->highest_lun)
324                 return 0;
325
326         if (drv->heads == 0)
327                 return 0;
328
329         vol_sz = drv->nr_blocks;
330         vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
331         vol_sz_frac *= 100;
332         sector_div(vol_sz_frac, ENG_GIG_FACTOR);
333
334         if (drv->raid_level > 5)
335                 drv->raid_level = RAID_UNKNOWN;
336         seq_printf(seq, "cciss/c%dd%d:"
337                         "\t%4u.%02uGB\tRAID %s\n",
338                         ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
339                         raid_label[drv->raid_level]);
340         return 0;
341 }
342
343 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
344 {
345         ctlr_info_t *h = seq->private;
346
347         if (*pos > h->highest_lun)
348                 return NULL;
349         *pos += 1;
350
351         return pos;
352 }
353
354 static void cciss_seq_stop(struct seq_file *seq, void *v)
355 {
356         ctlr_info_t *h = seq->private;
357
358         /* Only reset h->busy_configuring if we succeeded in setting
359          * it during cciss_seq_start. */
360         if (v == ERR_PTR(-EBUSY))
361                 return;
362
363         h->busy_configuring = 0;
364 }
365
366 static struct seq_operations cciss_seq_ops = {
367         .start = cciss_seq_start,
368         .show  = cciss_seq_show,
369         .next  = cciss_seq_next,
370         .stop  = cciss_seq_stop,
371 };
372
373 static int cciss_seq_open(struct inode *inode, struct file *file)
374 {
375         int ret = seq_open(file, &cciss_seq_ops);
376         struct seq_file *seq = file->private_data;
377
378         if (!ret)
379                 seq->private = PDE(inode)->data;
380
381         return ret;
382 }
383
384 static ssize_t
385 cciss_proc_write(struct file *file, const char __user *buf,
386                  size_t length, loff_t *ppos)
387 {
388         int err;
389         char *buffer;
390
391 #ifndef CONFIG_CISS_SCSI_TAPE
392         return -EINVAL;
393 #endif
394
395         if (!buf || length > PAGE_SIZE - 1)
396                 return -EINVAL;
397
398         buffer = (char *)__get_free_page(GFP_KERNEL);
399         if (!buffer)
400                 return -ENOMEM;
401
402         err = -EFAULT;
403         if (copy_from_user(buffer, buf, length))
404                 goto out;
405         buffer[length] = '\0';
406
407 #ifdef CONFIG_CISS_SCSI_TAPE
408         if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
409                 struct seq_file *seq = file->private_data;
410                 ctlr_info_t *h = seq->private;
411                 int rc;
412
413                 rc = cciss_engage_scsi(h->ctlr);
414                 if (rc != 0)
415                         err = -rc;
416                 else
417                         err = length;
418         } else
419 #endif /* CONFIG_CISS_SCSI_TAPE */
420                 err = -EINVAL;
421         /* might be nice to have "disengage" too, but it's not
422            safely possible. (only 1 module use count, lock issues.) */
423
424 out:
425         free_page((unsigned long)buffer);
426         return err;
427 }
428
429 static struct file_operations cciss_proc_fops = {
430         .owner   = THIS_MODULE,
431         .open    = cciss_seq_open,
432         .read    = seq_read,
433         .llseek  = seq_lseek,
434         .release = seq_release,
435         .write   = cciss_proc_write,
436 };
437
438 static void __devinit cciss_procinit(int i)
439 {
440         struct proc_dir_entry *pde;
441
442         if (proc_cciss == NULL)
443                 proc_cciss = proc_mkdir("driver/cciss", NULL);
444         if (!proc_cciss)
445                 return;
446         pde = proc_create_data(hba[i]->devname, S_IWUSR | S_IRUSR | S_IRGRP |
447                                         S_IROTH, proc_cciss,
448                                         &cciss_proc_fops, hba[i]);
449 }
450 #endif                          /* CONFIG_PROC_FS */
451
452 #define MAX_PRODUCT_NAME_LEN 19
453
454 #define to_hba(n) container_of(n, struct ctlr_info, dev)
455 #define to_drv(n) container_of(n, drive_info_struct, dev)
456
457 static struct device_type cciss_host_type = {
458         .name           = "cciss_host",
459 };
460
461 static ssize_t dev_show_unique_id(struct device *dev,
462                                  struct device_attribute *attr,
463                                  char *buf)
464 {
465         drive_info_struct *drv = to_drv(dev);
466         struct ctlr_info *h = to_hba(drv->dev.parent);
467         __u8 sn[16];
468         unsigned long flags;
469         int ret = 0;
470
471         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
472         if (h->busy_configuring)
473                 ret = -EBUSY;
474         else
475                 memcpy(sn, drv->serial_no, sizeof(sn));
476         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
477
478         if (ret)
479                 return ret;
480         else
481                 return snprintf(buf, 16 * 2 + 2,
482                                 "%02X%02X%02X%02X%02X%02X%02X%02X"
483                                 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
484                                 sn[0], sn[1], sn[2], sn[3],
485                                 sn[4], sn[5], sn[6], sn[7],
486                                 sn[8], sn[9], sn[10], sn[11],
487                                 sn[12], sn[13], sn[14], sn[15]);
488 }
489 DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
490
491 static ssize_t dev_show_vendor(struct device *dev,
492                                struct device_attribute *attr,
493                                char *buf)
494 {
495         drive_info_struct *drv = to_drv(dev);
496         struct ctlr_info *h = to_hba(drv->dev.parent);
497         char vendor[VENDOR_LEN + 1];
498         unsigned long flags;
499         int ret = 0;
500
501         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
502         if (h->busy_configuring)
503                 ret = -EBUSY;
504         else
505                 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
506         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
507
508         if (ret)
509                 return ret;
510         else
511                 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
512 }
513 DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
514
515 static ssize_t dev_show_model(struct device *dev,
516                               struct device_attribute *attr,
517                               char *buf)
518 {
519         drive_info_struct *drv = to_drv(dev);
520         struct ctlr_info *h = to_hba(drv->dev.parent);
521         char model[MODEL_LEN + 1];
522         unsigned long flags;
523         int ret = 0;
524
525         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
526         if (h->busy_configuring)
527                 ret = -EBUSY;
528         else
529                 memcpy(model, drv->model, MODEL_LEN + 1);
530         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
531
532         if (ret)
533                 return ret;
534         else
535                 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
536 }
537 DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
538
539 static ssize_t dev_show_rev(struct device *dev,
540                             struct device_attribute *attr,
541                             char *buf)
542 {
543         drive_info_struct *drv = to_drv(dev);
544         struct ctlr_info *h = to_hba(drv->dev.parent);
545         char rev[REV_LEN + 1];
546         unsigned long flags;
547         int ret = 0;
548
549         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
550         if (h->busy_configuring)
551                 ret = -EBUSY;
552         else
553                 memcpy(rev, drv->rev, REV_LEN + 1);
554         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
555
556         if (ret)
557                 return ret;
558         else
559                 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
560 }
561 DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
562
563 static struct attribute *cciss_dev_attrs[] = {
564         &dev_attr_unique_id.attr,
565         &dev_attr_model.attr,
566         &dev_attr_vendor.attr,
567         &dev_attr_rev.attr,
568         NULL
569 };
570
571 static struct attribute_group cciss_dev_attr_group = {
572         .attrs = cciss_dev_attrs,
573 };
574
575 static struct attribute_group *cciss_dev_attr_groups[] = {
576         &cciss_dev_attr_group,
577         NULL
578 };
579
580 static struct device_type cciss_dev_type = {
581         .name           = "cciss_device",
582         .groups         = cciss_dev_attr_groups,
583 };
584
585 static struct bus_type cciss_bus_type = {
586         .name           = "cciss",
587 };
588
589
590 /*
591  * Initialize sysfs entry for each controller.  This sets up and registers
592  * the 'cciss#' directory for each individual controller under
593  * /sys/bus/pci/devices/<dev>/.
594  */
595 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
596 {
597         device_initialize(&h->dev);
598         h->dev.type = &cciss_host_type;
599         h->dev.bus = &cciss_bus_type;
600         dev_set_name(&h->dev, "%s", h->devname);
601         h->dev.parent = &h->pdev->dev;
602
603         return device_add(&h->dev);
604 }
605
606 /*
607  * Remove sysfs entries for an hba.
608  */
609 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
610 {
611         device_del(&h->dev);
612 }
613
614 /*
615  * Initialize sysfs for each logical drive.  This sets up and registers
616  * the 'c#d#' directory for each individual logical drive under
617  * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
618  * /sys/block/cciss!c#d# to this entry.
619  */
620 static int cciss_create_ld_sysfs_entry(struct ctlr_info *h,
621                                        drive_info_struct *drv,
622                                        int drv_index)
623 {
624         device_initialize(&drv->dev);
625         drv->dev.type = &cciss_dev_type;
626         drv->dev.bus = &cciss_bus_type;
627         dev_set_name(&drv->dev, "c%dd%d", h->ctlr, drv_index);
628         drv->dev.parent = &h->dev;
629         return device_add(&drv->dev);
630 }
631
632 /*
633  * Remove sysfs entries for a logical drive.
634  */
635 static void cciss_destroy_ld_sysfs_entry(drive_info_struct *drv)
636 {
637         device_del(&drv->dev);
638 }
639
640 /*
641  * For operations that cannot sleep, a command block is allocated at init,
642  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
643  * which ones are free or in use.  For operations that can wait for kmalloc
644  * to possible sleep, this routine can be called with get_from_pool set to 0.
645  * cmd_free() MUST be called with a got_from_pool set to 0 if cmd_alloc was.
646  */
647 static CommandList_struct *cmd_alloc(ctlr_info_t *h, int get_from_pool)
648 {
649         CommandList_struct *c;
650         int i;
651         u64bit temp64;
652         dma_addr_t cmd_dma_handle, err_dma_handle;
653
654         if (!get_from_pool) {
655                 c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
656                         sizeof(CommandList_struct), &cmd_dma_handle);
657                 if (c == NULL)
658                         return NULL;
659                 memset(c, 0, sizeof(CommandList_struct));
660
661                 c->cmdindex = -1;
662
663                 c->err_info = (ErrorInfo_struct *)
664                     pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
665                             &err_dma_handle);
666
667                 if (c->err_info == NULL) {
668                         pci_free_consistent(h->pdev,
669                                 sizeof(CommandList_struct), c, cmd_dma_handle);
670                         return NULL;
671                 }
672                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
673         } else {                /* get it out of the controllers pool */
674
675                 do {
676                         i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
677                         if (i == h->nr_cmds)
678                                 return NULL;
679                 } while (test_and_set_bit
680                          (i & (BITS_PER_LONG - 1),
681                           h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
682 #ifdef CCISS_DEBUG
683                 printk(KERN_DEBUG "cciss: using command buffer %d\n", i);
684 #endif
685                 c = h->cmd_pool + i;
686                 memset(c, 0, sizeof(CommandList_struct));
687                 cmd_dma_handle = h->cmd_pool_dhandle
688                     + i * sizeof(CommandList_struct);
689                 c->err_info = h->errinfo_pool + i;
690                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
691                 err_dma_handle = h->errinfo_pool_dhandle
692                     + i * sizeof(ErrorInfo_struct);
693                 h->nr_allocs++;
694
695                 c->cmdindex = i;
696         }
697
698         INIT_HLIST_NODE(&c->list);
699         c->busaddr = (__u32) cmd_dma_handle;
700         temp64.val = (__u64) err_dma_handle;
701         c->ErrDesc.Addr.lower = temp64.val32.lower;
702         c->ErrDesc.Addr.upper = temp64.val32.upper;
703         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
704
705         c->ctlr = h->ctlr;
706         return c;
707 }
708
709 /*
710  * Frees a command block that was previously allocated with cmd_alloc().
711  */
712 static void cmd_free(ctlr_info_t *h, CommandList_struct *c, int got_from_pool)
713 {
714         int i;
715         u64bit temp64;
716
717         if (!got_from_pool) {
718                 temp64.val32.lower = c->ErrDesc.Addr.lower;
719                 temp64.val32.upper = c->ErrDesc.Addr.upper;
720                 pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
721                                     c->err_info, (dma_addr_t) temp64.val);
722                 pci_free_consistent(h->pdev, sizeof(CommandList_struct),
723                                     c, (dma_addr_t) c->busaddr);
724         } else {
725                 i = c - h->cmd_pool;
726                 clear_bit(i & (BITS_PER_LONG - 1),
727                           h->cmd_pool_bits + (i / BITS_PER_LONG));
728                 h->nr_frees++;
729         }
730 }
731
732 static inline ctlr_info_t *get_host(struct gendisk *disk)
733 {
734         return disk->queue->queuedata;
735 }
736
737 static inline drive_info_struct *get_drv(struct gendisk *disk)
738 {
739         return disk->private_data;
740 }
741
742 /*
743  * Open.  Make sure the device is really there.
744  */
745 static int cciss_open(struct block_device *bdev, fmode_t mode)
746 {
747         ctlr_info_t *host = get_host(bdev->bd_disk);
748         drive_info_struct *drv = get_drv(bdev->bd_disk);
749
750 #ifdef CCISS_DEBUG
751         printk(KERN_DEBUG "cciss_open %s\n", bdev->bd_disk->disk_name);
752 #endif                          /* CCISS_DEBUG */
753
754         if (host->busy_initializing || drv->busy_configuring)
755                 return -EBUSY;
756         /*
757          * Root is allowed to open raw volume zero even if it's not configured
758          * so array config can still work. Root is also allowed to open any
759          * volume that has a LUN ID, so it can issue IOCTL to reread the
760          * disk information.  I don't think I really like this
761          * but I'm already using way to many device nodes to claim another one
762          * for "raw controller".
763          */
764         if (drv->heads == 0) {
765                 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
766                         /* if not node 0 make sure it is a partition = 0 */
767                         if (MINOR(bdev->bd_dev) & 0x0f) {
768                                 return -ENXIO;
769                                 /* if it is, make sure we have a LUN ID */
770                         } else if (drv->LunID == 0) {
771                                 return -ENXIO;
772                         }
773                 }
774                 if (!capable(CAP_SYS_ADMIN))
775                         return -EPERM;
776         }
777         drv->usage_count++;
778         host->usage_count++;
779         return 0;
780 }
781
782 /*
783  * Close.  Sync first.
784  */
785 static int cciss_release(struct gendisk *disk, fmode_t mode)
786 {
787         ctlr_info_t *host = get_host(disk);
788         drive_info_struct *drv = get_drv(disk);
789
790 #ifdef CCISS_DEBUG
791         printk(KERN_DEBUG "cciss_release %s\n", disk->disk_name);
792 #endif                          /* CCISS_DEBUG */
793
794         drv->usage_count--;
795         host->usage_count--;
796         return 0;
797 }
798
799 #ifdef CONFIG_COMPAT
800
801 static int do_ioctl(struct block_device *bdev, fmode_t mode,
802                     unsigned cmd, unsigned long arg)
803 {
804         int ret;
805         lock_kernel();
806         ret = cciss_ioctl(bdev, mode, cmd, arg);
807         unlock_kernel();
808         return ret;
809 }
810
811 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
812                                   unsigned cmd, unsigned long arg);
813 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
814                                       unsigned cmd, unsigned long arg);
815
816 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
817                               unsigned cmd, unsigned long arg)
818 {
819         switch (cmd) {
820         case CCISS_GETPCIINFO:
821         case CCISS_GETINTINFO:
822         case CCISS_SETINTINFO:
823         case CCISS_GETNODENAME:
824         case CCISS_SETNODENAME:
825         case CCISS_GETHEARTBEAT:
826         case CCISS_GETBUSTYPES:
827         case CCISS_GETFIRMVER:
828         case CCISS_GETDRIVVER:
829         case CCISS_REVALIDVOLS:
830         case CCISS_DEREGDISK:
831         case CCISS_REGNEWDISK:
832         case CCISS_REGNEWD:
833         case CCISS_RESCANDISK:
834         case CCISS_GETLUNINFO:
835                 return do_ioctl(bdev, mode, cmd, arg);
836
837         case CCISS_PASSTHRU32:
838                 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
839         case CCISS_BIG_PASSTHRU32:
840                 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
841
842         default:
843                 return -ENOIOCTLCMD;
844         }
845 }
846
847 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
848                                   unsigned cmd, unsigned long arg)
849 {
850         IOCTL32_Command_struct __user *arg32 =
851             (IOCTL32_Command_struct __user *) arg;
852         IOCTL_Command_struct arg64;
853         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
854         int err;
855         u32 cp;
856
857         err = 0;
858         err |=
859             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
860                            sizeof(arg64.LUN_info));
861         err |=
862             copy_from_user(&arg64.Request, &arg32->Request,
863                            sizeof(arg64.Request));
864         err |=
865             copy_from_user(&arg64.error_info, &arg32->error_info,
866                            sizeof(arg64.error_info));
867         err |= get_user(arg64.buf_size, &arg32->buf_size);
868         err |= get_user(cp, &arg32->buf);
869         arg64.buf = compat_ptr(cp);
870         err |= copy_to_user(p, &arg64, sizeof(arg64));
871
872         if (err)
873                 return -EFAULT;
874
875         err = do_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
876         if (err)
877                 return err;
878         err |=
879             copy_in_user(&arg32->error_info, &p->error_info,
880                          sizeof(arg32->error_info));
881         if (err)
882                 return -EFAULT;
883         return err;
884 }
885
886 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
887                                       unsigned cmd, unsigned long arg)
888 {
889         BIG_IOCTL32_Command_struct __user *arg32 =
890             (BIG_IOCTL32_Command_struct __user *) arg;
891         BIG_IOCTL_Command_struct arg64;
892         BIG_IOCTL_Command_struct __user *p =
893             compat_alloc_user_space(sizeof(arg64));
894         int err;
895         u32 cp;
896
897         err = 0;
898         err |=
899             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
900                            sizeof(arg64.LUN_info));
901         err |=
902             copy_from_user(&arg64.Request, &arg32->Request,
903                            sizeof(arg64.Request));
904         err |=
905             copy_from_user(&arg64.error_info, &arg32->error_info,
906                            sizeof(arg64.error_info));
907         err |= get_user(arg64.buf_size, &arg32->buf_size);
908         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
909         err |= get_user(cp, &arg32->buf);
910         arg64.buf = compat_ptr(cp);
911         err |= copy_to_user(p, &arg64, sizeof(arg64));
912
913         if (err)
914                 return -EFAULT;
915
916         err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
917         if (err)
918                 return err;
919         err |=
920             copy_in_user(&arg32->error_info, &p->error_info,
921                          sizeof(arg32->error_info));
922         if (err)
923                 return -EFAULT;
924         return err;
925 }
926 #endif
927
928 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
929 {
930         drive_info_struct *drv = get_drv(bdev->bd_disk);
931
932         if (!drv->cylinders)
933                 return -ENXIO;
934
935         geo->heads = drv->heads;
936         geo->sectors = drv->sectors;
937         geo->cylinders = drv->cylinders;
938         return 0;
939 }
940
941 static void check_ioctl_unit_attention(ctlr_info_t *host, CommandList_struct *c)
942 {
943         if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
944                         c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
945                 (void)check_for_unit_attention(host, c);
946 }
947 /*
948  * ioctl
949  */
950 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
951                        unsigned int cmd, unsigned long arg)
952 {
953         struct gendisk *disk = bdev->bd_disk;
954         ctlr_info_t *host = get_host(disk);
955         drive_info_struct *drv = get_drv(disk);
956         int ctlr = host->ctlr;
957         void __user *argp = (void __user *)arg;
958
959 #ifdef CCISS_DEBUG
960         printk(KERN_DEBUG "cciss_ioctl: Called with cmd=%x %lx\n", cmd, arg);
961 #endif                          /* CCISS_DEBUG */
962
963         switch (cmd) {
964         case CCISS_GETPCIINFO:
965                 {
966                         cciss_pci_info_struct pciinfo;
967
968                         if (!arg)
969                                 return -EINVAL;
970                         pciinfo.domain = pci_domain_nr(host->pdev->bus);
971                         pciinfo.bus = host->pdev->bus->number;
972                         pciinfo.dev_fn = host->pdev->devfn;
973                         pciinfo.board_id = host->board_id;
974                         if (copy_to_user
975                             (argp, &pciinfo, sizeof(cciss_pci_info_struct)))
976                                 return -EFAULT;
977                         return 0;
978                 }
979         case CCISS_GETINTINFO:
980                 {
981                         cciss_coalint_struct intinfo;
982                         if (!arg)
983                                 return -EINVAL;
984                         intinfo.delay =
985                             readl(&host->cfgtable->HostWrite.CoalIntDelay);
986                         intinfo.count =
987                             readl(&host->cfgtable->HostWrite.CoalIntCount);
988                         if (copy_to_user
989                             (argp, &intinfo, sizeof(cciss_coalint_struct)))
990                                 return -EFAULT;
991                         return 0;
992                 }
993         case CCISS_SETINTINFO:
994                 {
995                         cciss_coalint_struct intinfo;
996                         unsigned long flags;
997                         int i;
998
999                         if (!arg)
1000                                 return -EINVAL;
1001                         if (!capable(CAP_SYS_ADMIN))
1002                                 return -EPERM;
1003                         if (copy_from_user
1004                             (&intinfo, argp, sizeof(cciss_coalint_struct)))
1005                                 return -EFAULT;
1006                         if ((intinfo.delay == 0) && (intinfo.count == 0))
1007                         {
1008 //                      printk("cciss_ioctl: delay and count cannot be 0\n");
1009                                 return -EINVAL;
1010                         }
1011                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1012                         /* Update the field, and then ring the doorbell */
1013                         writel(intinfo.delay,
1014                                &(host->cfgtable->HostWrite.CoalIntDelay));
1015                         writel(intinfo.count,
1016                                &(host->cfgtable->HostWrite.CoalIntCount));
1017                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1018
1019                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1020                                 if (!(readl(host->vaddr + SA5_DOORBELL)
1021                                       & CFGTBL_ChangeReq))
1022                                         break;
1023                                 /* delay and try again */
1024                                 udelay(1000);
1025                         }
1026                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1027                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1028                                 return -EAGAIN;
1029                         return 0;
1030                 }
1031         case CCISS_GETNODENAME:
1032                 {
1033                         NodeName_type NodeName;
1034                         int i;
1035
1036                         if (!arg)
1037                                 return -EINVAL;
1038                         for (i = 0; i < 16; i++)
1039                                 NodeName[i] =
1040                                     readb(&host->cfgtable->ServerName[i]);
1041                         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1042                                 return -EFAULT;
1043                         return 0;
1044                 }
1045         case CCISS_SETNODENAME:
1046                 {
1047                         NodeName_type NodeName;
1048                         unsigned long flags;
1049                         int i;
1050
1051                         if (!arg)
1052                                 return -EINVAL;
1053                         if (!capable(CAP_SYS_ADMIN))
1054                                 return -EPERM;
1055
1056                         if (copy_from_user
1057                             (NodeName, argp, sizeof(NodeName_type)))
1058                                 return -EFAULT;
1059
1060                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1061
1062                         /* Update the field, and then ring the doorbell */
1063                         for (i = 0; i < 16; i++)
1064                                 writeb(NodeName[i],
1065                                        &host->cfgtable->ServerName[i]);
1066
1067                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1068
1069                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1070                                 if (!(readl(host->vaddr + SA5_DOORBELL)
1071                                       & CFGTBL_ChangeReq))
1072                                         break;
1073                                 /* delay and try again */
1074                                 udelay(1000);
1075                         }
1076                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1077                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1078                                 return -EAGAIN;
1079                         return 0;
1080                 }
1081
1082         case CCISS_GETHEARTBEAT:
1083                 {
1084                         Heartbeat_type heartbeat;
1085
1086                         if (!arg)
1087                                 return -EINVAL;
1088                         heartbeat = readl(&host->cfgtable->HeartBeat);
1089                         if (copy_to_user
1090                             (argp, &heartbeat, sizeof(Heartbeat_type)))
1091                                 return -EFAULT;
1092                         return 0;
1093                 }
1094         case CCISS_GETBUSTYPES:
1095                 {
1096                         BusTypes_type BusTypes;
1097
1098                         if (!arg)
1099                                 return -EINVAL;
1100                         BusTypes = readl(&host->cfgtable->BusTypes);
1101                         if (copy_to_user
1102                             (argp, &BusTypes, sizeof(BusTypes_type)))
1103                                 return -EFAULT;
1104                         return 0;
1105                 }
1106         case CCISS_GETFIRMVER:
1107                 {
1108                         FirmwareVer_type firmware;
1109
1110                         if (!arg)
1111                                 return -EINVAL;
1112                         memcpy(firmware, host->firm_ver, 4);
1113
1114                         if (copy_to_user
1115                             (argp, firmware, sizeof(FirmwareVer_type)))
1116                                 return -EFAULT;
1117                         return 0;
1118                 }
1119         case CCISS_GETDRIVVER:
1120                 {
1121                         DriverVer_type DriverVer = DRIVER_VERSION;
1122
1123                         if (!arg)
1124                                 return -EINVAL;
1125
1126                         if (copy_to_user
1127                             (argp, &DriverVer, sizeof(DriverVer_type)))
1128                                 return -EFAULT;
1129                         return 0;
1130                 }
1131
1132         case CCISS_DEREGDISK:
1133         case CCISS_REGNEWD:
1134         case CCISS_REVALIDVOLS:
1135                 return rebuild_lun_table(host, 0);
1136
1137         case CCISS_GETLUNINFO:{
1138                         LogvolInfo_struct luninfo;
1139
1140                         luninfo.LunID = drv->LunID;
1141                         luninfo.num_opens = drv->usage_count;
1142                         luninfo.num_parts = 0;
1143                         if (copy_to_user(argp, &luninfo,
1144                                          sizeof(LogvolInfo_struct)))
1145                                 return -EFAULT;
1146                         return 0;
1147                 }
1148         case CCISS_PASSTHRU:
1149                 {
1150                         IOCTL_Command_struct iocommand;
1151                         CommandList_struct *c;
1152                         char *buff = NULL;
1153                         u64bit temp64;
1154                         unsigned long flags;
1155                         DECLARE_COMPLETION_ONSTACK(wait);
1156
1157                         if (!arg)
1158                                 return -EINVAL;
1159
1160                         if (!capable(CAP_SYS_RAWIO))
1161                                 return -EPERM;
1162
1163                         if (copy_from_user
1164                             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1165                                 return -EFAULT;
1166                         if ((iocommand.buf_size < 1) &&
1167                             (iocommand.Request.Type.Direction != XFER_NONE)) {
1168                                 return -EINVAL;
1169                         }
1170 #if 0                           /* 'buf_size' member is 16-bits, and always smaller than kmalloc limit */
1171                         /* Check kmalloc limits */
1172                         if (iocommand.buf_size > 128000)
1173                                 return -EINVAL;
1174 #endif
1175                         if (iocommand.buf_size > 0) {
1176                                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1177                                 if (buff == NULL)
1178                                         return -EFAULT;
1179                         }
1180                         if (iocommand.Request.Type.Direction == XFER_WRITE) {
1181                                 /* Copy the data into the buffer we created */
1182                                 if (copy_from_user
1183                                     (buff, iocommand.buf, iocommand.buf_size)) {
1184                                         kfree(buff);
1185                                         return -EFAULT;
1186                                 }
1187                         } else {
1188                                 memset(buff, 0, iocommand.buf_size);
1189                         }
1190                         if ((c = cmd_alloc(host, 0)) == NULL) {
1191                                 kfree(buff);
1192                                 return -ENOMEM;
1193                         }
1194                         // Fill in the command type
1195                         c->cmd_type = CMD_IOCTL_PEND;
1196                         // Fill in Command Header
1197                         c->Header.ReplyQueue = 0;       // unused in simple mode
1198                         if (iocommand.buf_size > 0)     // buffer to fill
1199                         {
1200                                 c->Header.SGList = 1;
1201                                 c->Header.SGTotal = 1;
1202                         } else  // no buffers to fill
1203                         {
1204                                 c->Header.SGList = 0;
1205                                 c->Header.SGTotal = 0;
1206                         }
1207                         c->Header.LUN = iocommand.LUN_info;
1208                         c->Header.Tag.lower = c->busaddr;       // use the kernel address the cmd block for tag
1209
1210                         // Fill in Request block
1211                         c->Request = iocommand.Request;
1212
1213                         // Fill in the scatter gather information
1214                         if (iocommand.buf_size > 0) {
1215                                 temp64.val = pci_map_single(host->pdev, buff,
1216                                         iocommand.buf_size,
1217                                         PCI_DMA_BIDIRECTIONAL);
1218                                 c->SG[0].Addr.lower = temp64.val32.lower;
1219                                 c->SG[0].Addr.upper = temp64.val32.upper;
1220                                 c->SG[0].Len = iocommand.buf_size;
1221                                 c->SG[0].Ext = 0;       // we are not chaining
1222                         }
1223                         c->waiting = &wait;
1224
1225                         /* Put the request on the tail of the request queue */
1226                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1227                         addQ(&host->reqQ, c);
1228                         host->Qdepth++;
1229                         start_io(host);
1230                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1231
1232                         wait_for_completion(&wait);
1233
1234                         /* unlock the buffers from DMA */
1235                         temp64.val32.lower = c->SG[0].Addr.lower;
1236                         temp64.val32.upper = c->SG[0].Addr.upper;
1237                         pci_unmap_single(host->pdev, (dma_addr_t) temp64.val,
1238                                          iocommand.buf_size,
1239                                          PCI_DMA_BIDIRECTIONAL);
1240
1241                         check_ioctl_unit_attention(host, c);
1242
1243                         /* Copy the error information out */
1244                         iocommand.error_info = *(c->err_info);
1245                         if (copy_to_user
1246                             (argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1247                                 kfree(buff);
1248                                 cmd_free(host, c, 0);
1249                                 return -EFAULT;
1250                         }
1251
1252                         if (iocommand.Request.Type.Direction == XFER_READ) {
1253                                 /* Copy the data out of the buffer we created */
1254                                 if (copy_to_user
1255                                     (iocommand.buf, buff, iocommand.buf_size)) {
1256                                         kfree(buff);
1257                                         cmd_free(host, c, 0);
1258                                         return -EFAULT;
1259                                 }
1260                         }
1261                         kfree(buff);
1262                         cmd_free(host, c, 0);
1263                         return 0;
1264                 }
1265         case CCISS_BIG_PASSTHRU:{
1266                         BIG_IOCTL_Command_struct *ioc;
1267                         CommandList_struct *c;
1268                         unsigned char **buff = NULL;
1269                         int *buff_size = NULL;
1270                         u64bit temp64;
1271                         unsigned long flags;
1272                         BYTE sg_used = 0;
1273                         int status = 0;
1274                         int i;
1275                         DECLARE_COMPLETION_ONSTACK(wait);
1276                         __u32 left;
1277                         __u32 sz;
1278                         BYTE __user *data_ptr;
1279
1280                         if (!arg)
1281                                 return -EINVAL;
1282                         if (!capable(CAP_SYS_RAWIO))
1283                                 return -EPERM;
1284                         ioc = (BIG_IOCTL_Command_struct *)
1285                             kmalloc(sizeof(*ioc), GFP_KERNEL);
1286                         if (!ioc) {
1287                                 status = -ENOMEM;
1288                                 goto cleanup1;
1289                         }
1290                         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1291                                 status = -EFAULT;
1292                                 goto cleanup1;
1293                         }
1294                         if ((ioc->buf_size < 1) &&
1295                             (ioc->Request.Type.Direction != XFER_NONE)) {
1296                                 status = -EINVAL;
1297                                 goto cleanup1;
1298                         }
1299                         /* Check kmalloc limits  using all SGs */
1300                         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1301                                 status = -EINVAL;
1302                                 goto cleanup1;
1303                         }
1304                         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1305                                 status = -EINVAL;
1306                                 goto cleanup1;
1307                         }
1308                         buff =
1309                             kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1310                         if (!buff) {
1311                                 status = -ENOMEM;
1312                                 goto cleanup1;
1313                         }
1314                         buff_size = kmalloc(MAXSGENTRIES * sizeof(int),
1315                                                    GFP_KERNEL);
1316                         if (!buff_size) {
1317                                 status = -ENOMEM;
1318                                 goto cleanup1;
1319                         }
1320                         left = ioc->buf_size;
1321                         data_ptr = ioc->buf;
1322                         while (left) {
1323                                 sz = (left >
1324                                       ioc->malloc_size) ? ioc->
1325                                     malloc_size : left;
1326                                 buff_size[sg_used] = sz;
1327                                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1328                                 if (buff[sg_used] == NULL) {
1329                                         status = -ENOMEM;
1330                                         goto cleanup1;
1331                                 }
1332                                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1333                                         if (copy_from_user
1334                                             (buff[sg_used], data_ptr, sz)) {
1335                                                 status = -EFAULT;
1336                                                 goto cleanup1;
1337                                         }
1338                                 } else {
1339                                         memset(buff[sg_used], 0, sz);
1340                                 }
1341                                 left -= sz;
1342                                 data_ptr += sz;
1343                                 sg_used++;
1344                         }
1345                         if ((c = cmd_alloc(host, 0)) == NULL) {
1346                                 status = -ENOMEM;
1347                                 goto cleanup1;
1348                         }
1349                         c->cmd_type = CMD_IOCTL_PEND;
1350                         c->Header.ReplyQueue = 0;
1351
1352                         if (ioc->buf_size > 0) {
1353                                 c->Header.SGList = sg_used;
1354                                 c->Header.SGTotal = sg_used;
1355                         } else {
1356                                 c->Header.SGList = 0;
1357                                 c->Header.SGTotal = 0;
1358                         }
1359                         c->Header.LUN = ioc->LUN_info;
1360                         c->Header.Tag.lower = c->busaddr;
1361
1362                         c->Request = ioc->Request;
1363                         if (ioc->buf_size > 0) {
1364                                 int i;
1365                                 for (i = 0; i < sg_used; i++) {
1366                                         temp64.val =
1367                                             pci_map_single(host->pdev, buff[i],
1368                                                     buff_size[i],
1369                                                     PCI_DMA_BIDIRECTIONAL);
1370                                         c->SG[i].Addr.lower =
1371                                             temp64.val32.lower;
1372                                         c->SG[i].Addr.upper =
1373                                             temp64.val32.upper;
1374                                         c->SG[i].Len = buff_size[i];
1375                                         c->SG[i].Ext = 0;       /* we are not chaining */
1376                                 }
1377                         }
1378                         c->waiting = &wait;
1379                         /* Put the request on the tail of the request queue */
1380                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1381                         addQ(&host->reqQ, c);
1382                         host->Qdepth++;
1383                         start_io(host);
1384                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1385                         wait_for_completion(&wait);
1386                         /* unlock the buffers from DMA */
1387                         for (i = 0; i < sg_used; i++) {
1388                                 temp64.val32.lower = c->SG[i].Addr.lower;
1389                                 temp64.val32.upper = c->SG[i].Addr.upper;
1390                                 pci_unmap_single(host->pdev,
1391                                         (dma_addr_t) temp64.val, buff_size[i],
1392                                         PCI_DMA_BIDIRECTIONAL);
1393                         }
1394                         check_ioctl_unit_attention(host, c);
1395                         /* Copy the error information out */
1396                         ioc->error_info = *(c->err_info);
1397                         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1398                                 cmd_free(host, c, 0);
1399                                 status = -EFAULT;
1400                                 goto cleanup1;
1401                         }
1402                         if (ioc->Request.Type.Direction == XFER_READ) {
1403                                 /* Copy the data out of the buffer we created */
1404                                 BYTE __user *ptr = ioc->buf;
1405                                 for (i = 0; i < sg_used; i++) {
1406                                         if (copy_to_user
1407                                             (ptr, buff[i], buff_size[i])) {
1408                                                 cmd_free(host, c, 0);
1409                                                 status = -EFAULT;
1410                                                 goto cleanup1;
1411                                         }
1412                                         ptr += buff_size[i];
1413                                 }
1414                         }
1415                         cmd_free(host, c, 0);
1416                         status = 0;
1417                       cleanup1:
1418                         if (buff) {
1419                                 for (i = 0; i < sg_used; i++)
1420                                         kfree(buff[i]);
1421                                 kfree(buff);
1422                         }
1423                         kfree(buff_size);
1424                         kfree(ioc);
1425                         return status;
1426                 }
1427
1428         /* scsi_cmd_ioctl handles these, below, though some are not */
1429         /* very meaningful for cciss.  SG_IO is the main one people want. */
1430
1431         case SG_GET_VERSION_NUM:
1432         case SG_SET_TIMEOUT:
1433         case SG_GET_TIMEOUT:
1434         case SG_GET_RESERVED_SIZE:
1435         case SG_SET_RESERVED_SIZE:
1436         case SG_EMULATED_HOST:
1437         case SG_IO:
1438         case SCSI_IOCTL_SEND_COMMAND:
1439                 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
1440
1441         /* scsi_cmd_ioctl would normally handle these, below, but */
1442         /* they aren't a good fit for cciss, as CD-ROMs are */
1443         /* not supported, and we don't have any bus/target/lun */
1444         /* which we present to the kernel. */
1445
1446         case CDROM_SEND_PACKET:
1447         case CDROMCLOSETRAY:
1448         case CDROMEJECT:
1449         case SCSI_IOCTL_GET_IDLUN:
1450         case SCSI_IOCTL_GET_BUS_NUMBER:
1451         default:
1452                 return -ENOTTY;
1453         }
1454 }
1455
1456 static void cciss_check_queues(ctlr_info_t *h)
1457 {
1458         int start_queue = h->next_to_run;
1459         int i;
1460
1461         /* check to see if we have maxed out the number of commands that can
1462          * be placed on the queue.  If so then exit.  We do this check here
1463          * in case the interrupt we serviced was from an ioctl and did not
1464          * free any new commands.
1465          */
1466         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1467                 return;
1468
1469         /* We have room on the queue for more commands.  Now we need to queue
1470          * them up.  We will also keep track of the next queue to run so
1471          * that every queue gets a chance to be started first.
1472          */
1473         for (i = 0; i < h->highest_lun + 1; i++) {
1474                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1475                 /* make sure the disk has been added and the drive is real
1476                  * because this can be called from the middle of init_one.
1477                  */
1478                 if (!(h->drv[curr_queue].queue) || !(h->drv[curr_queue].heads))
1479                         continue;
1480                 blk_start_queue(h->gendisk[curr_queue]->queue);
1481
1482                 /* check to see if we have maxed out the number of commands
1483                  * that can be placed on the queue.
1484                  */
1485                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1486                         if (curr_queue == start_queue) {
1487                                 h->next_to_run =
1488                                     (start_queue + 1) % (h->highest_lun + 1);
1489                                 break;
1490                         } else {
1491                                 h->next_to_run = curr_queue;
1492                                 break;
1493                         }
1494                 }
1495         }
1496 }
1497
1498 static void cciss_softirq_done(struct request *rq)
1499 {
1500         CommandList_struct *cmd = rq->completion_data;
1501         ctlr_info_t *h = hba[cmd->ctlr];
1502         unsigned long flags;
1503         u64bit temp64;
1504         int i, ddir;
1505
1506         if (cmd->Request.Type.Direction == XFER_READ)
1507                 ddir = PCI_DMA_FROMDEVICE;
1508         else
1509                 ddir = PCI_DMA_TODEVICE;
1510
1511         /* command did not need to be retried */
1512         /* unmap the DMA mapping for all the scatter gather elements */
1513         for (i = 0; i < cmd->Header.SGList; i++) {
1514                 temp64.val32.lower = cmd->SG[i].Addr.lower;
1515                 temp64.val32.upper = cmd->SG[i].Addr.upper;
1516                 pci_unmap_page(h->pdev, temp64.val, cmd->SG[i].Len, ddir);
1517         }
1518
1519 #ifdef CCISS_DEBUG
1520         printk("Done with %p\n", rq);
1521 #endif                          /* CCISS_DEBUG */
1522
1523         /* set the residual count for pc requests */
1524         if (blk_pc_request(rq))
1525                 rq->resid_len = cmd->err_info->ResidualCnt;
1526
1527         blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1528
1529         spin_lock_irqsave(&h->lock, flags);
1530         cmd_free(h, cmd, 1);
1531         cciss_check_queues(h);
1532         spin_unlock_irqrestore(&h->lock, flags);
1533 }
1534
1535 static void log_unit_to_scsi3addr(ctlr_info_t *h, unsigned char scsi3addr[],
1536         uint32_t log_unit)
1537 {
1538         log_unit = h->drv[log_unit].LunID & 0x03fff;
1539         memset(&scsi3addr[4], 0, 4);
1540         memcpy(&scsi3addr[0], &log_unit, 4);
1541         scsi3addr[3] |= 0x40;
1542 }
1543
1544 /* This function gets the SCSI vendor, model, and revision of a logical drive
1545  * via the inquiry page 0.  Model, vendor, and rev are set to empty strings if
1546  * they cannot be read.
1547  */
1548 static void cciss_get_device_descr(int ctlr, int logvol, int withirq,
1549                                    char *vendor, char *model, char *rev)
1550 {
1551         int rc;
1552         InquiryData_struct *inq_buf;
1553         unsigned char scsi3addr[8];
1554
1555         *vendor = '\0';
1556         *model = '\0';
1557         *rev = '\0';
1558
1559         inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1560         if (!inq_buf)
1561                 return;
1562
1563         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1564         if (withirq)
1565                 rc = sendcmd_withirq(CISS_INQUIRY, ctlr, inq_buf,
1566                              sizeof(InquiryData_struct), 0,
1567                                 scsi3addr, TYPE_CMD);
1568         else
1569                 rc = sendcmd(CISS_INQUIRY, ctlr, inq_buf,
1570                              sizeof(InquiryData_struct), 0,
1571                                 scsi3addr, TYPE_CMD);
1572         if (rc == IO_OK) {
1573                 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1574                 vendor[VENDOR_LEN] = '\0';
1575                 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1576                 model[MODEL_LEN] = '\0';
1577                 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1578                 rev[REV_LEN] = '\0';
1579         }
1580
1581         kfree(inq_buf);
1582         return;
1583 }
1584
1585 /* This function gets the serial number of a logical drive via
1586  * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1587  * number cannot be had, for whatever reason, 16 bytes of 0xff
1588  * are returned instead.
1589  */
1590 static void cciss_get_serial_no(int ctlr, int logvol, int withirq,
1591                                 unsigned char *serial_no, int buflen)
1592 {
1593 #define PAGE_83_INQ_BYTES 64
1594         int rc;
1595         unsigned char *buf;
1596         unsigned char scsi3addr[8];
1597
1598         if (buflen > 16)
1599                 buflen = 16;
1600         memset(serial_no, 0xff, buflen);
1601         buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1602         if (!buf)
1603                 return;
1604         memset(serial_no, 0, buflen);
1605         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1606         if (withirq)
1607                 rc = sendcmd_withirq(CISS_INQUIRY, ctlr, buf,
1608                         PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1609         else
1610                 rc = sendcmd(CISS_INQUIRY, ctlr, buf,
1611                         PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1612         if (rc == IO_OK)
1613                 memcpy(serial_no, &buf[8], buflen);
1614         kfree(buf);
1615         return;
1616 }
1617
1618 static void cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1619                                 int drv_index)
1620 {
1621         disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1622         sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1623         disk->major = h->major;
1624         disk->first_minor = drv_index << NWD_SHIFT;
1625         disk->fops = &cciss_fops;
1626         disk->private_data = &h->drv[drv_index];
1627         disk->driverfs_dev = &h->drv[drv_index].dev;
1628
1629         /* Set up queue information */
1630         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1631
1632         /* This is a hardware imposed limit. */
1633         blk_queue_max_hw_segments(disk->queue, MAXSGENTRIES);
1634
1635         /* This is a limit in the driver and could be eliminated. */
1636         blk_queue_max_phys_segments(disk->queue, MAXSGENTRIES);
1637
1638         blk_queue_max_sectors(disk->queue, h->cciss_max_sectors);
1639
1640         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1641
1642         disk->queue->queuedata = h;
1643
1644         blk_queue_logical_block_size(disk->queue,
1645                                      h->drv[drv_index].block_size);
1646
1647         /* Make sure all queue data is written out before */
1648         /* setting h->drv[drv_index].queue, as setting this */
1649         /* allows the interrupt handler to start the queue */
1650         wmb();
1651         h->drv[drv_index].queue = disk->queue;
1652         add_disk(disk);
1653 }
1654
1655 /* This function will check the usage_count of the drive to be updated/added.
1656  * If the usage_count is zero and it is a heretofore unknown drive, or,
1657  * the drive's capacity, geometry, or serial number has changed,
1658  * then the drive information will be updated and the disk will be
1659  * re-registered with the kernel.  If these conditions don't hold,
1660  * then it will be left alone for the next reboot.  The exception to this
1661  * is disk 0 which will always be left registered with the kernel since it
1662  * is also the controller node.  Any changes to disk 0 will show up on
1663  * the next reboot.
1664  */
1665 static void cciss_update_drive_info(int ctlr, int drv_index, int first_time)
1666 {
1667         ctlr_info_t *h = hba[ctlr];
1668         struct gendisk *disk;
1669         InquiryData_struct *inq_buff = NULL;
1670         unsigned int block_size;
1671         sector_t total_size;
1672         unsigned long flags = 0;
1673         int ret = 0;
1674         drive_info_struct *drvinfo;
1675         int was_only_controller_node;
1676
1677         /* Get information about the disk and modify the driver structure */
1678         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1679         drvinfo = kmalloc(sizeof(*drvinfo), GFP_KERNEL);
1680         if (inq_buff == NULL || drvinfo == NULL)
1681                 goto mem_msg;
1682
1683         /* See if we're trying to update the "controller node"
1684          * this will happen the when the first logical drive gets
1685          * created by ACU.
1686          */
1687         was_only_controller_node = (drv_index == 0 &&
1688                                 h->drv[0].raid_level == -1);
1689
1690         /* testing to see if 16-byte CDBs are already being used */
1691         if (h->cciss_read == CCISS_READ_16) {
1692                 cciss_read_capacity_16(h->ctlr, drv_index, 1,
1693                         &total_size, &block_size);
1694
1695         } else {
1696                 cciss_read_capacity(ctlr, drv_index, 1,
1697                                     &total_size, &block_size);
1698
1699                 /* if read_capacity returns all F's this volume is >2TB */
1700                 /* in size so we switch to 16-byte CDB's for all */
1701                 /* read/write ops */
1702                 if (total_size == 0xFFFFFFFFULL) {
1703                         cciss_read_capacity_16(ctlr, drv_index, 1,
1704                         &total_size, &block_size);
1705                         h->cciss_read = CCISS_READ_16;
1706                         h->cciss_write = CCISS_WRITE_16;
1707                 } else {
1708                         h->cciss_read = CCISS_READ_10;
1709                         h->cciss_write = CCISS_WRITE_10;
1710                 }
1711         }
1712
1713         cciss_geometry_inquiry(ctlr, drv_index, 1, total_size, block_size,
1714                                inq_buff, drvinfo);
1715         drvinfo->block_size = block_size;
1716         drvinfo->nr_blocks = total_size + 1;
1717
1718         cciss_get_device_descr(ctlr, drv_index, 1, drvinfo->vendor,
1719                                 drvinfo->model, drvinfo->rev);
1720         cciss_get_serial_no(ctlr, drv_index, 1, drvinfo->serial_no,
1721                         sizeof(drvinfo->serial_no));
1722
1723         /* Is it the same disk we already know, and nothing's changed? */
1724         if (h->drv[drv_index].raid_level != -1 &&
1725                 ((memcmp(drvinfo->serial_no,
1726                                 h->drv[drv_index].serial_no, 16) == 0) &&
1727                 drvinfo->block_size == h->drv[drv_index].block_size &&
1728                 drvinfo->nr_blocks == h->drv[drv_index].nr_blocks &&
1729                 drvinfo->heads == h->drv[drv_index].heads &&
1730                 drvinfo->sectors == h->drv[drv_index].sectors &&
1731                 drvinfo->cylinders == h->drv[drv_index].cylinders))
1732                         /* The disk is unchanged, nothing to update */
1733                         goto freeret;
1734
1735         /* If we get here it's not the same disk, or something's changed,
1736          * so we need to * deregister it, and re-register it, if it's not
1737          * in use.
1738          * If the disk already exists then deregister it before proceeding
1739          * (unless it's the first disk (for the controller node).
1740          */
1741         if (h->drv[drv_index].raid_level != -1 && drv_index != 0) {
1742                 printk(KERN_WARNING "disk %d has changed.\n", drv_index);
1743                 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1744                 h->drv[drv_index].busy_configuring = 1;
1745                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1746
1747                 /* deregister_disk sets h->drv[drv_index].queue = NULL
1748                  * which keeps the interrupt handler from starting
1749                  * the queue.
1750                  */
1751                 ret = deregister_disk(h, drv_index, 0);
1752                 h->drv[drv_index].busy_configuring = 0;
1753         }
1754
1755         /* If the disk is in use return */
1756         if (ret)
1757                 goto freeret;
1758
1759         /* Save the new information from cciss_geometry_inquiry
1760          * and serial number inquiry.
1761          */
1762         h->drv[drv_index].block_size = drvinfo->block_size;
1763         h->drv[drv_index].nr_blocks = drvinfo->nr_blocks;
1764         h->drv[drv_index].heads = drvinfo->heads;
1765         h->drv[drv_index].sectors = drvinfo->sectors;
1766         h->drv[drv_index].cylinders = drvinfo->cylinders;
1767         h->drv[drv_index].raid_level = drvinfo->raid_level;
1768         memcpy(h->drv[drv_index].serial_no, drvinfo->serial_no, 16);
1769         memcpy(h->drv[drv_index].vendor, drvinfo->vendor, VENDOR_LEN + 1);
1770         memcpy(h->drv[drv_index].model, drvinfo->model, MODEL_LEN + 1);
1771         memcpy(h->drv[drv_index].rev, drvinfo->rev, REV_LEN + 1);
1772
1773         ++h->num_luns;
1774         disk = h->gendisk[drv_index];
1775         set_capacity(disk, h->drv[drv_index].nr_blocks);
1776
1777         /* If it's not disk 0 (drv_index != 0)
1778          * or if it was disk 0, but there was previously
1779          * no actual corresponding configured logical drive
1780          * (raid_leve == -1) then we want to update the
1781          * logical drive's information.
1782          */
1783         if (drv_index || first_time)
1784                 cciss_add_disk(h, disk, drv_index);
1785
1786 freeret:
1787         kfree(inq_buff);
1788         kfree(drvinfo);
1789         return;
1790 mem_msg:
1791         printk(KERN_ERR "cciss: out of memory\n");
1792         goto freeret;
1793 }
1794
1795 /* This function will find the first index of the controllers drive array
1796  * that has a -1 for the raid_level and will return that index.  This is
1797  * where new drives will be added.  If the index to be returned is greater
1798  * than the highest_lun index for the controller then highest_lun is set
1799  * to this new index.  If there are no available indexes then -1 is returned.
1800  * "controller_node" is used to know if this is a real logical drive, or just
1801  * the controller node, which determines if this counts towards highest_lun.
1802  */
1803 static int cciss_find_free_drive_index(int ctlr, int controller_node)
1804 {
1805         int i;
1806
1807         for (i = 0; i < CISS_MAX_LUN; i++) {
1808                 if (hba[ctlr]->drv[i].raid_level == -1) {
1809                         if (i > hba[ctlr]->highest_lun)
1810                                 if (!controller_node)
1811                                         hba[ctlr]->highest_lun = i;
1812                         return i;
1813                 }
1814         }
1815         return -1;
1816 }
1817
1818 /* cciss_add_gendisk finds a free hba[]->drv structure
1819  * and allocates a gendisk if needed, and sets the lunid
1820  * in the drvinfo structure.   It returns the index into
1821  * the ->drv[] array, or -1 if none are free.
1822  * is_controller_node indicates whether highest_lun should
1823  * count this disk, or if it's only being added to provide
1824  * a means to talk to the controller in case no logical
1825  * drives have yet been configured.
1826  */
1827 static int cciss_add_gendisk(ctlr_info_t *h, __u32 lunid, int controller_node)
1828 {
1829         int drv_index;
1830
1831         drv_index = cciss_find_free_drive_index(h->ctlr, controller_node);
1832         if (drv_index == -1)
1833                 return -1;
1834         /*Check if the gendisk needs to be allocated */
1835         if (!h->gendisk[drv_index]) {
1836                 h->gendisk[drv_index] =
1837                         alloc_disk(1 << NWD_SHIFT);
1838                 if (!h->gendisk[drv_index]) {
1839                         printk(KERN_ERR "cciss%d: could not "
1840                                 "allocate a new disk %d\n",
1841                                 h->ctlr, drv_index);
1842                         return -1;
1843                 }
1844         }
1845         h->drv[drv_index].LunID = lunid;
1846         if (cciss_create_ld_sysfs_entry(h, &h->drv[drv_index], drv_index))
1847                 goto err_free_disk;
1848
1849         /* Don't need to mark this busy because nobody */
1850         /* else knows about this disk yet to contend */
1851         /* for access to it. */
1852         h->drv[drv_index].busy_configuring = 0;
1853         wmb();
1854         return drv_index;
1855
1856 err_free_disk:
1857         put_disk(h->gendisk[drv_index]);
1858         h->gendisk[drv_index] = NULL;
1859         return -1;
1860 }
1861
1862 /* This is for the special case of a controller which
1863  * has no logical drives.  In this case, we still need
1864  * to register a disk so the controller can be accessed
1865  * by the Array Config Utility.
1866  */
1867 static void cciss_add_controller_node(ctlr_info_t *h)
1868 {
1869         struct gendisk *disk;
1870         int drv_index;
1871
1872         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
1873                 return;
1874
1875         drv_index = cciss_add_gendisk(h, 0, 1);
1876         if (drv_index == -1) {
1877                 printk(KERN_WARNING "cciss%d: could not "
1878                         "add disk 0.\n", h->ctlr);
1879                 return;
1880         }
1881         h->drv[drv_index].block_size = 512;
1882         h->drv[drv_index].nr_blocks = 0;
1883         h->drv[drv_index].heads = 0;
1884         h->drv[drv_index].sectors = 0;
1885         h->drv[drv_index].cylinders = 0;
1886         h->drv[drv_index].raid_level = -1;
1887         memset(h->drv[drv_index].serial_no, 0, 16);
1888         disk = h->gendisk[drv_index];
1889         cciss_add_disk(h, disk, drv_index);
1890 }
1891
1892 /* This function will add and remove logical drives from the Logical
1893  * drive array of the controller and maintain persistency of ordering
1894  * so that mount points are preserved until the next reboot.  This allows
1895  * for the removal of logical drives in the middle of the drive array
1896  * without a re-ordering of those drives.
1897  * INPUT
1898  * h            = The controller to perform the operations on
1899  */
1900 static int rebuild_lun_table(ctlr_info_t *h, int first_time)
1901 {
1902         int ctlr = h->ctlr;
1903         int num_luns;
1904         ReportLunData_struct *ld_buff = NULL;
1905         int return_code;
1906         int listlength = 0;
1907         int i;
1908         int drv_found;
1909         int drv_index = 0;
1910         __u32 lunid = 0;
1911         unsigned long flags;
1912
1913         if (!capable(CAP_SYS_RAWIO))
1914                 return -EPERM;
1915
1916         /* Set busy_configuring flag for this operation */
1917         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1918         if (h->busy_configuring) {
1919                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1920                 return -EBUSY;
1921         }
1922         h->busy_configuring = 1;
1923         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1924
1925         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
1926         if (ld_buff == NULL)
1927                 goto mem_msg;
1928
1929         return_code = sendcmd_withirq(CISS_REPORT_LOG, ctlr, ld_buff,
1930                                       sizeof(ReportLunData_struct),
1931                                       0, CTLR_LUNID, TYPE_CMD);
1932
1933         if (return_code == IO_OK)
1934                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
1935         else {  /* reading number of logical volumes failed */
1936                 printk(KERN_WARNING "cciss: report logical volume"
1937                        " command failed\n");
1938                 listlength = 0;
1939                 goto freeret;
1940         }
1941
1942         num_luns = listlength / 8;      /* 8 bytes per entry */
1943         if (num_luns > CISS_MAX_LUN) {
1944                 num_luns = CISS_MAX_LUN;
1945                 printk(KERN_WARNING "cciss: more luns configured"
1946                        " on controller than can be handled by"
1947                        " this driver.\n");
1948         }
1949
1950         if (num_luns == 0)
1951                 cciss_add_controller_node(h);
1952
1953         /* Compare controller drive array to driver's drive array
1954          * to see if any drives are missing on the controller due
1955          * to action of Array Config Utility (user deletes drive)
1956          * and deregister logical drives which have disappeared.
1957          */
1958         for (i = 0; i <= h->highest_lun; i++) {
1959                 int j;
1960                 drv_found = 0;
1961
1962                 /* skip holes in the array from already deleted drives */
1963                 if (h->drv[i].raid_level == -1)
1964                         continue;
1965
1966                 for (j = 0; j < num_luns; j++) {
1967                         memcpy(&lunid, &ld_buff->LUN[j][0], 4);
1968                         lunid = le32_to_cpu(lunid);
1969                         if (h->drv[i].LunID == lunid) {
1970                                 drv_found = 1;
1971                                 break;
1972                         }
1973                 }
1974                 if (!drv_found) {
1975                         /* Deregister it from the OS, it's gone. */
1976                         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1977                         h->drv[i].busy_configuring = 1;
1978                         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1979                         return_code = deregister_disk(h, i, 1);
1980                         cciss_destroy_ld_sysfs_entry(&h->drv[i]);
1981                         h->drv[i].busy_configuring = 0;
1982                 }
1983         }
1984
1985         /* Compare controller drive array to driver's drive array.
1986          * Check for updates in the drive information and any new drives
1987          * on the controller due to ACU adding logical drives, or changing
1988          * a logical drive's size, etc.  Reregister any new/changed drives
1989          */
1990         for (i = 0; i < num_luns; i++) {
1991                 int j;
1992
1993                 drv_found = 0;
1994
1995                 memcpy(&lunid, &ld_buff->LUN[i][0], 4);
1996                 lunid = le32_to_cpu(lunid);
1997
1998                 /* Find if the LUN is already in the drive array
1999                  * of the driver.  If so then update its info
2000                  * if not in use.  If it does not exist then find
2001                  * the first free index and add it.
2002                  */
2003                 for (j = 0; j <= h->highest_lun; j++) {
2004                         if (h->drv[j].raid_level != -1 &&
2005                                 h->drv[j].LunID == lunid) {
2006                                 drv_index = j;
2007                                 drv_found = 1;
2008                                 break;
2009                         }
2010                 }
2011
2012                 /* check if the drive was found already in the array */
2013                 if (!drv_found) {
2014                         drv_index = cciss_add_gendisk(h, lunid, 0);
2015                         if (drv_index == -1)
2016                                 goto freeret;
2017                 }
2018                 cciss_update_drive_info(ctlr, drv_index, first_time);
2019         }               /* end for */
2020
2021 freeret:
2022         kfree(ld_buff);
2023         h->busy_configuring = 0;
2024         /* We return -1 here to tell the ACU that we have registered/updated
2025          * all of the drives that we can and to keep it from calling us
2026          * additional times.
2027          */
2028         return -1;
2029 mem_msg:
2030         printk(KERN_ERR "cciss: out of memory\n");
2031         h->busy_configuring = 0;
2032         goto freeret;
2033 }
2034
2035 /* This function will deregister the disk and it's queue from the
2036  * kernel.  It must be called with the controller lock held and the
2037  * drv structures busy_configuring flag set.  It's parameters are:
2038  *
2039  * disk = This is the disk to be deregistered
2040  * drv  = This is the drive_info_struct associated with the disk to be
2041  *        deregistered.  It contains information about the disk used
2042  *        by the driver.
2043  * clear_all = This flag determines whether or not the disk information
2044  *             is going to be completely cleared out and the highest_lun
2045  *             reset.  Sometimes we want to clear out information about
2046  *             the disk in preparation for re-adding it.  In this case
2047  *             the highest_lun should be left unchanged and the LunID
2048  *             should not be cleared.
2049 */
2050 static int deregister_disk(ctlr_info_t *h, int drv_index,
2051                            int clear_all)
2052 {
2053         int i;
2054         struct gendisk *disk;
2055         drive_info_struct *drv;
2056
2057         if (!capable(CAP_SYS_RAWIO))
2058                 return -EPERM;
2059
2060         drv = &h->drv[drv_index];
2061         disk = h->gendisk[drv_index];
2062
2063         /* make sure logical volume is NOT is use */
2064         if (clear_all || (h->gendisk[0] == disk)) {
2065                 if (drv->usage_count > 1)
2066                         return -EBUSY;
2067         } else if (drv->usage_count > 0)
2068                 return -EBUSY;
2069
2070         /* invalidate the devices and deregister the disk.  If it is disk
2071          * zero do not deregister it but just zero out it's values.  This
2072          * allows us to delete disk zero but keep the controller registered.
2073          */
2074         if (h->gendisk[0] != disk) {
2075                 struct request_queue *q = disk->queue;
2076                 if (disk->flags & GENHD_FL_UP)
2077                         del_gendisk(disk);
2078                 if (q) {
2079                         blk_cleanup_queue(q);
2080                         /* Set drv->queue to NULL so that we do not try
2081                          * to call blk_start_queue on this queue in the
2082                          * interrupt handler
2083                          */
2084                         drv->queue = NULL;
2085                 }
2086                 /* If clear_all is set then we are deleting the logical
2087                  * drive, not just refreshing its info.  For drives
2088                  * other than disk 0 we will call put_disk.  We do not
2089                  * do this for disk 0 as we need it to be able to
2090                  * configure the controller.
2091                  */
2092                 if (clear_all){
2093                         /* This isn't pretty, but we need to find the
2094                          * disk in our array and NULL our the pointer.
2095                          * This is so that we will call alloc_disk if
2096                          * this index is used again later.
2097                          */
2098                         for (i=0; i < CISS_MAX_LUN; i++){
2099                                 if (h->gendisk[i] == disk) {
2100                                         h->gendisk[i] = NULL;
2101                                         break;
2102                                 }
2103                         }
2104                         put_disk(disk);
2105                 }
2106         } else {
2107                 set_capacity(disk, 0);
2108         }
2109
2110         --h->num_luns;
2111         /* zero out the disk size info */
2112         drv->nr_blocks = 0;
2113         drv->block_size = 0;
2114         drv->heads = 0;
2115         drv->sectors = 0;
2116         drv->cylinders = 0;
2117         drv->raid_level = -1;   /* This can be used as a flag variable to
2118                                  * indicate that this element of the drive
2119                                  * array is free.
2120                                  */
2121
2122         if (clear_all) {
2123                 /* check to see if it was the last disk */
2124                 if (drv == h->drv + h->highest_lun) {
2125                         /* if so, find the new hightest lun */
2126                         int i, newhighest = -1;
2127                         for (i = 0; i <= h->highest_lun; i++) {
2128                                 /* if the disk has size > 0, it is available */
2129                                 if (h->drv[i].heads)
2130                                         newhighest = i;
2131                         }
2132                         h->highest_lun = newhighest;
2133                 }
2134
2135                 drv->LunID = 0;
2136         }
2137         return 0;
2138 }
2139
2140 static int fill_cmd(CommandList_struct *c, __u8 cmd, int ctlr, void *buff,
2141                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2142                 int cmd_type)
2143 {
2144         ctlr_info_t *h = hba[ctlr];
2145         u64bit buff_dma_handle;
2146         int status = IO_OK;
2147
2148         c->cmd_type = CMD_IOCTL_PEND;
2149         c->Header.ReplyQueue = 0;
2150         if (buff != NULL) {
2151                 c->Header.SGList = 1;
2152                 c->Header.SGTotal = 1;
2153         } else {
2154                 c->Header.SGList = 0;
2155                 c->Header.SGTotal = 0;
2156         }
2157         c->Header.Tag.lower = c->busaddr;
2158         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2159
2160         c->Request.Type.Type = cmd_type;
2161         if (cmd_type == TYPE_CMD) {
2162                 switch (cmd) {
2163                 case CISS_INQUIRY:
2164                         /* are we trying to read a vital product page */
2165                         if (page_code != 0) {
2166                                 c->Request.CDB[1] = 0x01;
2167                                 c->Request.CDB[2] = page_code;
2168                         }
2169                         c->Request.CDBLen = 6;
2170                         c->Request.Type.Attribute = ATTR_SIMPLE;
2171                         c->Request.Type.Direction = XFER_READ;
2172                         c->Request.Timeout = 0;
2173                         c->Request.CDB[0] = CISS_INQUIRY;
2174                         c->Request.CDB[4] = size & 0xFF;
2175                         break;
2176                 case CISS_REPORT_LOG:
2177                 case CISS_REPORT_PHYS:
2178                         /* Talking to controller so It's a physical command
2179                            mode = 00 target = 0.  Nothing to write.
2180                          */
2181                         c->Request.CDBLen = 12;
2182                         c->Request.Type.Attribute = ATTR_SIMPLE;
2183                         c->Request.Type.Direction = XFER_READ;
2184                         c->Request.Timeout = 0;
2185                         c->Request.CDB[0] = cmd;
2186                         c->Request.CDB[6] = (size >> 24) & 0xFF;        //MSB
2187                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2188                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2189                         c->Request.CDB[9] = size & 0xFF;
2190                         break;
2191
2192                 case CCISS_READ_CAPACITY:
2193                         c->Request.CDBLen = 10;
2194                         c->Request.Type.Attribute = ATTR_SIMPLE;
2195                         c->Request.Type.Direction = XFER_READ;
2196                         c->Request.Timeout = 0;
2197                         c->Request.CDB[0] = cmd;
2198                         break;
2199                 case CCISS_READ_CAPACITY_16:
2200                         c->Request.CDBLen = 16;
2201                         c->Request.Type.Attribute = ATTR_SIMPLE;
2202                         c->Request.Type.Direction = XFER_READ;
2203                         c->Request.Timeout = 0;
2204                         c->Request.CDB[0] = cmd;
2205                         c->Request.CDB[1] = 0x10;
2206                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2207                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2208                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2209                         c->Request.CDB[13] = size & 0xFF;
2210                         c->Request.Timeout = 0;
2211                         c->Request.CDB[0] = cmd;
2212                         break;
2213                 case CCISS_CACHE_FLUSH:
2214                         c->Request.CDBLen = 12;
2215                         c->Request.Type.Attribute = ATTR_SIMPLE;
2216                         c->Request.Type.Direction = XFER_WRITE;
2217                         c->Request.Timeout = 0;
2218                         c->Request.CDB[0] = BMIC_WRITE;
2219                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2220                         break;
2221                 case TEST_UNIT_READY:
2222                         c->Request.CDBLen = 6;
2223                         c->Request.Type.Attribute = ATTR_SIMPLE;
2224                         c->Request.Type.Direction = XFER_NONE;
2225                         c->Request.Timeout = 0;
2226                         break;
2227                 default:
2228                         printk(KERN_WARNING
2229                                "cciss%d:  Unknown Command 0x%c\n", ctlr, cmd);
2230                         return IO_ERROR;
2231                 }
2232         } else if (cmd_type == TYPE_MSG) {
2233                 switch (cmd) {
2234                 case 0: /* ABORT message */
2235                         c->Request.CDBLen = 12;
2236                         c->Request.Type.Attribute = ATTR_SIMPLE;
2237                         c->Request.Type.Direction = XFER_WRITE;
2238                         c->Request.Timeout = 0;
2239                         c->Request.CDB[0] = cmd;        /* abort */
2240                         c->Request.CDB[1] = 0;  /* abort a command */
2241                         /* buff contains the tag of the command to abort */
2242                         memcpy(&c->Request.CDB[4], buff, 8);
2243                         break;
2244                 case 1: /* RESET message */
2245                         c->Request.CDBLen = 16;
2246                         c->Request.Type.Attribute = ATTR_SIMPLE;
2247                         c->Request.Type.Direction = XFER_NONE;
2248                         c->Request.Timeout = 0;
2249                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2250                         c->Request.CDB[0] = cmd;        /* reset */
2251                         c->Request.CDB[1] = 0x03;       /* reset a target */
2252                         break;
2253                 case 3: /* No-Op message */
2254                         c->Request.CDBLen = 1;
2255                         c->Request.Type.Attribute = ATTR_SIMPLE;
2256                         c->Request.Type.Direction = XFER_WRITE;
2257                         c->Request.Timeout = 0;
2258                         c->Request.CDB[0] = cmd;
2259                         break;
2260                 default:
2261                         printk(KERN_WARNING
2262                                "cciss%d: unknown message type %d\n", ctlr, cmd);
2263                         return IO_ERROR;
2264                 }
2265         } else {
2266                 printk(KERN_WARNING
2267                        "cciss%d: unknown command type %d\n", ctlr, cmd_type);
2268                 return IO_ERROR;
2269         }
2270         /* Fill in the scatter gather information */
2271         if (size > 0) {
2272                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2273                                                              buff, size,
2274                                                              PCI_DMA_BIDIRECTIONAL);
2275                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2276                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2277                 c->SG[0].Len = size;
2278                 c->SG[0].Ext = 0;       /* we are not chaining */
2279         }
2280         return status;
2281 }
2282
2283 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2284 {
2285         switch (c->err_info->ScsiStatus) {
2286         case SAM_STAT_GOOD:
2287                 return IO_OK;
2288         case SAM_STAT_CHECK_CONDITION:
2289                 switch (0xf & c->err_info->SenseInfo[2]) {
2290                 case 0: return IO_OK; /* no sense */
2291                 case 1: return IO_OK; /* recovered error */
2292                 default:
2293                         printk(KERN_WARNING "cciss%d: cmd 0x%02x "
2294                                 "check condition, sense key = 0x%02x\n",
2295                                 h->ctlr, c->Request.CDB[0],
2296                                 c->err_info->SenseInfo[2]);
2297                 }
2298                 break;
2299         default:
2300                 printk(KERN_WARNING "cciss%d: cmd 0x%02x"
2301                         "scsi status = 0x%02x\n", h->ctlr,
2302                         c->Request.CDB[0], c->err_info->ScsiStatus);
2303                 break;
2304         }
2305         return IO_ERROR;
2306 }
2307
2308 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2309 {
2310         int return_status = IO_OK;
2311
2312         if (c->err_info->CommandStatus == CMD_SUCCESS)
2313                 return IO_OK;
2314
2315         switch (c->err_info->CommandStatus) {
2316         case CMD_TARGET_STATUS:
2317                 return_status = check_target_status(h, c);
2318                 break;
2319         case CMD_DATA_UNDERRUN:
2320         case CMD_DATA_OVERRUN:
2321                 /* expected for inquiry and report lun commands */
2322                 break;
2323         case CMD_INVALID:
2324                 printk(KERN_WARNING "cciss: cmd 0x%02x is "
2325                        "reported invalid\n", c->Request.CDB[0]);
2326                 return_status = IO_ERROR;
2327                 break;
2328         case CMD_PROTOCOL_ERR:
2329                 printk(KERN_WARNING "cciss: cmd 0x%02x has "
2330                        "protocol error \n", c->Request.CDB[0]);
2331                 return_status = IO_ERROR;
2332                 break;
2333         case CMD_HARDWARE_ERR:
2334                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2335                        " hardware error\n", c->Request.CDB[0]);
2336                 return_status = IO_ERROR;
2337                 break;
2338         case CMD_CONNECTION_LOST:
2339                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2340                        "connection lost\n", c->Request.CDB[0]);
2341                 return_status = IO_ERROR;
2342                 break;
2343         case CMD_ABORTED:
2344                 printk(KERN_WARNING "cciss: cmd 0x%02x was "
2345                        "aborted\n", c->Request.CDB[0]);
2346                 return_status = IO_ERROR;
2347                 break;
2348         case CMD_ABORT_FAILED:
2349                 printk(KERN_WARNING "cciss: cmd 0x%02x reports "
2350                        "abort failed\n", c->Request.CDB[0]);
2351                 return_status = IO_ERROR;
2352                 break;
2353         case CMD_UNSOLICITED_ABORT:
2354                 printk(KERN_WARNING
2355                        "cciss%d: unsolicited abort 0x%02x\n", h->ctlr,
2356                         c->Request.CDB[0]);
2357                 return_status = IO_NEEDS_RETRY;
2358                 break;
2359         default:
2360                 printk(KERN_WARNING "cciss: cmd 0x%02x returned "
2361                        "unknown status %x\n", c->Request.CDB[0],
2362                        c->err_info->CommandStatus);
2363                 return_status = IO_ERROR;
2364         }
2365         return return_status;
2366 }
2367
2368 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2369         int attempt_retry)
2370 {
2371         DECLARE_COMPLETION_ONSTACK(wait);
2372         u64bit buff_dma_handle;
2373         unsigned long flags;
2374         int return_status = IO_OK;
2375
2376 resend_cmd2:
2377         c->waiting = &wait;
2378         /* Put the request on the tail of the queue and send it */
2379         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2380         addQ(&h->reqQ, c);
2381         h->Qdepth++;
2382         start_io(h);
2383         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2384
2385         wait_for_completion(&wait);
2386
2387         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2388                 goto command_done;
2389
2390         return_status = process_sendcmd_error(h, c);
2391
2392         if (return_status == IO_NEEDS_RETRY &&
2393                 c->retry_count < MAX_CMD_RETRIES) {
2394                 printk(KERN_WARNING "cciss%d: retrying 0x%02x\n", h->ctlr,
2395                         c->Request.CDB[0]);
2396                 c->retry_count++;
2397                 /* erase the old error information */
2398                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2399                 return_status = IO_OK;
2400                 INIT_COMPLETION(wait);
2401                 goto resend_cmd2;
2402         }
2403
2404 command_done:
2405         /* unlock the buffers from DMA */
2406         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2407         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2408         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2409                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2410         return return_status;
2411 }
2412
2413 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
2414                            __u8 page_code, unsigned char scsi3addr[],
2415                         int cmd_type)
2416 {
2417         ctlr_info_t *h = hba[ctlr];
2418         CommandList_struct *c;
2419         int return_status;
2420
2421         c = cmd_alloc(h, 0);
2422         if (!c)
2423                 return -ENOMEM;
2424         return_status = fill_cmd(c, cmd, ctlr, buff, size, page_code,
2425                 scsi3addr, cmd_type);
2426         if (return_status == IO_OK)
2427                 return_status = sendcmd_withirq_core(h, c, 1);
2428
2429         cmd_free(h, c, 0);
2430         return return_status;
2431 }
2432
2433 static void cciss_geometry_inquiry(int ctlr, int logvol,
2434                                    int withirq, sector_t total_size,
2435                                    unsigned int block_size,
2436                                    InquiryData_struct *inq_buff,
2437                                    drive_info_struct *drv)
2438 {
2439         int return_code;
2440         unsigned long t;
2441         unsigned char scsi3addr[8];
2442
2443         memset(inq_buff, 0, sizeof(InquiryData_struct));
2444         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2445         if (withirq)
2446                 return_code = sendcmd_withirq(CISS_INQUIRY, ctlr,
2447                                               inq_buff, sizeof(*inq_buff),
2448                                               0xC1, scsi3addr, TYPE_CMD);
2449         else
2450                 return_code = sendcmd(CISS_INQUIRY, ctlr, inq_buff,
2451                                       sizeof(*inq_buff), 0xC1, scsi3addr,
2452                                       TYPE_CMD);
2453         if (return_code == IO_OK) {
2454                 if (inq_buff->data_byte[8] == 0xFF) {
2455                         printk(KERN_WARNING
2456                                "cciss: reading geometry failed, volume "
2457                                "does not support reading geometry\n");
2458                         drv->heads = 255;
2459                         drv->sectors = 32;      // Sectors per track
2460                         drv->cylinders = total_size + 1;
2461                         drv->raid_level = RAID_UNKNOWN;
2462                 } else {
2463                         drv->heads = inq_buff->data_byte[6];
2464                         drv->sectors = inq_buff->data_byte[7];
2465                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2466                         drv->cylinders += inq_buff->data_byte[5];
2467                         drv->raid_level = inq_buff->data_byte[8];
2468                 }
2469                 drv->block_size = block_size;
2470                 drv->nr_blocks = total_size + 1;
2471                 t = drv->heads * drv->sectors;
2472                 if (t > 1) {
2473                         sector_t real_size = total_size + 1;
2474                         unsigned long rem = sector_div(real_size, t);
2475                         if (rem)
2476                                 real_size++;
2477                         drv->cylinders = real_size;
2478                 }
2479         } else {                /* Get geometry failed */
2480                 printk(KERN_WARNING "cciss: reading geometry failed\n");
2481         }
2482         printk(KERN_INFO "      heads=%d, sectors=%d, cylinders=%d\n\n",
2483                drv->heads, drv->sectors, drv->cylinders);
2484 }
2485
2486 static void
2487 cciss_read_capacity(int ctlr, int logvol, int withirq, sector_t *total_size,
2488                     unsigned int *block_size)
2489 {
2490         ReadCapdata_struct *buf;
2491         int return_code;
2492         unsigned char scsi3addr[8];
2493
2494         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2495         if (!buf) {
2496                 printk(KERN_WARNING "cciss: out of memory\n");
2497                 return;
2498         }
2499
2500         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2501         if (withirq)
2502                 return_code = sendcmd_withirq(CCISS_READ_CAPACITY,
2503                                 ctlr, buf, sizeof(ReadCapdata_struct),
2504                                         0, scsi3addr, TYPE_CMD);
2505         else
2506                 return_code = sendcmd(CCISS_READ_CAPACITY,
2507                                 ctlr, buf, sizeof(ReadCapdata_struct),
2508                                         0, scsi3addr, TYPE_CMD);
2509         if (return_code == IO_OK) {
2510                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2511                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2512         } else {                /* read capacity command failed */
2513                 printk(KERN_WARNING "cciss: read capacity failed\n");
2514                 *total_size = 0;
2515                 *block_size = BLOCK_SIZE;
2516         }
2517         if (*total_size != 0)
2518                 printk(KERN_INFO "      blocks= %llu block_size= %d\n",
2519                 (unsigned long long)*total_size+1, *block_size);
2520         kfree(buf);
2521 }
2522
2523 static void
2524 cciss_read_capacity_16(int ctlr, int logvol, int withirq, sector_t *total_size,                                 unsigned int *block_size)
2525 {
2526         ReadCapdata_struct_16 *buf;
2527         int return_code;
2528         unsigned char scsi3addr[8];
2529
2530         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2531         if (!buf) {
2532                 printk(KERN_WARNING "cciss: out of memory\n");
2533                 return;
2534         }
2535
2536         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2537         if (withirq) {
2538                 return_code = sendcmd_withirq(CCISS_READ_CAPACITY_16,
2539                         ctlr, buf, sizeof(ReadCapdata_struct_16),
2540                                 0, scsi3addr, TYPE_CMD);
2541         }
2542         else {
2543                 return_code = sendcmd(CCISS_READ_CAPACITY_16,
2544                         ctlr, buf, sizeof(ReadCapdata_struct_16),
2545                                 0, scsi3addr, TYPE_CMD);
2546         }
2547         if (return_code == IO_OK) {
2548                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2549                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2550         } else {                /* read capacity command failed */
2551                 printk(KERN_WARNING "cciss: read capacity failed\n");
2552                 *total_size = 0;
2553                 *block_size = BLOCK_SIZE;
2554         }
2555         printk(KERN_INFO "      blocks= %llu block_size= %d\n",
2556                (unsigned long long)*total_size+1, *block_size);
2557         kfree(buf);
2558 }
2559
2560 static int cciss_revalidate(struct gendisk *disk)
2561 {
2562         ctlr_info_t *h = get_host(disk);
2563         drive_info_struct *drv = get_drv(disk);
2564         int logvol;
2565         int FOUND = 0;
2566         unsigned int block_size;
2567         sector_t total_size;
2568         InquiryData_struct *inq_buff = NULL;
2569
2570         for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
2571                 if (h->drv[logvol].LunID == drv->LunID) {
2572                         FOUND = 1;
2573                         break;
2574                 }
2575         }
2576
2577         if (!FOUND)
2578                 return 1;
2579
2580         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2581         if (inq_buff == NULL) {
2582                 printk(KERN_WARNING "cciss: out of memory\n");
2583                 return 1;
2584         }
2585         if (h->cciss_read == CCISS_READ_10) {
2586                 cciss_read_capacity(h->ctlr, logvol, 1,
2587                                         &total_size, &block_size);
2588         } else {
2589                 cciss_read_capacity_16(h->ctlr, logvol, 1,
2590                                         &total_size, &block_size);
2591         }
2592         cciss_geometry_inquiry(h->ctlr, logvol, 1, total_size, block_size,
2593                                inq_buff, drv);
2594
2595         blk_queue_logical_block_size(drv->queue, drv->block_size);
2596         set_capacity(disk, drv->nr_blocks);
2597
2598         kfree(inq_buff);
2599         return 0;
2600 }
2601
2602 /*
2603  *   Wait polling for a command to complete.
2604  *   The memory mapped FIFO is polled for the completion.
2605  *   Used only at init time, interrupts from the HBA are disabled.
2606  */
2607 static unsigned long pollcomplete(int ctlr)
2608 {
2609         unsigned long done;
2610         int i;
2611
2612         /* Wait (up to 20 seconds) for a command to complete */
2613
2614         for (i = 20 * HZ; i > 0; i--) {
2615                 done = hba[ctlr]->access.command_completed(hba[ctlr]);
2616                 if (done == FIFO_EMPTY)
2617                         schedule_timeout_uninterruptible(1);
2618                 else
2619                         return done;
2620         }
2621         /* Invalid address to tell caller we ran out of time */
2622         return 1;
2623 }
2624
2625 /* Send command c to controller h and poll for it to complete.
2626  * Turns interrupts off on the board.  Used at driver init time
2627  * and during SCSI error recovery.
2628  */
2629 static int sendcmd_core(ctlr_info_t *h, CommandList_struct *c)
2630 {
2631         int i;
2632         unsigned long complete;
2633         int status = IO_ERROR;
2634         u64bit buff_dma_handle;
2635
2636 resend_cmd1:
2637
2638         /* Disable interrupt on the board. */
2639         h->access.set_intr_mask(h, CCISS_INTR_OFF);
2640
2641         /* Make sure there is room in the command FIFO */
2642         /* Actually it should be completely empty at this time */
2643         /* unless we are in here doing error handling for the scsi */
2644         /* tape side of the driver. */
2645         for (i = 200000; i > 0; i--) {
2646                 /* if fifo isn't full go */
2647                 if (!(h->access.fifo_full(h)))
2648                         break;
2649                 udelay(10);
2650                 printk(KERN_WARNING "cciss cciss%d: SendCmd FIFO full,"
2651                        " waiting!\n", h->ctlr);
2652         }
2653         h->access.submit_command(h, c); /* Send the cmd */
2654         do {
2655                 complete = pollcomplete(h->ctlr);
2656
2657 #ifdef CCISS_DEBUG
2658                 printk(KERN_DEBUG "cciss: command completed\n");
2659 #endif                          /* CCISS_DEBUG */
2660
2661                 if (complete == 1) {
2662                         printk(KERN_WARNING
2663                                "cciss cciss%d: SendCmd Timeout out, "
2664                                "No command list address returned!\n", h->ctlr);
2665                         status = IO_ERROR;
2666                         break;
2667                 }
2668
2669                 /* Make sure it's the command we're expecting. */
2670                 if ((complete & ~CISS_ERROR_BIT) != c->busaddr) {
2671                         printk(KERN_WARNING "cciss%d: Unexpected command "
2672                                 "completion.\n", h->ctlr);
2673                         continue;
2674                 }
2675
2676                 /* It is our command.  If no error, we're done. */
2677                 if (!(complete & CISS_ERROR_BIT)) {
2678                         status = IO_OK;
2679                         break;
2680                 }
2681
2682                 /* There is an error... */
2683
2684                 /* if data overrun or underun on Report command ignore it */
2685                 if (((c->Request.CDB[0] == CISS_REPORT_LOG) ||
2686                      (c->Request.CDB[0] == CISS_REPORT_PHYS) ||
2687                      (c->Request.CDB[0] == CISS_INQUIRY)) &&
2688                         ((c->err_info->CommandStatus == CMD_DATA_OVERRUN) ||
2689                          (c->err_info->CommandStatus == CMD_DATA_UNDERRUN))) {
2690                         complete = c->busaddr;
2691                         status = IO_OK;
2692                         break;
2693                 }
2694
2695                 if (c->err_info->CommandStatus == CMD_UNSOLICITED_ABORT) {
2696                         printk(KERN_WARNING "cciss%d: unsolicited abort %p\n",
2697                                 h->ctlr, c);
2698                         if (c->retry_count < MAX_CMD_RETRIES) {
2699                                 printk(KERN_WARNING "cciss%d: retrying %p\n",
2700                                    h->ctlr, c);
2701                                 c->retry_count++;
2702                                 /* erase the old error information */
2703                                 memset(c->err_info, 0, sizeof(c->err_info));
2704                                 goto resend_cmd1;
2705                         }
2706                         printk(KERN_WARNING "cciss%d: retried %p too many "
2707                                 "times\n", h->ctlr, c);
2708                         status = IO_ERROR;
2709                         break;
2710                 }
2711
2712                 if (c->err_info->CommandStatus == CMD_UNABORTABLE) {
2713                         printk(KERN_WARNING "cciss%d: command could not be "
2714                                 "aborted.\n", h->ctlr);
2715                         status = IO_ERROR;
2716                         break;
2717                 }
2718
2719                 if (c->err_info->CommandStatus == CMD_TARGET_STATUS) {
2720                         status = check_target_status(h, c);
2721                         break;
2722                 }
2723
2724                 printk(KERN_WARNING "cciss%d: sendcmd error\n", h->ctlr);
2725                 printk(KERN_WARNING "cmd = 0x%02x, CommandStatus = 0x%02x\n",
2726                         c->Request.CDB[0], c->err_info->CommandStatus);
2727                 status = IO_ERROR;
2728                 break;
2729
2730         } while (1);
2731
2732         /* unlock the data buffer from DMA */
2733         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2734         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2735         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2736                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2737         return status;
2738 }
2739
2740 /*
2741  * Send a command to the controller, and wait for it to complete.
2742  * Used at init time, and during SCSI error recovery.
2743  */
2744 static int sendcmd(__u8 cmd, int ctlr, void *buff, size_t size,
2745         __u8 page_code, unsigned char *scsi3addr, int cmd_type)
2746 {
2747         CommandList_struct *c;
2748         int status;
2749
2750         c = cmd_alloc(hba[ctlr], 1);
2751         if (!c) {
2752                 printk(KERN_WARNING "cciss: unable to get memory");
2753                 return IO_ERROR;
2754         }
2755         status = fill_cmd(c, cmd, ctlr, buff, size, page_code,
2756                 scsi3addr, cmd_type);
2757         if (status == IO_OK)
2758                 status = sendcmd_core(hba[ctlr], c);
2759         cmd_free(hba[ctlr], c, 1);
2760         return status;
2761 }
2762
2763 /*
2764  * Map (physical) PCI mem into (virtual) kernel space
2765  */
2766 static void __iomem *remap_pci_mem(ulong base, ulong size)
2767 {
2768         ulong page_base = ((ulong) base) & PAGE_MASK;
2769         ulong page_offs = ((ulong) base) - page_base;
2770         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2771
2772         return page_remapped ? (page_remapped + page_offs) : NULL;
2773 }
2774
2775 /*
2776  * Takes jobs of the Q and sends them to the hardware, then puts it on
2777  * the Q to wait for completion.
2778  */
2779 static void start_io(ctlr_info_t *h)
2780 {
2781         CommandList_struct *c;
2782
2783         while (!hlist_empty(&h->reqQ)) {
2784                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
2785                 /* can't do anything if fifo is full */
2786                 if ((h->access.fifo_full(h))) {
2787                         printk(KERN_WARNING "cciss: fifo full\n");
2788                         break;
2789                 }
2790
2791                 /* Get the first entry from the Request Q */
2792                 removeQ(c);
2793                 h->Qdepth--;
2794
2795                 /* Tell the controller execute command */
2796                 h->access.submit_command(h, c);
2797
2798                 /* Put job onto the completed Q */
2799                 addQ(&h->cmpQ, c);
2800         }
2801 }
2802
2803 /* Assumes that CCISS_LOCK(h->ctlr) is held. */
2804 /* Zeros out the error record and then resends the command back */
2805 /* to the controller */
2806 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2807 {
2808         /* erase the old error information */
2809         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2810
2811         /* add it to software queue and then send it to the controller */
2812         addQ(&h->reqQ, c);
2813         h->Qdepth++;
2814         if (h->Qdepth > h->maxQsinceinit)
2815                 h->maxQsinceinit = h->Qdepth;
2816
2817         start_io(h);
2818 }
2819
2820 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
2821         unsigned int msg_byte, unsigned int host_byte,
2822         unsigned int driver_byte)
2823 {
2824         /* inverse of macros in scsi.h */
2825         return (scsi_status_byte & 0xff) |
2826                 ((msg_byte & 0xff) << 8) |
2827                 ((host_byte & 0xff) << 16) |
2828                 ((driver_byte & 0xff) << 24);
2829 }
2830
2831 static inline int evaluate_target_status(ctlr_info_t *h,
2832                         CommandList_struct *cmd, int *retry_cmd)
2833 {
2834         unsigned char sense_key;
2835         unsigned char status_byte, msg_byte, host_byte, driver_byte;
2836         int error_value;
2837
2838         *retry_cmd = 0;
2839         /* If we get in here, it means we got "target status", that is, scsi status */
2840         status_byte = cmd->err_info->ScsiStatus;
2841         driver_byte = DRIVER_OK;
2842         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
2843
2844         if (blk_pc_request(cmd->rq))
2845                 host_byte = DID_PASSTHROUGH;
2846         else
2847                 host_byte = DID_OK;
2848
2849         error_value = make_status_bytes(status_byte, msg_byte,
2850                 host_byte, driver_byte);
2851
2852         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
2853                 if (!blk_pc_request(cmd->rq))
2854                         printk(KERN_WARNING "cciss: cmd %p "
2855                                "has SCSI Status 0x%x\n",
2856                                cmd, cmd->err_info->ScsiStatus);
2857                 return error_value;
2858         }
2859
2860         /* check the sense key */
2861         sense_key = 0xf & cmd->err_info->SenseInfo[2];
2862         /* no status or recovered error */
2863         if (((sense_key == 0x0) || (sense_key == 0x1)) && !blk_pc_request(cmd->rq))
2864                 error_value = 0;
2865
2866         if (check_for_unit_attention(h, cmd)) {
2867                 *retry_cmd = !blk_pc_request(cmd->rq);
2868                 return 0;
2869         }
2870
2871         if (!blk_pc_request(cmd->rq)) { /* Not SG_IO or similar? */
2872                 if (error_value != 0)
2873                         printk(KERN_WARNING "cciss: cmd %p has CHECK CONDITION"
2874                                " sense key = 0x%x\n", cmd, sense_key);
2875                 return error_value;
2876         }
2877
2878         /* SG_IO or similar, copy sense data back */
2879         if (cmd->rq->sense) {
2880                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
2881                         cmd->rq->sense_len = cmd->err_info->SenseLen;
2882                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
2883                         cmd->rq->sense_len);
2884         } else
2885                 cmd->rq->sense_len = 0;
2886
2887         return error_value;
2888 }
2889
2890 /* checks the status of the job and calls complete buffers to mark all
2891  * buffers for the completed job. Note that this function does not need
2892  * to hold the hba/queue lock.
2893  */
2894 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
2895                                     int timeout)
2896 {
2897         int retry_cmd = 0;
2898         struct request *rq = cmd->rq;
2899
2900         rq->errors = 0;
2901
2902         if (timeout)
2903                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
2904
2905         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
2906                 goto after_error_processing;
2907
2908         switch (cmd->err_info->CommandStatus) {
2909         case CMD_TARGET_STATUS:
2910                 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
2911                 break;
2912         case CMD_DATA_UNDERRUN:
2913                 if (blk_fs_request(cmd->rq)) {
2914                         printk(KERN_WARNING "cciss: cmd %p has"
2915                                " completed with data underrun "
2916                                "reported\n", cmd);
2917                         cmd->rq->resid_len = cmd->err_info->ResidualCnt;
2918                 }
2919                 break;
2920         case CMD_DATA_OVERRUN:
2921                 if (blk_fs_request(cmd->rq))
2922                         printk(KERN_WARNING "cciss: cmd %p has"
2923                                " completed with data overrun "
2924                                "reported\n", cmd);
2925                 break;
2926         case CMD_INVALID:
2927                 printk(KERN_WARNING "cciss: cmd %p is "
2928                        "reported invalid\n", cmd);
2929                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2930                         cmd->err_info->CommandStatus, DRIVER_OK,
2931                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2932                 break;
2933         case CMD_PROTOCOL_ERR:
2934                 printk(KERN_WARNING "cciss: cmd %p has "
2935                        "protocol error \n", cmd);
2936                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2937                         cmd->err_info->CommandStatus, DRIVER_OK,
2938                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2939                 break;
2940         case CMD_HARDWARE_ERR:
2941                 printk(KERN_WARNING "cciss: cmd %p had "
2942                        " hardware error\n", cmd);
2943                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2944                         cmd->err_info->CommandStatus, DRIVER_OK,
2945                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2946                 break;
2947         case CMD_CONNECTION_LOST:
2948                 printk(KERN_WARNING "cciss: cmd %p had "
2949                        "connection lost\n", cmd);
2950                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2951                         cmd->err_info->CommandStatus, DRIVER_OK,
2952                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2953                 break;
2954         case CMD_ABORTED:
2955                 printk(KERN_WARNING "cciss: cmd %p was "
2956                        "aborted\n", cmd);
2957                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2958                         cmd->err_info->CommandStatus, DRIVER_OK,
2959                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
2960                 break;
2961         case CMD_ABORT_FAILED:
2962                 printk(KERN_WARNING "cciss: cmd %p reports "
2963                        "abort failed\n", cmd);
2964                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2965                         cmd->err_info->CommandStatus, DRIVER_OK,
2966                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2967                 break;
2968         case CMD_UNSOLICITED_ABORT:
2969                 printk(KERN_WARNING "cciss%d: unsolicited "
2970                        "abort %p\n", h->ctlr, cmd);
2971                 if (cmd->retry_count < MAX_CMD_RETRIES) {
2972                         retry_cmd = 1;
2973                         printk(KERN_WARNING
2974                                "cciss%d: retrying %p\n", h->ctlr, cmd);
2975                         cmd->retry_count++;
2976                 } else
2977                         printk(KERN_WARNING
2978                                "cciss%d: %p retried too "
2979                                "many times\n", h->ctlr, cmd);
2980                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2981                         cmd->err_info->CommandStatus, DRIVER_OK,
2982                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
2983                 break;
2984         case CMD_TIMEOUT:
2985                 printk(KERN_WARNING "cciss: cmd %p timedout\n", cmd);
2986                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2987                         cmd->err_info->CommandStatus, DRIVER_OK,
2988                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2989                 break;
2990         default:
2991                 printk(KERN_WARNING "cciss: cmd %p returned "
2992                        "unknown status %x\n", cmd,
2993                        cmd->err_info->CommandStatus);
2994                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2995                         cmd->err_info->CommandStatus, DRIVER_OK,
2996                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2997         }
2998
2999 after_error_processing:
3000
3001         /* We need to return this command */
3002         if (retry_cmd) {
3003                 resend_cciss_cmd(h, cmd);
3004                 return;
3005         }
3006         cmd->rq->completion_data = cmd;
3007         blk_complete_request(cmd->rq);
3008 }
3009
3010 /*
3011  * Get a request and submit it to the controller.
3012  */
3013 static void do_cciss_request(struct request_queue *q)
3014 {
3015         ctlr_info_t *h = q->queuedata;
3016         CommandList_struct *c;
3017         sector_t start_blk;
3018         int seg;
3019         struct request *creq;
3020         u64bit temp64;
3021         struct scatterlist tmp_sg[MAXSGENTRIES];
3022         drive_info_struct *drv;
3023         int i, dir;
3024
3025         /* We call start_io here in case there is a command waiting on the
3026          * queue that has not been sent.
3027          */
3028         if (blk_queue_plugged(q))
3029                 goto startio;
3030
3031       queue:
3032         creq = blk_peek_request(q);
3033         if (!creq)
3034                 goto startio;
3035
3036         BUG_ON(creq->nr_phys_segments > MAXSGENTRIES);
3037
3038         if ((c = cmd_alloc(h, 1)) == NULL)
3039                 goto full;
3040
3041         blk_start_request(creq);
3042
3043         spin_unlock_irq(q->queue_lock);
3044
3045         c->cmd_type = CMD_RWREQ;
3046         c->rq = creq;
3047
3048         /* fill in the request */
3049         drv = creq->rq_disk->private_data;
3050         c->Header.ReplyQueue = 0;       // unused in simple mode
3051         /* got command from pool, so use the command block index instead */
3052         /* for direct lookups. */
3053         /* The first 2 bits are reserved for controller error reporting. */
3054         c->Header.Tag.lower = (c->cmdindex << 3);
3055         c->Header.Tag.lower |= 0x04;    /* flag for direct lookup. */
3056         c->Header.LUN.LogDev.VolId = drv->LunID;
3057         c->Header.LUN.LogDev.Mode = 1;
3058         c->Request.CDBLen = 10; // 12 byte commands not in FW yet;
3059         c->Request.Type.Type = TYPE_CMD;        // It is a command.
3060         c->Request.Type.Attribute = ATTR_SIMPLE;
3061         c->Request.Type.Direction =
3062             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3063         c->Request.Timeout = 0; // Don't time out
3064         c->Request.CDB[0] =
3065             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3066         start_blk = blk_rq_pos(creq);
3067 #ifdef CCISS_DEBUG
3068         printk(KERN_DEBUG "ciss: sector =%d nr_sectors=%d\n",
3069                (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3070 #endif                          /* CCISS_DEBUG */
3071
3072         sg_init_table(tmp_sg, MAXSGENTRIES);
3073         seg = blk_rq_map_sg(q, creq, tmp_sg);
3074
3075         /* get the DMA records for the setup */
3076         if (c->Request.Type.Direction == XFER_READ)
3077                 dir = PCI_DMA_FROMDEVICE;
3078         else
3079                 dir = PCI_DMA_TODEVICE;
3080
3081         for (i = 0; i < seg; i++) {
3082                 c->SG[i].Len = tmp_sg[i].length;
3083                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3084                                                   tmp_sg[i].offset,
3085                                                   tmp_sg[i].length, dir);
3086                 c->SG[i].Addr.lower = temp64.val32.lower;
3087                 c->SG[i].Addr.upper = temp64.val32.upper;
3088                 c->SG[i].Ext = 0;       // we are not chaining
3089         }
3090         /* track how many SG entries we are using */
3091         if (seg > h->maxSG)
3092                 h->maxSG = seg;
3093
3094 #ifdef CCISS_DEBUG
3095         printk(KERN_DEBUG "cciss: Submitting %u sectors in %d segments\n",
3096                blk_rq_sectors(creq), seg);
3097 #endif                          /* CCISS_DEBUG */
3098
3099         c->Header.SGList = c->Header.SGTotal = seg;
3100         if (likely(blk_fs_request(creq))) {
3101                 if(h->cciss_read == CCISS_READ_10) {
3102                         c->Request.CDB[1] = 0;
3103                         c->Request.CDB[2] = (start_blk >> 24) & 0xff;   //MSB
3104                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3105                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3106                         c->Request.CDB[5] = start_blk & 0xff;
3107                         c->Request.CDB[6] = 0;  // (sect >> 24) & 0xff; MSB
3108                         c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3109                         c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3110                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3111                 } else {
3112                         u32 upper32 = upper_32_bits(start_blk);
3113
3114                         c->Request.CDBLen = 16;
3115                         c->Request.CDB[1]= 0;
3116                         c->Request.CDB[2]= (upper32 >> 24) & 0xff;      //MSB
3117                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3118                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3119                         c->Request.CDB[5]= upper32 & 0xff;
3120                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3121                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3122                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3123                         c->Request.CDB[9]= start_blk & 0xff;
3124                         c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3125                         c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3126                         c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3127                         c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3128                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
3129                 }
3130         } else if (blk_pc_request(creq)) {
3131                 c->Request.CDBLen = creq->cmd_len;
3132                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3133         } else {
3134                 printk(KERN_WARNING "cciss%d: bad request type %d\n", h->ctlr, creq->cmd_type);
3135                 BUG();
3136         }
3137
3138         spin_lock_irq(q->queue_lock);
3139
3140         addQ(&h->reqQ, c);
3141         h->Qdepth++;
3142         if (h->Qdepth > h->maxQsinceinit)
3143                 h->maxQsinceinit = h->Qdepth;
3144
3145         goto queue;
3146 full:
3147         blk_stop_queue(q);
3148 startio:
3149         /* We will already have the driver lock here so not need
3150          * to lock it.
3151          */
3152         start_io(h);
3153 }
3154
3155 static inline unsigned long get_next_completion(ctlr_info_t *h)
3156 {
3157         return h->access.command_completed(h);
3158 }
3159
3160 static inline int interrupt_pending(ctlr_info_t *h)
3161 {
3162         return h->access.intr_pending(h);
3163 }
3164
3165 static inline long interrupt_not_for_us(ctlr_info_t *h)
3166 {
3167         return (((h->access.intr_pending(h) == 0) ||
3168                  (h->interrupts_enabled == 0)));
3169 }
3170
3171 static irqreturn_t do_cciss_intr(int irq, void *dev_id)
3172 {
3173         ctlr_info_t *h = dev_id;
3174         CommandList_struct *c;
3175         unsigned long flags;
3176         __u32 a, a1, a2;
3177
3178         if (interrupt_not_for_us(h))
3179                 return IRQ_NONE;
3180         /*
3181          * If there are completed commands in the completion queue,
3182          * we had better do something about it.
3183          */
3184         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
3185         while (interrupt_pending(h)) {
3186                 while ((a = get_next_completion(h)) != FIFO_EMPTY) {
3187                         a1 = a;
3188                         if ((a & 0x04)) {
3189                                 a2 = (a >> 3);
3190                                 if (a2 >= h->nr_cmds) {
3191                                         printk(KERN_WARNING
3192                                                "cciss: controller cciss%d failed, stopping.\n",
3193                                                h->ctlr);
3194                                         fail_all_cmds(h->ctlr);
3195                                         return IRQ_HANDLED;
3196                                 }
3197
3198                                 c = h->cmd_pool + a2;
3199                                 a = c->busaddr;
3200
3201                         } else {
3202                                 struct hlist_node *tmp;
3203
3204                                 a &= ~3;
3205                                 c = NULL;
3206                                 hlist_for_each_entry(c, tmp, &h->cmpQ, list) {
3207                                         if (c->busaddr == a)
3208                                                 break;
3209                                 }
3210                         }
3211                         /*
3212                          * If we've found the command, take it off the
3213                          * completion Q and free it
3214                          */
3215                         if (c && c->busaddr == a) {
3216                                 removeQ(c);
3217                                 if (c->cmd_type == CMD_RWREQ) {
3218                                         complete_command(h, c, 0);
3219                                 } else if (c->cmd_type == CMD_IOCTL_PEND) {
3220                                         complete(c->waiting);
3221                                 }
3222 #                               ifdef CONFIG_CISS_SCSI_TAPE
3223                                 else if (c->cmd_type == CMD_SCSI)
3224                                         complete_scsi_command(c, 0, a1);
3225 #                               endif
3226                                 continue;
3227                         }
3228                 }
3229         }
3230
3231         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
3232         return IRQ_HANDLED;
3233 }
3234
3235 static int scan_thread(void *data)
3236 {
3237         ctlr_info_t *h = data;
3238         int rc;
3239         DECLARE_COMPLETION_ONSTACK(wait);
3240         h->rescan_wait = &wait;
3241
3242         for (;;) {
3243                 rc = wait_for_completion_interruptible(&wait);
3244                 if (kthread_should_stop())
3245                         break;
3246                 if (!rc)
3247                         rebuild_lun_table(h, 0);
3248         }
3249         return 0;
3250 }
3251
3252 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3253 {
3254         if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3255                 return 0;
3256
3257         switch (c->err_info->SenseInfo[12]) {
3258         case STATE_CHANGED:
3259                 printk(KERN_WARNING "cciss%d: a state change "
3260                         "detected, command retried\n", h->ctlr);
3261                 return 1;
3262         break;
3263         case LUN_FAILED:
3264                 printk(KERN_WARNING "cciss%d: LUN failure "
3265                         "detected, action required\n", h->ctlr);
3266                 return 1;
3267         break;
3268         case REPORT_LUNS_CHANGED:
3269                 printk(KERN_WARNING "cciss%d: report LUN data "
3270                         "changed\n", h->ctlr);
3271                 if (h->rescan_wait)
3272                         complete(h->rescan_wait);
3273                 return 1;
3274         break;
3275         case POWER_OR_RESET:
3276                 printk(KERN_WARNING "cciss%d: a power on "
3277                         "or device reset detected\n", h->ctlr);
3278                 return 1;
3279         break;
3280         case UNIT_ATTENTION_CLEARED:
3281                 printk(KERN_WARNING "cciss%d: unit attention "
3282                     "cleared by another initiator\n", h->ctlr);
3283                 return 1;
3284         break;
3285         default:
3286                 printk(KERN_WARNING "cciss%d: unknown "
3287                         "unit attention detected\n", h->ctlr);
3288                                 return 1;
3289         }
3290 }
3291
3292 /*
3293  *  We cannot read the structure directly, for portability we must use
3294  *   the io functions.
3295  *   This is for debug only.
3296  */
3297 #ifdef CCISS_DEBUG
3298 static void print_cfg_table(CfgTable_struct *tb)
3299 {
3300         int i;
3301         char temp_name[17];
3302
3303         printk("Controller Configuration information\n");
3304         printk("------------------------------------\n");
3305         for (i = 0; i < 4; i++)
3306                 temp_name[i] = readb(&(tb->Signature[i]));
3307         temp_name[4] = '\0';
3308         printk("   Signature = %s\n", temp_name);
3309         printk("   Spec Number = %d\n", readl(&(tb->SpecValence)));
3310         printk("   Transport methods supported = 0x%x\n",
3311                readl(&(tb->TransportSupport)));
3312         printk("   Transport methods active = 0x%x\n",
3313                readl(&(tb->TransportActive)));
3314         printk("   Requested transport Method = 0x%x\n",
3315                readl(&(tb->HostWrite.TransportRequest)));
3316         printk("   Coalesce Interrupt Delay = 0x%x\n",
3317                readl(&(tb->HostWrite.CoalIntDelay)));
3318         printk("   Coalesce Interrupt Count = 0x%x\n",
3319                readl(&(tb->HostWrite.CoalIntCount)));
3320         printk("   Max outstanding commands = 0x%d\n",
3321                readl(&(tb->CmdsOutMax)));
3322         printk("   Bus Types = 0x%x\n", readl(&(tb->BusTypes)));
3323         for (i = 0; i < 16; i++)
3324                 temp_name[i] = readb(&(tb->ServerName[i]));
3325         temp_name[16] = '\0';
3326         printk("   Server Name = %s\n", temp_name);
3327         printk("   Heartbeat Counter = 0x%x\n\n\n", readl(&(tb->HeartBeat)));
3328 }
3329 #endif                          /* CCISS_DEBUG */
3330
3331 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3332 {
3333         int i, offset, mem_type, bar_type;
3334         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3335                 return 0;
3336         offset = 0;
3337         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3338                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3339                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3340                         offset += 4;
3341                 else {
3342                         mem_type = pci_resource_flags(pdev, i) &
3343                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3344                         switch (mem_type) {
3345                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
3346                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3347                                 offset += 4;    /* 32 bit */
3348                                 break;
3349                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
3350                                 offset += 8;
3351                                 break;
3352                         default:        /* reserved in PCI 2.2 */
3353                                 printk(KERN_WARNING
3354                                        "Base address is invalid\n");
3355                                 return -1;
3356                                 break;
3357                         }
3358                 }
3359                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3360                         return i + 1;
3361         }
3362         return -1;
3363 }
3364
3365 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
3366  * controllers that are capable. If not, we use IO-APIC mode.
3367  */
3368
3369 static void __devinit cciss_interrupt_mode(ctlr_info_t *c,
3370                                            struct pci_dev *pdev, __u32 board_id)
3371 {
3372 #ifdef CONFIG_PCI_MSI
3373         int err;
3374         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
3375         {0, 2}, {0, 3}
3376         };
3377
3378         /* Some boards advertise MSI but don't really support it */
3379         if ((board_id == 0x40700E11) ||
3380             (board_id == 0x40800E11) ||
3381             (board_id == 0x40820E11) || (board_id == 0x40830E11))
3382                 goto default_int_mode;
3383
3384         if (pci_find_capability(pdev, PCI_CAP_ID_MSIX)) {
3385                 err = pci_enable_msix(pdev, cciss_msix_entries, 4);
3386                 if (!err) {
3387                         c->intr[0] = cciss_msix_entries[0].vector;
3388                         c->intr[1] = cciss_msix_entries[1].vector;
3389                         c->intr[2] = cciss_msix_entries[2].vector;
3390                         c->intr[3] = cciss_msix_entries[3].vector;
3391                         c->msix_vector = 1;
3392                         return;
3393                 }
3394                 if (err > 0) {
3395                         printk(KERN_WARNING "cciss: only %d MSI-X vectors "
3396                                "available\n", err);
3397                         goto default_int_mode;
3398                 } else {
3399                         printk(KERN_WARNING "cciss: MSI-X init failed %d\n",
3400                                err);
3401                         goto default_int_mode;
3402                 }
3403         }
3404         if (pci_find_capability(pdev, PCI_CAP_ID_MSI)) {
3405                 if (!pci_enable_msi(pdev)) {
3406                         c->msi_vector = 1;
3407                 } else {
3408                         printk(KERN_WARNING "cciss: MSI init failed\n");
3409                 }
3410         }
3411 default_int_mode:
3412 #endif                          /* CONFIG_PCI_MSI */
3413         /* if we get here we're going to use the default interrupt mode */
3414         c->intr[SIMPLE_MODE_INT] = pdev->irq;
3415         return;
3416 }
3417
3418 static int __devinit cciss_pci_init(ctlr_info_t *c, struct pci_dev *pdev)
3419 {
3420         ushort subsystem_vendor_id, subsystem_device_id, command;
3421         __u32 board_id, scratchpad = 0;
3422         __u64 cfg_offset;
3423         __u32 cfg_base_addr;
3424         __u64 cfg_base_addr_index;
3425         int i, err;
3426
3427         /* check to see if controller has been disabled */
3428         /* BEFORE trying to enable it */
3429         (void)pci_read_config_word(pdev, PCI_COMMAND, &command);
3430         if (!(command & 0x02)) {
3431                 printk(KERN_WARNING
3432                        "cciss: controller appears to be disabled\n");
3433                 return -ENODEV;
3434         }
3435
3436         err = pci_enable_device(pdev);
3437         if (err) {
3438                 printk(KERN_ERR "cciss: Unable to Enable PCI device\n");
3439                 return err;
3440         }
3441
3442         err = pci_request_regions(pdev, "cciss");
3443         if (err) {
3444                 printk(KERN_ERR "cciss: Cannot obtain PCI resources, "
3445                        "aborting\n");
3446                 return err;
3447         }
3448
3449         subsystem_vendor_id = pdev->subsystem_vendor;
3450         subsystem_device_id = pdev->subsystem_device;
3451         board_id = (((__u32) (subsystem_device_id << 16) & 0xffff0000) |
3452                     subsystem_vendor_id);
3453
3454 #ifdef CCISS_DEBUG
3455         printk("command = %x\n", command);
3456         printk("irq = %x\n", pdev->irq);
3457         printk("board_id = %x\n", board_id);
3458 #endif                          /* CCISS_DEBUG */
3459
3460 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
3461  * else we use the IO-APIC interrupt assigned to us by system ROM.
3462  */
3463         cciss_interrupt_mode(c, pdev, board_id);
3464
3465         /* find the memory BAR */
3466         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3467                 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM)
3468                         break;
3469         }
3470         if (i == DEVICE_COUNT_RESOURCE) {
3471                 printk(KERN_WARNING "cciss: No memory BAR found\n");
3472                 err = -ENODEV;
3473                 goto err_out_free_res;
3474         }
3475
3476         c->paddr = pci_resource_start(pdev, i); /* addressing mode bits
3477                                                  * already removed
3478                                                  */
3479
3480 #ifdef CCISS_DEBUG
3481         printk("address 0 = %lx\n", c->paddr);
3482 #endif                          /* CCISS_DEBUG */
3483         c->vaddr = remap_pci_mem(c->paddr, 0x250);
3484
3485         /* Wait for the board to become ready.  (PCI hotplug needs this.)
3486          * We poll for up to 120 secs, once per 100ms. */
3487         for (i = 0; i < 1200; i++) {
3488                 scratchpad = readl(c->vaddr + SA5_SCRATCHPAD_OFFSET);
3489                 if (scratchpad == CCISS_FIRMWARE_READY)
3490                         break;
3491                 set_current_state(TASK_INTERRUPTIBLE);
3492                 schedule_timeout(HZ / 10);      /* wait 100ms */
3493         }
3494         if (scratchpad != CCISS_FIRMWARE_READY) {
3495                 printk(KERN_WARNING "cciss: Board not ready.  Timed out.\n");
3496                 err = -ENODEV;
3497                 goto err_out_free_res;
3498         }
3499
3500         /* get the address index number */
3501         cfg_base_addr = readl(c->vaddr + SA5_CTCFG_OFFSET);
3502         cfg_base_addr &= (__u32) 0x0000ffff;
3503 #ifdef CCISS_DEBUG
3504         printk("cfg base address = %x\n", cfg_base_addr);
3505 #endif                          /* CCISS_DEBUG */
3506         cfg_base_addr_index = find_PCI_BAR_index(pdev, cfg_base_addr);
3507 #ifdef CCISS_DEBUG
3508         printk("cfg base address index = %llx\n",
3509                 (unsigned long long)cfg_base_addr_index);
3510 #endif                          /* CCISS_DEBUG */
3511         if (cfg_base_addr_index == -1) {
3512                 printk(KERN_WARNING "cciss: Cannot find cfg_base_addr_index\n");
3513                 err = -ENODEV;
3514                 goto err_out_free_res;
3515         }
3516
3517         cfg_offset = readl(c->vaddr + SA5_CTMEM_OFFSET);
3518 #ifdef CCISS_DEBUG
3519         printk("cfg offset = %llx\n", (unsigned long long)cfg_offset);
3520 #endif                          /* CCISS_DEBUG */
3521         c->cfgtable = remap_pci_mem(pci_resource_start(pdev,
3522                                                        cfg_base_addr_index) +
3523                                     cfg_offset, sizeof(CfgTable_struct));
3524         c->board_id = board_id;
3525
3526 #ifdef CCISS_DEBUG
3527         print_cfg_table(c->cfgtable);
3528 #endif                          /* CCISS_DEBUG */
3529
3530         /* Some controllers support Zero Memory Raid (ZMR).
3531          * When configured in ZMR mode the number of supported
3532          * commands drops to 64. So instead of just setting an
3533          * arbitrary value we make the driver a little smarter.
3534          * We read the config table to tell us how many commands
3535          * are supported on the controller then subtract 4 to
3536          * leave a little room for ioctl calls.
3537          */
3538         c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
3539         for (i = 0; i < ARRAY_SIZE(products); i++) {
3540                 if (board_id == products[i].board_id) {
3541                         c->product_name = products[i].product_name;
3542                         c->access = *(products[i].access);
3543                         c->nr_cmds = c->max_commands - 4;
3544                         break;
3545                 }
3546         }
3547         if ((readb(&c->cfgtable->Signature[0]) != 'C') ||
3548             (readb(&c->cfgtable->Signature[1]) != 'I') ||
3549             (readb(&c->cfgtable->Signature[2]) != 'S') ||
3550             (readb(&c->cfgtable->Signature[3]) != 'S')) {
3551                 printk("Does not appear to be a valid CISS config table\n");
3552                 err = -ENODEV;
3553                 goto err_out_free_res;
3554         }
3555         /* We didn't find the controller in our list. We know the
3556          * signature is valid. If it's an HP device let's try to
3557          * bind to the device and fire it up. Otherwise we bail.
3558          */
3559         if (i == ARRAY_SIZE(products)) {
3560                 if (subsystem_vendor_id == PCI_VENDOR_ID_HP) {
3561                         c->product_name = products[i-1].product_name;
3562                         c->access = *(products[i-1].access);
3563                         c->nr_cmds = c->max_commands - 4;
3564                         printk(KERN_WARNING "cciss: This is an unknown "
3565                                 "Smart Array controller.\n"
3566                                 "cciss: Please update to the latest driver "
3567                                 "available from www.hp.com.\n");
3568                 } else {
3569                         printk(KERN_WARNING "cciss: Sorry, I don't know how"
3570                                 " to access the Smart Array controller %08lx\n"
3571                                         , (unsigned long)board_id);
3572                         err = -ENODEV;
3573                         goto err_out_free_res;
3574                 }
3575         }
3576 #ifdef CONFIG_X86
3577         {
3578                 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
3579                 __u32 prefetch;
3580                 prefetch = readl(&(c->cfgtable->SCSI_Prefetch));
3581                 prefetch |= 0x100;
3582                 writel(prefetch, &(c->cfgtable->SCSI_Prefetch));
3583         }
3584 #endif
3585
3586         /* Disabling DMA prefetch and refetch for the P600.
3587          * An ASIC bug may result in accesses to invalid memory addresses.
3588          * We've disabled prefetch for some time now. Testing with XEN
3589          * kernels revealed a bug in the refetch if dom0 resides on a P600.
3590          */
3591         if(board_id == 0x3225103C) {
3592                 __u32 dma_prefetch;
3593                 __u32 dma_refetch;
3594                 dma_prefetch = readl(c->vaddr + I2O_DMA1_CFG);
3595                 dma_prefetch |= 0x8000;
3596                 writel(dma_prefetch, c->vaddr + I2O_DMA1_CFG);
3597                 pci_read_config_dword(pdev, PCI_COMMAND_PARITY, &dma_refetch);
3598                 dma_refetch |= 0x1;
3599                 pci_write_config_dword(pdev, PCI_COMMAND_PARITY, dma_refetch);
3600         }
3601
3602 #ifdef CCISS_DEBUG
3603         printk("Trying to put board into Simple mode\n");
3604 #endif                          /* CCISS_DEBUG */
3605         c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
3606         /* Update the field, and then ring the doorbell */
3607         writel(CFGTBL_Trans_Simple, &(c->cfgtable->HostWrite.TransportRequest));
3608         writel(CFGTBL_ChangeReq, c->vaddr + SA5_DOORBELL);
3609
3610         /* under certain very rare conditions, this can take awhile.
3611          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3612          * as we enter this code.) */
3613         for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3614                 if (!(readl(c->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3615                         break;
3616                 /* delay and try again */
3617                 set_current_state(TASK_INTERRUPTIBLE);
3618                 schedule_timeout(10);
3619         }
3620
3621 #ifdef CCISS_DEBUG
3622         printk(KERN_DEBUG "I counter got to %d %x\n", i,
3623                readl(c->vaddr + SA5_DOORBELL));
3624 #endif                          /* CCISS_DEBUG */
3625 #ifdef CCISS_DEBUG
3626         print_cfg_table(c->cfgtable);
3627 #endif                          /* CCISS_DEBUG */
3628
3629         if (!(readl(&(c->cfgtable->TransportActive)) & CFGTBL_Trans_Simple)) {
3630                 printk(KERN_WARNING "cciss: unable to get board into"
3631                        " simple mode\n");
3632                 err = -ENODEV;
3633                 goto err_out_free_res;
3634         }
3635         return 0;
3636
3637 err_out_free_res:
3638         /*
3639          * Deliberately omit pci_disable_device(): it does something nasty to
3640          * Smart Array controllers that pci_enable_device does not undo
3641          */
3642         pci_release_regions(pdev);
3643         return err;
3644 }
3645
3646 /* Function to find the first free pointer into our hba[] array
3647  * Returns -1 if no free entries are left.
3648  */
3649 static int alloc_cciss_hba(void)
3650 {
3651         int i;
3652
3653         for (i = 0; i < MAX_CTLR; i++) {
3654                 if (!hba[i]) {
3655                         ctlr_info_t *p;
3656
3657                         p = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
3658                         if (!p)
3659                                 goto Enomem;
3660                         hba[i] = p;
3661                         return i;
3662                 }
3663         }
3664         printk(KERN_WARNING "cciss: This driver supports a maximum"
3665                " of %d controllers.\n", MAX_CTLR);
3666         return -1;
3667 Enomem:
3668         printk(KERN_ERR "cciss: out of memory.\n");
3669         return -1;
3670 }
3671
3672 static void free_hba(int i)
3673 {
3674         ctlr_info_t *p = hba[i];
3675         int n;
3676
3677         hba[i] = NULL;
3678         for (n = 0; n < CISS_MAX_LUN; n++)
3679                 put_disk(p->gendisk[n]);
3680         kfree(p);
3681 }
3682
3683 /* Send a message CDB to the firmware. */
3684 static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
3685 {
3686         typedef struct {
3687                 CommandListHeader_struct CommandHeader;
3688                 RequestBlock_struct Request;
3689                 ErrDescriptor_struct ErrorDescriptor;
3690         } Command;
3691         static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
3692         Command *cmd;
3693         dma_addr_t paddr64;
3694         uint32_t paddr32, tag;
3695         void __iomem *vaddr;
3696         int i, err;
3697
3698         vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
3699         if (vaddr == NULL)
3700                 return -ENOMEM;
3701
3702         /* The Inbound Post Queue only accepts 32-bit physical addresses for the
3703            CCISS commands, so they must be allocated from the lower 4GiB of
3704            memory. */
3705         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
3706         if (err) {
3707                 iounmap(vaddr);
3708                 return -ENOMEM;
3709         }
3710
3711         cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
3712         if (cmd == NULL) {
3713                 iounmap(vaddr);
3714                 return -ENOMEM;
3715         }
3716
3717         /* This must fit, because of the 32-bit consistent DMA mask.  Also,
3718            although there's no guarantee, we assume that the address is at
3719            least 4-byte aligned (most likely, it's page-aligned). */
3720         paddr32 = paddr64;
3721
3722         cmd->CommandHeader.ReplyQueue = 0;
3723         cmd->CommandHeader.SGList = 0;
3724         cmd->CommandHeader.SGTotal = 0;
3725         cmd->CommandHeader.Tag.lower = paddr32;
3726         cmd->CommandHeader.Tag.upper = 0;
3727         memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
3728
3729         cmd->Request.CDBLen = 16;
3730         cmd->Request.Type.Type = TYPE_MSG;
3731         cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
3732         cmd->Request.Type.Direction = XFER_NONE;
3733         cmd->Request.Timeout = 0; /* Don't time out */
3734         cmd->Request.CDB[0] = opcode;
3735         cmd->Request.CDB[1] = type;
3736         memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
3737
3738         cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
3739         cmd->ErrorDescriptor.Addr.upper = 0;
3740         cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
3741
3742         writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
3743
3744         for (i = 0; i < 10; i++) {
3745                 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
3746                 if ((tag & ~3) == paddr32)
3747                         break;
3748                 schedule_timeout_uninterruptible(HZ);
3749         }
3750
3751         iounmap(vaddr);
3752
3753         /* we leak the DMA buffer here ... no choice since the controller could
3754            still complete the command. */
3755         if (i == 10) {
3756                 printk(KERN_ERR "cciss: controller message %02x:%02x timed out\n",
3757                         opcode, type);
3758                 return -ETIMEDOUT;
3759         }
3760
3761         pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
3762
3763         if (tag & 2) {
3764                 printk(KERN_ERR "cciss: controller message %02x:%02x failed\n",
3765                         opcode, type);
3766                 return -EIO;
3767         }
3768
3769         printk(KERN_INFO "cciss: controller message %02x:%02x succeeded\n",
3770                 opcode, type);
3771         return 0;
3772 }
3773
3774 #define cciss_soft_reset_controller(p) cciss_message(p, 1, 0)
3775 #define cciss_noop(p) cciss_message(p, 3, 0)
3776
3777 static __devinit int cciss_reset_msi(struct pci_dev *pdev)
3778 {
3779 /* the #defines are stolen from drivers/pci/msi.h. */
3780 #define msi_control_reg(base)           (base + PCI_MSI_FLAGS)
3781 #define PCI_MSIX_FLAGS_ENABLE           (1 << 15)
3782
3783         int pos;
3784         u16 control = 0;
3785
3786         pos = pci_find_capability(pdev, PCI_CAP_ID_MSI);
3787         if (pos) {
3788                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
3789                 if (control & PCI_MSI_FLAGS_ENABLE) {
3790                         printk(KERN_INFO "cciss: resetting MSI\n");
3791                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSI_FLAGS_ENABLE);
3792                 }
3793         }
3794
3795         pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
3796         if (pos) {
3797                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
3798                 if (control & PCI_MSIX_FLAGS_ENABLE) {
3799                         printk(KERN_INFO "cciss: resetting MSI-X\n");
3800                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSIX_FLAGS_ENABLE);
3801                 }
3802         }
3803
3804         return 0;
3805 }
3806
3807 /* This does a hard reset of the controller using PCI power management
3808  * states. */
3809 static __devinit int cciss_hard_reset_controller(struct pci_dev *pdev)
3810 {
3811         u16 pmcsr, saved_config_space[32];
3812         int i, pos;
3813
3814         printk(KERN_INFO "cciss: using PCI PM to reset controller\n");
3815
3816         /* This is very nearly the same thing as
3817
3818            pci_save_state(pci_dev);
3819            pci_set_power_state(pci_dev, PCI_D3hot);
3820            pci_set_power_state(pci_dev, PCI_D0);
3821            pci_restore_state(pci_dev);
3822
3823            but we can't use these nice canned kernel routines on
3824            kexec, because they also check the MSI/MSI-X state in PCI
3825            configuration space and do the wrong thing when it is
3826            set/cleared.  Also, the pci_save/restore_state functions
3827            violate the ordering requirements for restoring the
3828            configuration space from the CCISS document (see the
3829            comment below).  So we roll our own .... */
3830
3831         for (i = 0; i < 32; i++)
3832                 pci_read_config_word(pdev, 2*i, &saved_config_space[i]);
3833
3834         pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
3835         if (pos == 0) {
3836                 printk(KERN_ERR "cciss_reset_controller: PCI PM not supported\n");
3837                 return -ENODEV;
3838         }
3839
3840         /* Quoting from the Open CISS Specification: "The Power
3841          * Management Control/Status Register (CSR) controls the power
3842          * state of the device.  The normal operating state is D0,
3843          * CSR=00h.  The software off state is D3, CSR=03h.  To reset
3844          * the controller, place the interface device in D3 then to
3845          * D0, this causes a secondary PCI reset which will reset the
3846          * controller." */
3847
3848         /* enter the D3hot power management state */
3849         pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
3850         pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
3851         pmcsr |= PCI_D3hot;
3852         pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
3853
3854         schedule_timeout_uninterruptible(HZ >> 1);
3855
3856         /* enter the D0 power management state */
3857         pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
3858         pmcsr |= PCI_D0;
3859         pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
3860
3861         schedule_timeout_uninterruptible(HZ >> 1);
3862
3863         /* Restore the PCI configuration space.  The Open CISS
3864          * Specification says, "Restore the PCI Configuration
3865          * Registers, offsets 00h through 60h. It is important to
3866          * restore the command register, 16-bits at offset 04h,
3867          * last. Do not restore the configuration status register,
3868          * 16-bits at offset 06h."  Note that the offset is 2*i. */
3869         for (i = 0; i < 32; i++) {
3870                 if (i == 2 || i == 3)
3871                         continue;
3872                 pci_write_config_word(pdev, 2*i, saved_config_space[i]);
3873         }
3874         wmb();
3875         pci_write_config_word(pdev, 4, saved_config_space[2]);
3876
3877         return 0;
3878 }
3879
3880 /*
3881  *  This is it.  Find all the controllers and register them.  I really hate
3882  *  stealing all these major device numbers.
3883  *  returns the number of block devices registered.
3884  */
3885 static int __devinit cciss_init_one(struct pci_dev *pdev,
3886                                     const struct pci_device_id *ent)
3887 {
3888         int i;
3889         int j = 0;
3890         int rc;
3891         int dac, return_code;
3892         InquiryData_struct *inq_buff;
3893
3894         if (reset_devices) {
3895                 /* Reset the controller with a PCI power-cycle */
3896                 if (cciss_hard_reset_controller(pdev) || cciss_reset_msi(pdev))
3897                         return -ENODEV;
3898
3899                 /* Now try to get the controller to respond to a no-op. Some
3900                    devices (notably the HP Smart Array 5i Controller) need
3901                    up to 30 seconds to respond. */
3902                 for (i=0; i<30; i++) {
3903                         if (cciss_noop(pdev) == 0)
3904                                 break;
3905
3906                         schedule_timeout_uninterruptible(HZ);
3907                 }
3908                 if (i == 30) {
3909                         printk(KERN_ERR "cciss: controller seems dead\n");
3910                         return -EBUSY;
3911                 }
3912         }
3913
3914         i = alloc_cciss_hba();
3915         if (i < 0)
3916                 return -1;
3917
3918         hba[i]->busy_initializing = 1;
3919         INIT_HLIST_HEAD(&hba[i]->cmpQ);
3920         INIT_HLIST_HEAD(&hba[i]->reqQ);
3921
3922         if (cciss_pci_init(hba[i], pdev) != 0)
3923                 goto clean0;
3924
3925         sprintf(hba[i]->devname, "cciss%d", i);
3926         hba[i]->ctlr = i;
3927         hba[i]->pdev = pdev;
3928
3929         if (cciss_create_hba_sysfs_entry(hba[i]))
3930                 goto clean0;
3931
3932         /* configure PCI DMA stuff */
3933         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
3934                 dac = 1;
3935         else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
3936                 dac = 0;
3937         else {
3938                 printk(KERN_ERR "cciss: no suitable DMA available\n");
3939                 goto clean1;
3940         }
3941
3942         /*
3943          * register with the major number, or get a dynamic major number
3944          * by passing 0 as argument.  This is done for greater than
3945          * 8 controller support.
3946          */
3947         if (i < MAX_CTLR_ORIG)
3948                 hba[i]->major = COMPAQ_CISS_MAJOR + i;
3949         rc = register_blkdev(hba[i]->major, hba[i]->devname);
3950         if (rc == -EBUSY || rc == -EINVAL) {
3951                 printk(KERN_ERR
3952                        "cciss:  Unable to get major number %d for %s "
3953                        "on hba %d\n", hba[i]->major, hba[i]->devname, i);
3954                 goto clean1;
3955         } else {
3956                 if (i >= MAX_CTLR_ORIG)
3957                         hba[i]->major = rc;
3958         }
3959
3960         /* make sure the board interrupts are off */
3961         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_OFF);
3962         if (request_irq(hba[i]->intr[SIMPLE_MODE_INT], do_cciss_intr,
3963                         IRQF_DISABLED | IRQF_SHARED, hba[i]->devname, hba[i])) {
3964                 printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
3965                        hba[i]->intr[SIMPLE_MODE_INT], hba[i]->devname);
3966                 goto clean2;
3967         }
3968
3969         printk(KERN_INFO "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
3970                hba[i]->devname, pdev->device, pci_name(pdev),
3971                hba[i]->intr[SIMPLE_MODE_INT], dac ? "" : " not");
3972
3973         hba[i]->cmd_pool_bits =
3974             kmalloc(DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
3975                         * sizeof(unsigned long), GFP_KERNEL);
3976         hba[i]->cmd_pool = (CommandList_struct *)
3977             pci_alloc_consistent(hba[i]->pdev,
3978                     hba[i]->nr_cmds * sizeof(CommandList_struct),
3979                     &(hba[i]->cmd_pool_dhandle));
3980         hba[i]->errinfo_pool = (ErrorInfo_struct *)
3981             pci_alloc_consistent(hba[i]->pdev,
3982                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
3983                     &(hba[i]->errinfo_pool_dhandle));
3984         if ((hba[i]->cmd_pool_bits == NULL)
3985             || (hba[i]->cmd_pool == NULL)
3986             || (hba[i]->errinfo_pool == NULL)) {
3987                 printk(KERN_ERR "cciss: out of memory");
3988                 goto clean4;
3989         }
3990         spin_lock_init(&hba[i]->lock);
3991
3992         /* Initialize the pdev driver private data.
3993            have it point to hba[i].  */
3994         pci_set_drvdata(pdev, hba[i]);
3995         /* command and error info recs zeroed out before
3996            they are used */
3997         memset(hba[i]->cmd_pool_bits, 0,
3998                DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
3999                         * sizeof(unsigned long));
4000
4001         hba[i]->num_luns = 0;
4002         hba[i]->highest_lun = -1;
4003         for (j = 0; j < CISS_MAX_LUN; j++) {
4004                 hba[i]->drv[j].raid_level = -1;
4005                 hba[i]->drv[j].queue = NULL;
4006                 hba[i]->gendisk[j] = NULL;
4007         }
4008
4009         cciss_scsi_setup(i);
4010
4011         /* Turn the interrupts on so we can service requests */
4012         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_ON);
4013
4014         /* Get the firmware version */
4015         inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
4016         if (inq_buff == NULL) {
4017                 printk(KERN_ERR "cciss: out of memory\n");
4018                 goto clean4;
4019         }
4020
4021         return_code = sendcmd_withirq(CISS_INQUIRY, i, inq_buff,
4022                 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
4023         if (return_code == IO_OK) {
4024                 hba[i]->firm_ver[0] = inq_buff->data_byte[32];
4025                 hba[i]->firm_ver[1] = inq_buff->data_byte[33];
4026                 hba[i]->firm_ver[2] = inq_buff->data_byte[34];
4027                 hba[i]->firm_ver[3] = inq_buff->data_byte[35];
4028         } else {         /* send command failed */
4029                 printk(KERN_WARNING "cciss: unable to determine firmware"
4030                         " version of controller\n");
4031         }
4032         kfree(inq_buff);
4033
4034         cciss_procinit(i);
4035
4036         hba[i]->cciss_max_sectors = 2048;
4037
4038         hba[i]->busy_initializing = 0;
4039
4040         rebuild_lun_table(hba[i], 1);
4041         hba[i]->cciss_scan_thread = kthread_run(scan_thread, hba[i],
4042                                 "cciss_scan%02d", i);
4043         if (IS_ERR(hba[i]->cciss_scan_thread))
4044                 return PTR_ERR(hba[i]->cciss_scan_thread);
4045
4046         return 1;
4047
4048 clean4:
4049         kfree(hba[i]->cmd_pool_bits);
4050         if (hba[i]->cmd_pool)
4051                 pci_free_consistent(hba[i]->pdev,
4052                                     hba[i]->nr_cmds * sizeof(CommandList_struct),
4053                                     hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4054         if (hba[i]->errinfo_pool)
4055                 pci_free_consistent(hba[i]->pdev,
4056                                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4057                                     hba[i]->errinfo_pool,
4058                                     hba[i]->errinfo_pool_dhandle);
4059         free_irq(hba[i]->intr[SIMPLE_MODE_INT], hba[i]);
4060 clean2:
4061         unregister_blkdev(hba[i]->major, hba[i]->devname);
4062 clean1:
4063         cciss_destroy_hba_sysfs_entry(hba[i]);
4064 clean0:
4065         hba[i]->busy_initializing = 0;
4066         /* cleanup any queues that may have been initialized */
4067         for (j=0; j <= hba[i]->highest_lun; j++){
4068                 drive_info_struct *drv = &(hba[i]->drv[j]);
4069                 if (drv->queue)
4070                         blk_cleanup_queue(drv->queue);
4071         }
4072         /*
4073          * Deliberately omit pci_disable_device(): it does something nasty to
4074          * Smart Array controllers that pci_enable_device does not undo
4075          */
4076         pci_release_regions(pdev);
4077         pci_set_drvdata(pdev, NULL);
4078         free_hba(i);
4079         return -1;
4080 }
4081
4082 static void cciss_shutdown(struct pci_dev *pdev)
4083 {
4084         ctlr_info_t *tmp_ptr;
4085         int i;
4086         char flush_buf[4];
4087         int return_code;
4088
4089         tmp_ptr = pci_get_drvdata(pdev);
4090         if (tmp_ptr == NULL)
4091                 return;
4092         i = tmp_ptr->ctlr;
4093         if (hba[i] == NULL)
4094                 return;
4095
4096         /* Turn board interrupts off  and send the flush cache command */
4097         /* sendcmd will turn off interrupt, and send the flush...
4098          * To write all data in the battery backed cache to disks */
4099         memset(flush_buf, 0, 4);
4100         return_code = sendcmd(CCISS_CACHE_FLUSH, i, flush_buf, 4, 0,
4101                 CTLR_LUNID, TYPE_CMD);
4102         if (return_code == IO_OK) {
4103                 printk(KERN_INFO "Completed flushing cache on controller %d\n", i);
4104         } else {
4105                 printk(KERN_WARNING "Error flushing cache on controller %d\n", i);
4106         }
4107         free_irq(hba[i]->intr[2], hba[i]);
4108 }
4109
4110 static void __devexit cciss_remove_one(struct pci_dev *pdev)
4111 {
4112         ctlr_info_t *tmp_ptr;
4113         int i, j;
4114
4115         if (pci_get_drvdata(pdev) == NULL) {
4116                 printk(KERN_ERR "cciss: Unable to remove device \n");
4117                 return;
4118         }
4119
4120         tmp_ptr = pci_get_drvdata(pdev);
4121         i = tmp_ptr->ctlr;
4122         if (hba[i] == NULL) {
4123                 printk(KERN_ERR "cciss: device appears to "
4124                        "already be removed \n");
4125                 return;
4126         }
4127
4128         kthread_stop(hba[i]->cciss_scan_thread);
4129
4130         remove_proc_entry(hba[i]->devname, proc_cciss);
4131         unregister_blkdev(hba[i]->major, hba[i]->devname);
4132
4133         /* remove it from the disk list */
4134         for (j = 0; j < CISS_MAX_LUN; j++) {
4135                 struct gendisk *disk = hba[i]->gendisk[j];
4136                 if (disk) {
4137                         struct request_queue *q = disk->queue;
4138
4139                         if (disk->flags & GENHD_FL_UP)
4140                                 del_gendisk(disk);
4141                         if (q)
4142                                 blk_cleanup_queue(q);
4143                 }
4144         }
4145
4146 #ifdef CONFIG_CISS_SCSI_TAPE
4147         cciss_unregister_scsi(i);       /* unhook from SCSI subsystem */
4148 #endif
4149
4150         cciss_shutdown(pdev);
4151
4152 #ifdef CONFIG_PCI_MSI
4153         if (hba[i]->msix_vector)
4154                 pci_disable_msix(hba[i]->pdev);
4155         else if (hba[i]->msi_vector)
4156                 pci_disable_msi(hba[i]->pdev);
4157 #endif                          /* CONFIG_PCI_MSI */
4158
4159         iounmap(hba[i]->vaddr);
4160
4161         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(CommandList_struct),
4162                             hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4163         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4164                             hba[i]->errinfo_pool, hba[i]->errinfo_pool_dhandle);
4165         kfree(hba[i]->cmd_pool_bits);
4166         /*
4167          * Deliberately omit pci_disable_device(): it does something nasty to
4168          * Smart Array controllers that pci_enable_device does not undo
4169          */
4170         pci_release_regions(pdev);
4171         pci_set_drvdata(pdev, NULL);
4172         cciss_destroy_hba_sysfs_entry(hba[i]);
4173         free_hba(i);
4174 }
4175
4176 static struct pci_driver cciss_pci_driver = {
4177         .name = "cciss",
4178         .probe = cciss_init_one,
4179         .remove = __devexit_p(cciss_remove_one),
4180         .id_table = cciss_pci_device_id,        /* id_table */
4181         .shutdown = cciss_shutdown,
4182 };
4183
4184 /*
4185  *  This is it.  Register the PCI driver information for the cards we control
4186  *  the OS will call our registered routines when it finds one of our cards.
4187  */
4188 static int __init cciss_init(void)
4189 {
4190         int err;
4191
4192         /*
4193          * The hardware requires that commands are aligned on a 64-bit
4194          * boundary. Given that we use pci_alloc_consistent() to allocate an
4195          * array of them, the size must be a multiple of 8 bytes.
4196          */
4197         BUILD_BUG_ON(sizeof(CommandList_struct) % 8);
4198
4199         printk(KERN_INFO DRIVER_NAME "\n");
4200
4201         err = bus_register(&cciss_bus_type);
4202         if (err)
4203                 return err;
4204
4205         /* Register for our PCI devices */
4206         err = pci_register_driver(&cciss_pci_driver);
4207         if (err)
4208                 goto err_bus_register;
4209
4210         return 0;
4211
4212 err_bus_register:
4213         bus_unregister(&cciss_bus_type);
4214         return err;
4215 }
4216
4217 static void __exit cciss_cleanup(void)
4218 {
4219         int i;
4220
4221         pci_unregister_driver(&cciss_pci_driver);
4222         /* double check that all controller entrys have been removed */
4223         for (i = 0; i < MAX_CTLR; i++) {
4224                 if (hba[i] != NULL) {
4225                         printk(KERN_WARNING "cciss: had to remove"
4226                                " controller %d\n", i);
4227                         cciss_remove_one(hba[i]->pdev);
4228                 }
4229         }
4230         remove_proc_entry("driver/cciss", NULL);
4231         bus_unregister(&cciss_bus_type);
4232 }
4233
4234 static void fail_all_cmds(unsigned long ctlr)
4235 {
4236         /* If we get here, the board is apparently dead. */
4237         ctlr_info_t *h = hba[ctlr];
4238         CommandList_struct *c;
4239         unsigned long flags;
4240
4241         printk(KERN_WARNING "cciss%d: controller not responding.\n", h->ctlr);
4242         h->alive = 0;           /* the controller apparently died... */
4243
4244         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
4245
4246         pci_disable_device(h->pdev);    /* Make sure it is really dead. */
4247
4248         /* move everything off the request queue onto the completed queue */
4249         while (!hlist_empty(&h->reqQ)) {
4250                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
4251                 removeQ(c);
4252                 h->Qdepth--;
4253                 addQ(&h->cmpQ, c);
4254         }
4255
4256         /* Now, fail everything on the completed queue with a HW error */
4257         while (!hlist_empty(&h->cmpQ)) {
4258                 c = hlist_entry(h->cmpQ.first, CommandList_struct, list);
4259                 removeQ(c);
4260                 if (c->cmd_type != CMD_MSG_STALE)
4261                         c->err_info->CommandStatus = CMD_HARDWARE_ERR;
4262                 if (c->cmd_type == CMD_RWREQ) {
4263                         complete_command(h, c, 0);
4264                 } else if (c->cmd_type == CMD_IOCTL_PEND)
4265                         complete(c->waiting);
4266 #ifdef CONFIG_CISS_SCSI_TAPE
4267                 else if (c->cmd_type == CMD_SCSI)
4268                         complete_scsi_command(c, 0, 0);
4269 #endif
4270         }
4271         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
4272         return;
4273 }
4274
4275 module_init(cciss_init);
4276 module_exit(cciss_cleanup);