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