rbd: warn on update_snaps failure on notify
[linux-2.6.git] / drivers / block / cciss.c
1 /*
2  *    Disk Array driver for HP Smart Array controllers.
3  *    (C) Copyright 2000, 2007 Hewlett-Packard Development Company, L.P.
4  *
5  *    This program is free software; you can redistribute it and/or modify
6  *    it under the terms of the GNU General Public License as published by
7  *    the Free Software Foundation; version 2 of the License.
8  *
9  *    This program is distributed in the hope that it will be useful,
10  *    but WITHOUT ANY WARRANTY; without even the implied warranty of
11  *    MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12  *    General Public License for more details.
13  *
14  *    You should have received a copy of the GNU General Public License
15  *    along with this program; if not, write to the Free Software
16  *    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
17  *    02111-1307, USA.
18  *
19  *    Questions/Comments/Bugfixes to iss_storagedev@hp.com
20  *
21  */
22
23 #include <linux/module.h>
24 #include <linux/interrupt.h>
25 #include <linux/types.h>
26 #include <linux/pci.h>
27 #include <linux/kernel.h>
28 #include <linux/slab.h>
29 #include <linux/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/jiffies.h>
39 #include <linux/hdreg.h>
40 #include <linux/spinlock.h>
41 #include <linux/compat.h>
42 #include <linux/mutex.h>
43 #include <asm/uaccess.h>
44 #include <asm/io.h>
45
46 #include <linux/dma-mapping.h>
47 #include <linux/blkdev.h>
48 #include <linux/genhd.h>
49 #include <linux/completion.h>
50 #include <scsi/scsi.h>
51 #include <scsi/sg.h>
52 #include <scsi/scsi_ioctl.h>
53 #include <linux/cdrom.h>
54 #include <linux/scatterlist.h>
55 #include <linux/kthread.h>
56
57 #define CCISS_DRIVER_VERSION(maj,min,submin) ((maj<<16)|(min<<8)|(submin))
58 #define DRIVER_NAME "HP CISS Driver (v 3.6.26)"
59 #define DRIVER_VERSION CCISS_DRIVER_VERSION(3, 6, 26)
60
61 /* Embedded module documentation macros - see modules.h */
62 MODULE_AUTHOR("Hewlett-Packard Company");
63 MODULE_DESCRIPTION("Driver for HP Smart Array Controllers");
64 MODULE_SUPPORTED_DEVICE("HP Smart Array Controllers");
65 MODULE_VERSION("3.6.26");
66 MODULE_LICENSE("GPL");
67
68 static DEFINE_MUTEX(cciss_mutex);
69 static struct proc_dir_entry *proc_cciss;
70
71 #include "cciss_cmd.h"
72 #include "cciss.h"
73 #include <linux/cciss_ioctl.h>
74
75 /* define the PCI info for the cards we can control */
76 static const struct pci_device_id cciss_pci_device_id[] = {
77         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISS,  0x0E11, 0x4070},
78         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4080},
79         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4082},
80         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4083},
81         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x4091},
82         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409A},
83         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409B},
84         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409C},
85         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409D},
86         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSA,     0x103C, 0x3225},
87         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3223},
88         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3234},
89         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3235},
90         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3211},
91         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3212},
92         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3213},
93         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3214},
94         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3215},
95         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3237},
96         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x323D},
97         {0,}
98 };
99
100 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
101
102 /*  board_id = Subsystem Device ID & Vendor ID
103  *  product = Marketing Name for the board
104  *  access = Address of the struct of function pointers
105  */
106 static struct board_type products[] = {
107         {0x40700E11, "Smart Array 5300", &SA5_access},
108         {0x40800E11, "Smart Array 5i", &SA5B_access},
109         {0x40820E11, "Smart Array 532", &SA5B_access},
110         {0x40830E11, "Smart Array 5312", &SA5B_access},
111         {0x409A0E11, "Smart Array 641", &SA5_access},
112         {0x409B0E11, "Smart Array 642", &SA5_access},
113         {0x409C0E11, "Smart Array 6400", &SA5_access},
114         {0x409D0E11, "Smart Array 6400 EM", &SA5_access},
115         {0x40910E11, "Smart Array 6i", &SA5_access},
116         {0x3225103C, "Smart Array P600", &SA5_access},
117         {0x3223103C, "Smart Array P800", &SA5_access},
118         {0x3234103C, "Smart Array P400", &SA5_access},
119         {0x3235103C, "Smart Array P400i", &SA5_access},
120         {0x3211103C, "Smart Array E200i", &SA5_access},
121         {0x3212103C, "Smart Array E200", &SA5_access},
122         {0x3213103C, "Smart Array E200i", &SA5_access},
123         {0x3214103C, "Smart Array E200i", &SA5_access},
124         {0x3215103C, "Smart Array E200i", &SA5_access},
125         {0x3237103C, "Smart Array E500", &SA5_access},
126         {0x3223103C, "Smart Array P800", &SA5_access},
127         {0x3234103C, "Smart Array P400", &SA5_access},
128         {0x323D103C, "Smart Array P700m", &SA5_access},
129 };
130
131 /* How long to wait (in milliseconds) for board to go into simple mode */
132 #define MAX_CONFIG_WAIT 30000
133 #define MAX_IOCTL_CONFIG_WAIT 1000
134
135 /*define how many times we will try a command because of bus resets */
136 #define MAX_CMD_RETRIES 3
137
138 #define MAX_CTLR        32
139
140 /* Originally cciss driver only supports 8 major numbers */
141 #define MAX_CTLR_ORIG   8
142
143 static ctlr_info_t *hba[MAX_CTLR];
144
145 static struct task_struct *cciss_scan_thread;
146 static DEFINE_MUTEX(scan_mutex);
147 static LIST_HEAD(scan_q);
148
149 static void do_cciss_request(struct request_queue *q);
150 static irqreturn_t do_cciss_intx(int irq, void *dev_id);
151 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id);
152 static int cciss_open(struct block_device *bdev, fmode_t mode);
153 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode);
154 static int cciss_release(struct gendisk *disk, fmode_t mode);
155 static int do_ioctl(struct block_device *bdev, fmode_t mode,
156                     unsigned int cmd, unsigned long arg);
157 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
158                        unsigned int cmd, unsigned long arg);
159 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
160
161 static int cciss_revalidate(struct gendisk *disk);
162 static int rebuild_lun_table(ctlr_info_t *h, int first_time, int via_ioctl);
163 static int deregister_disk(ctlr_info_t *h, int drv_index,
164                            int clear_all, int via_ioctl);
165
166 static void cciss_read_capacity(ctlr_info_t *h, int logvol,
167                         sector_t *total_size, unsigned int *block_size);
168 static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
169                         sector_t *total_size, unsigned int *block_size);
170 static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
171                         sector_t total_size,
172                         unsigned int block_size, InquiryData_struct *inq_buff,
173                                    drive_info_struct *drv);
174 static void __devinit cciss_interrupt_mode(ctlr_info_t *);
175 static void start_io(ctlr_info_t *h);
176 static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
177                         __u8 page_code, unsigned char scsi3addr[],
178                         int cmd_type);
179 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
180         int attempt_retry);
181 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c);
182
183 static int add_to_scan_list(struct ctlr_info *h);
184 static int scan_thread(void *data);
185 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c);
186 static void cciss_hba_release(struct device *dev);
187 static void cciss_device_release(struct device *dev);
188 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index);
189 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index);
190 static inline u32 next_command(ctlr_info_t *h);
191 static int __devinit cciss_find_cfg_addrs(struct pci_dev *pdev,
192         void __iomem *vaddr, u32 *cfg_base_addr, u64 *cfg_base_addr_index,
193         u64 *cfg_offset);
194 static int __devinit cciss_pci_find_memory_BAR(struct pci_dev *pdev,
195         unsigned long *memory_bar);
196 static inline u32 cciss_tag_discard_error_bits(ctlr_info_t *h, u32 tag);
197
198 /* performant mode helper functions */
199 static void  calc_bucket_map(int *bucket, int num_buckets, int nsgs,
200                                 int *bucket_map);
201 static void cciss_put_controller_into_performant_mode(ctlr_info_t *h);
202
203 #ifdef CONFIG_PROC_FS
204 static void cciss_procinit(ctlr_info_t *h);
205 #else
206 static void cciss_procinit(ctlr_info_t *h)
207 {
208 }
209 #endif                          /* CONFIG_PROC_FS */
210
211 #ifdef CONFIG_COMPAT
212 static int cciss_compat_ioctl(struct block_device *, fmode_t,
213                               unsigned, unsigned long);
214 #endif
215
216 static const struct block_device_operations cciss_fops = {
217         .owner = THIS_MODULE,
218         .open = cciss_unlocked_open,
219         .release = cciss_release,
220         .ioctl = do_ioctl,
221         .getgeo = cciss_getgeo,
222 #ifdef CONFIG_COMPAT
223         .compat_ioctl = cciss_compat_ioctl,
224 #endif
225         .revalidate_disk = cciss_revalidate,
226 };
227
228 /* set_performant_mode: Modify the tag for cciss performant
229  * set bit 0 for pull model, bits 3-1 for block fetch
230  * register number
231  */
232 static void set_performant_mode(ctlr_info_t *h, CommandList_struct *c)
233 {
234         if (likely(h->transMethod & CFGTBL_Trans_Performant))
235                 c->busaddr |= 1 | (h->blockFetchTable[c->Header.SGList] << 1);
236 }
237
238 /*
239  * Enqueuing and dequeuing functions for cmdlists.
240  */
241 static inline void addQ(struct list_head *list, CommandList_struct *c)
242 {
243         list_add_tail(&c->list, list);
244 }
245
246 static inline void removeQ(CommandList_struct *c)
247 {
248         /*
249          * After kexec/dump some commands might still
250          * be in flight, which the firmware will try
251          * to complete. Resetting the firmware doesn't work
252          * with old fw revisions, so we have to mark
253          * them off as 'stale' to prevent the driver from
254          * falling over.
255          */
256         if (WARN_ON(list_empty(&c->list))) {
257                 c->cmd_type = CMD_MSG_STALE;
258                 return;
259         }
260
261         list_del_init(&c->list);
262 }
263
264 static void enqueue_cmd_and_start_io(ctlr_info_t *h,
265         CommandList_struct *c)
266 {
267         unsigned long flags;
268         set_performant_mode(h, c);
269         spin_lock_irqsave(&h->lock, flags);
270         addQ(&h->reqQ, c);
271         h->Qdepth++;
272         if (h->Qdepth > h->maxQsinceinit)
273                 h->maxQsinceinit = h->Qdepth;
274         start_io(h);
275         spin_unlock_irqrestore(&h->lock, flags);
276 }
277
278 static void cciss_free_sg_chain_blocks(SGDescriptor_struct **cmd_sg_list,
279         int nr_cmds)
280 {
281         int i;
282
283         if (!cmd_sg_list)
284                 return;
285         for (i = 0; i < nr_cmds; i++) {
286                 kfree(cmd_sg_list[i]);
287                 cmd_sg_list[i] = NULL;
288         }
289         kfree(cmd_sg_list);
290 }
291
292 static SGDescriptor_struct **cciss_allocate_sg_chain_blocks(
293         ctlr_info_t *h, int chainsize, int nr_cmds)
294 {
295         int j;
296         SGDescriptor_struct **cmd_sg_list;
297
298         if (chainsize <= 0)
299                 return NULL;
300
301         cmd_sg_list = kmalloc(sizeof(*cmd_sg_list) * nr_cmds, GFP_KERNEL);
302         if (!cmd_sg_list)
303                 return NULL;
304
305         /* Build up chain blocks for each command */
306         for (j = 0; j < nr_cmds; j++) {
307                 /* Need a block of chainsized s/g elements. */
308                 cmd_sg_list[j] = kmalloc((chainsize *
309                         sizeof(*cmd_sg_list[j])), GFP_KERNEL);
310                 if (!cmd_sg_list[j]) {
311                         dev_err(&h->pdev->dev, "Cannot get memory "
312                                 "for s/g chains.\n");
313                         goto clean;
314                 }
315         }
316         return cmd_sg_list;
317 clean:
318         cciss_free_sg_chain_blocks(cmd_sg_list, nr_cmds);
319         return NULL;
320 }
321
322 static void cciss_unmap_sg_chain_block(ctlr_info_t *h, CommandList_struct *c)
323 {
324         SGDescriptor_struct *chain_sg;
325         u64bit temp64;
326
327         if (c->Header.SGTotal <= h->max_cmd_sgentries)
328                 return;
329
330         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
331         temp64.val32.lower = chain_sg->Addr.lower;
332         temp64.val32.upper = chain_sg->Addr.upper;
333         pci_unmap_single(h->pdev, temp64.val, chain_sg->Len, PCI_DMA_TODEVICE);
334 }
335
336 static void cciss_map_sg_chain_block(ctlr_info_t *h, CommandList_struct *c,
337         SGDescriptor_struct *chain_block, int len)
338 {
339         SGDescriptor_struct *chain_sg;
340         u64bit temp64;
341
342         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
343         chain_sg->Ext = CCISS_SG_CHAIN;
344         chain_sg->Len = len;
345         temp64.val = pci_map_single(h->pdev, chain_block, len,
346                                 PCI_DMA_TODEVICE);
347         chain_sg->Addr.lower = temp64.val32.lower;
348         chain_sg->Addr.upper = temp64.val32.upper;
349 }
350
351 #include "cciss_scsi.c"         /* For SCSI tape support */
352
353 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
354         "UNKNOWN"
355 };
356 #define RAID_UNKNOWN (ARRAY_SIZE(raid_label)-1)
357
358 #ifdef CONFIG_PROC_FS
359
360 /*
361  * Report information about this controller.
362  */
363 #define ENG_GIG 1000000000
364 #define ENG_GIG_FACTOR (ENG_GIG/512)
365 #define ENGAGE_SCSI     "engage scsi"
366
367 static void cciss_seq_show_header(struct seq_file *seq)
368 {
369         ctlr_info_t *h = seq->private;
370
371         seq_printf(seq, "%s: HP %s Controller\n"
372                 "Board ID: 0x%08lx\n"
373                 "Firmware Version: %c%c%c%c\n"
374                 "IRQ: %d\n"
375                 "Logical drives: %d\n"
376                 "Current Q depth: %d\n"
377                 "Current # commands on controller: %d\n"
378                 "Max Q depth since init: %d\n"
379                 "Max # commands on controller since init: %d\n"
380                 "Max SG entries since init: %d\n",
381                 h->devname,
382                 h->product_name,
383                 (unsigned long)h->board_id,
384                 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
385                 h->firm_ver[3], (unsigned int)h->intr[PERF_MODE_INT],
386                 h->num_luns,
387                 h->Qdepth, h->commands_outstanding,
388                 h->maxQsinceinit, h->max_outstanding, h->maxSG);
389
390 #ifdef CONFIG_CISS_SCSI_TAPE
391         cciss_seq_tape_report(seq, h);
392 #endif /* CONFIG_CISS_SCSI_TAPE */
393 }
394
395 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
396 {
397         ctlr_info_t *h = seq->private;
398         unsigned long flags;
399
400         /* prevent displaying bogus info during configuration
401          * or deconfiguration of a logical volume
402          */
403         spin_lock_irqsave(&h->lock, flags);
404         if (h->busy_configuring) {
405                 spin_unlock_irqrestore(&h->lock, flags);
406                 return ERR_PTR(-EBUSY);
407         }
408         h->busy_configuring = 1;
409         spin_unlock_irqrestore(&h->lock, flags);
410
411         if (*pos == 0)
412                 cciss_seq_show_header(seq);
413
414         return pos;
415 }
416
417 static int cciss_seq_show(struct seq_file *seq, void *v)
418 {
419         sector_t vol_sz, vol_sz_frac;
420         ctlr_info_t *h = seq->private;
421         unsigned ctlr = h->ctlr;
422         loff_t *pos = v;
423         drive_info_struct *drv = h->drv[*pos];
424
425         if (*pos > h->highest_lun)
426                 return 0;
427
428         if (drv == NULL) /* it's possible for h->drv[] to have holes. */
429                 return 0;
430
431         if (drv->heads == 0)
432                 return 0;
433
434         vol_sz = drv->nr_blocks;
435         vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
436         vol_sz_frac *= 100;
437         sector_div(vol_sz_frac, ENG_GIG_FACTOR);
438
439         if (drv->raid_level < 0 || drv->raid_level > RAID_UNKNOWN)
440                 drv->raid_level = RAID_UNKNOWN;
441         seq_printf(seq, "cciss/c%dd%d:"
442                         "\t%4u.%02uGB\tRAID %s\n",
443                         ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
444                         raid_label[drv->raid_level]);
445         return 0;
446 }
447
448 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
449 {
450         ctlr_info_t *h = seq->private;
451
452         if (*pos > h->highest_lun)
453                 return NULL;
454         *pos += 1;
455
456         return pos;
457 }
458
459 static void cciss_seq_stop(struct seq_file *seq, void *v)
460 {
461         ctlr_info_t *h = seq->private;
462
463         /* Only reset h->busy_configuring if we succeeded in setting
464          * it during cciss_seq_start. */
465         if (v == ERR_PTR(-EBUSY))
466                 return;
467
468         h->busy_configuring = 0;
469 }
470
471 static const struct seq_operations cciss_seq_ops = {
472         .start = cciss_seq_start,
473         .show  = cciss_seq_show,
474         .next  = cciss_seq_next,
475         .stop  = cciss_seq_stop,
476 };
477
478 static int cciss_seq_open(struct inode *inode, struct file *file)
479 {
480         int ret = seq_open(file, &cciss_seq_ops);
481         struct seq_file *seq = file->private_data;
482
483         if (!ret)
484                 seq->private = PDE(inode)->data;
485
486         return ret;
487 }
488
489 static ssize_t
490 cciss_proc_write(struct file *file, const char __user *buf,
491                  size_t length, loff_t *ppos)
492 {
493         int err;
494         char *buffer;
495
496 #ifndef CONFIG_CISS_SCSI_TAPE
497         return -EINVAL;
498 #endif
499
500         if (!buf || length > PAGE_SIZE - 1)
501                 return -EINVAL;
502
503         buffer = (char *)__get_free_page(GFP_KERNEL);
504         if (!buffer)
505                 return -ENOMEM;
506
507         err = -EFAULT;
508         if (copy_from_user(buffer, buf, length))
509                 goto out;
510         buffer[length] = '\0';
511
512 #ifdef CONFIG_CISS_SCSI_TAPE
513         if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
514                 struct seq_file *seq = file->private_data;
515                 ctlr_info_t *h = seq->private;
516
517                 err = cciss_engage_scsi(h);
518                 if (err == 0)
519                         err = length;
520         } else
521 #endif /* CONFIG_CISS_SCSI_TAPE */
522                 err = -EINVAL;
523         /* might be nice to have "disengage" too, but it's not
524            safely possible. (only 1 module use count, lock issues.) */
525
526 out:
527         free_page((unsigned long)buffer);
528         return err;
529 }
530
531 static const struct file_operations cciss_proc_fops = {
532         .owner   = THIS_MODULE,
533         .open    = cciss_seq_open,
534         .read    = seq_read,
535         .llseek  = seq_lseek,
536         .release = seq_release,
537         .write   = cciss_proc_write,
538 };
539
540 static void __devinit cciss_procinit(ctlr_info_t *h)
541 {
542         struct proc_dir_entry *pde;
543
544         if (proc_cciss == NULL)
545                 proc_cciss = proc_mkdir("driver/cciss", NULL);
546         if (!proc_cciss)
547                 return;
548         pde = proc_create_data(h->devname, S_IWUSR | S_IRUSR | S_IRGRP |
549                                         S_IROTH, proc_cciss,
550                                         &cciss_proc_fops, h);
551 }
552 #endif                          /* CONFIG_PROC_FS */
553
554 #define MAX_PRODUCT_NAME_LEN 19
555
556 #define to_hba(n) container_of(n, struct ctlr_info, dev)
557 #define to_drv(n) container_of(n, drive_info_struct, dev)
558
559 /* List of controllers which cannot be reset on kexec with reset_devices */
560 static u32 unresettable_controller[] = {
561         0x324a103C, /* Smart Array P712m */
562         0x324b103C, /* SmartArray P711m */
563         0x3223103C, /* Smart Array P800 */
564         0x3234103C, /* Smart Array P400 */
565         0x3235103C, /* Smart Array P400i */
566         0x3211103C, /* Smart Array E200i */
567         0x3212103C, /* Smart Array E200 */
568         0x3213103C, /* Smart Array E200i */
569         0x3214103C, /* Smart Array E200i */
570         0x3215103C, /* Smart Array E200i */
571         0x3237103C, /* Smart Array E500 */
572         0x323D103C, /* Smart Array P700m */
573         0x409C0E11, /* Smart Array 6400 */
574         0x409D0E11, /* Smart Array 6400 EM */
575 };
576
577 static int ctlr_is_resettable(struct ctlr_info *h)
578 {
579         int i;
580
581         for (i = 0; i < ARRAY_SIZE(unresettable_controller); i++)
582                 if (unresettable_controller[i] == h->board_id)
583                         return 0;
584         return 1;
585 }
586
587 static ssize_t host_show_resettable(struct device *dev,
588                                     struct device_attribute *attr,
589                                     char *buf)
590 {
591         struct ctlr_info *h = to_hba(dev);
592
593         return snprintf(buf, 20, "%d\n", ctlr_is_resettable(h));
594 }
595 static DEVICE_ATTR(resettable, S_IRUGO, host_show_resettable, NULL);
596
597 static ssize_t host_store_rescan(struct device *dev,
598                                  struct device_attribute *attr,
599                                  const char *buf, size_t count)
600 {
601         struct ctlr_info *h = to_hba(dev);
602
603         add_to_scan_list(h);
604         wake_up_process(cciss_scan_thread);
605         wait_for_completion_interruptible(&h->scan_wait);
606
607         return count;
608 }
609 static DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
610
611 static ssize_t dev_show_unique_id(struct device *dev,
612                                  struct device_attribute *attr,
613                                  char *buf)
614 {
615         drive_info_struct *drv = to_drv(dev);
616         struct ctlr_info *h = to_hba(drv->dev.parent);
617         __u8 sn[16];
618         unsigned long flags;
619         int ret = 0;
620
621         spin_lock_irqsave(&h->lock, flags);
622         if (h->busy_configuring)
623                 ret = -EBUSY;
624         else
625                 memcpy(sn, drv->serial_no, sizeof(sn));
626         spin_unlock_irqrestore(&h->lock, flags);
627
628         if (ret)
629                 return ret;
630         else
631                 return snprintf(buf, 16 * 2 + 2,
632                                 "%02X%02X%02X%02X%02X%02X%02X%02X"
633                                 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
634                                 sn[0], sn[1], sn[2], sn[3],
635                                 sn[4], sn[5], sn[6], sn[7],
636                                 sn[8], sn[9], sn[10], sn[11],
637                                 sn[12], sn[13], sn[14], sn[15]);
638 }
639 static DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
640
641 static ssize_t dev_show_vendor(struct device *dev,
642                                struct device_attribute *attr,
643                                char *buf)
644 {
645         drive_info_struct *drv = to_drv(dev);
646         struct ctlr_info *h = to_hba(drv->dev.parent);
647         char vendor[VENDOR_LEN + 1];
648         unsigned long flags;
649         int ret = 0;
650
651         spin_lock_irqsave(&h->lock, flags);
652         if (h->busy_configuring)
653                 ret = -EBUSY;
654         else
655                 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
656         spin_unlock_irqrestore(&h->lock, flags);
657
658         if (ret)
659                 return ret;
660         else
661                 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
662 }
663 static DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
664
665 static ssize_t dev_show_model(struct device *dev,
666                               struct device_attribute *attr,
667                               char *buf)
668 {
669         drive_info_struct *drv = to_drv(dev);
670         struct ctlr_info *h = to_hba(drv->dev.parent);
671         char model[MODEL_LEN + 1];
672         unsigned long flags;
673         int ret = 0;
674
675         spin_lock_irqsave(&h->lock, flags);
676         if (h->busy_configuring)
677                 ret = -EBUSY;
678         else
679                 memcpy(model, drv->model, MODEL_LEN + 1);
680         spin_unlock_irqrestore(&h->lock, flags);
681
682         if (ret)
683                 return ret;
684         else
685                 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
686 }
687 static DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
688
689 static ssize_t dev_show_rev(struct device *dev,
690                             struct device_attribute *attr,
691                             char *buf)
692 {
693         drive_info_struct *drv = to_drv(dev);
694         struct ctlr_info *h = to_hba(drv->dev.parent);
695         char rev[REV_LEN + 1];
696         unsigned long flags;
697         int ret = 0;
698
699         spin_lock_irqsave(&h->lock, flags);
700         if (h->busy_configuring)
701                 ret = -EBUSY;
702         else
703                 memcpy(rev, drv->rev, REV_LEN + 1);
704         spin_unlock_irqrestore(&h->lock, flags);
705
706         if (ret)
707                 return ret;
708         else
709                 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
710 }
711 static DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
712
713 static ssize_t cciss_show_lunid(struct device *dev,
714                                 struct device_attribute *attr, char *buf)
715 {
716         drive_info_struct *drv = to_drv(dev);
717         struct ctlr_info *h = to_hba(drv->dev.parent);
718         unsigned long flags;
719         unsigned char lunid[8];
720
721         spin_lock_irqsave(&h->lock, flags);
722         if (h->busy_configuring) {
723                 spin_unlock_irqrestore(&h->lock, flags);
724                 return -EBUSY;
725         }
726         if (!drv->heads) {
727                 spin_unlock_irqrestore(&h->lock, flags);
728                 return -ENOTTY;
729         }
730         memcpy(lunid, drv->LunID, sizeof(lunid));
731         spin_unlock_irqrestore(&h->lock, flags);
732         return snprintf(buf, 20, "0x%02x%02x%02x%02x%02x%02x%02x%02x\n",
733                 lunid[0], lunid[1], lunid[2], lunid[3],
734                 lunid[4], lunid[5], lunid[6], lunid[7]);
735 }
736 static DEVICE_ATTR(lunid, S_IRUGO, cciss_show_lunid, NULL);
737
738 static ssize_t cciss_show_raid_level(struct device *dev,
739                                      struct device_attribute *attr, char *buf)
740 {
741         drive_info_struct *drv = to_drv(dev);
742         struct ctlr_info *h = to_hba(drv->dev.parent);
743         int raid;
744         unsigned long flags;
745
746         spin_lock_irqsave(&h->lock, flags);
747         if (h->busy_configuring) {
748                 spin_unlock_irqrestore(&h->lock, flags);
749                 return -EBUSY;
750         }
751         raid = drv->raid_level;
752         spin_unlock_irqrestore(&h->lock, flags);
753         if (raid < 0 || raid > RAID_UNKNOWN)
754                 raid = RAID_UNKNOWN;
755
756         return snprintf(buf, strlen(raid_label[raid]) + 7, "RAID %s\n",
757                         raid_label[raid]);
758 }
759 static DEVICE_ATTR(raid_level, S_IRUGO, cciss_show_raid_level, NULL);
760
761 static ssize_t cciss_show_usage_count(struct device *dev,
762                                       struct device_attribute *attr, char *buf)
763 {
764         drive_info_struct *drv = to_drv(dev);
765         struct ctlr_info *h = to_hba(drv->dev.parent);
766         unsigned long flags;
767         int count;
768
769         spin_lock_irqsave(&h->lock, flags);
770         if (h->busy_configuring) {
771                 spin_unlock_irqrestore(&h->lock, flags);
772                 return -EBUSY;
773         }
774         count = drv->usage_count;
775         spin_unlock_irqrestore(&h->lock, flags);
776         return snprintf(buf, 20, "%d\n", count);
777 }
778 static DEVICE_ATTR(usage_count, S_IRUGO, cciss_show_usage_count, NULL);
779
780 static struct attribute *cciss_host_attrs[] = {
781         &dev_attr_rescan.attr,
782         &dev_attr_resettable.attr,
783         NULL
784 };
785
786 static struct attribute_group cciss_host_attr_group = {
787         .attrs = cciss_host_attrs,
788 };
789
790 static const struct attribute_group *cciss_host_attr_groups[] = {
791         &cciss_host_attr_group,
792         NULL
793 };
794
795 static struct device_type cciss_host_type = {
796         .name           = "cciss_host",
797         .groups         = cciss_host_attr_groups,
798         .release        = cciss_hba_release,
799 };
800
801 static struct attribute *cciss_dev_attrs[] = {
802         &dev_attr_unique_id.attr,
803         &dev_attr_model.attr,
804         &dev_attr_vendor.attr,
805         &dev_attr_rev.attr,
806         &dev_attr_lunid.attr,
807         &dev_attr_raid_level.attr,
808         &dev_attr_usage_count.attr,
809         NULL
810 };
811
812 static struct attribute_group cciss_dev_attr_group = {
813         .attrs = cciss_dev_attrs,
814 };
815
816 static const struct attribute_group *cciss_dev_attr_groups[] = {
817         &cciss_dev_attr_group,
818         NULL
819 };
820
821 static struct device_type cciss_dev_type = {
822         .name           = "cciss_device",
823         .groups         = cciss_dev_attr_groups,
824         .release        = cciss_device_release,
825 };
826
827 static struct bus_type cciss_bus_type = {
828         .name           = "cciss",
829 };
830
831 /*
832  * cciss_hba_release is called when the reference count
833  * of h->dev goes to zero.
834  */
835 static void cciss_hba_release(struct device *dev)
836 {
837         /*
838          * nothing to do, but need this to avoid a warning
839          * about not having a release handler from lib/kref.c.
840          */
841 }
842
843 /*
844  * Initialize sysfs entry for each controller.  This sets up and registers
845  * the 'cciss#' directory for each individual controller under
846  * /sys/bus/pci/devices/<dev>/.
847  */
848 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
849 {
850         device_initialize(&h->dev);
851         h->dev.type = &cciss_host_type;
852         h->dev.bus = &cciss_bus_type;
853         dev_set_name(&h->dev, "%s", h->devname);
854         h->dev.parent = &h->pdev->dev;
855
856         return device_add(&h->dev);
857 }
858
859 /*
860  * Remove sysfs entries for an hba.
861  */
862 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
863 {
864         device_del(&h->dev);
865         put_device(&h->dev); /* final put. */
866 }
867
868 /* cciss_device_release is called when the reference count
869  * of h->drv[x]dev goes to zero.
870  */
871 static void cciss_device_release(struct device *dev)
872 {
873         drive_info_struct *drv = to_drv(dev);
874         kfree(drv);
875 }
876
877 /*
878  * Initialize sysfs for each logical drive.  This sets up and registers
879  * the 'c#d#' directory for each individual logical drive under
880  * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
881  * /sys/block/cciss!c#d# to this entry.
882  */
883 static long cciss_create_ld_sysfs_entry(struct ctlr_info *h,
884                                        int drv_index)
885 {
886         struct device *dev;
887
888         if (h->drv[drv_index]->device_initialized)
889                 return 0;
890
891         dev = &h->drv[drv_index]->dev;
892         device_initialize(dev);
893         dev->type = &cciss_dev_type;
894         dev->bus = &cciss_bus_type;
895         dev_set_name(dev, "c%dd%d", h->ctlr, drv_index);
896         dev->parent = &h->dev;
897         h->drv[drv_index]->device_initialized = 1;
898         return device_add(dev);
899 }
900
901 /*
902  * Remove sysfs entries for a logical drive.
903  */
904 static void cciss_destroy_ld_sysfs_entry(struct ctlr_info *h, int drv_index,
905         int ctlr_exiting)
906 {
907         struct device *dev = &h->drv[drv_index]->dev;
908
909         /* special case for c*d0, we only destroy it on controller exit */
910         if (drv_index == 0 && !ctlr_exiting)
911                 return;
912
913         device_del(dev);
914         put_device(dev); /* the "final" put. */
915         h->drv[drv_index] = NULL;
916 }
917
918 /*
919  * For operations that cannot sleep, a command block is allocated at init,
920  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
921  * which ones are free or in use.
922  */
923 static CommandList_struct *cmd_alloc(ctlr_info_t *h)
924 {
925         CommandList_struct *c;
926         int i;
927         u64bit temp64;
928         dma_addr_t cmd_dma_handle, err_dma_handle;
929
930         do {
931                 i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
932                 if (i == h->nr_cmds)
933                         return NULL;
934         } while (test_and_set_bit(i & (BITS_PER_LONG - 1),
935                   h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
936         c = h->cmd_pool + i;
937         memset(c, 0, sizeof(CommandList_struct));
938         cmd_dma_handle = h->cmd_pool_dhandle + i * sizeof(CommandList_struct);
939         c->err_info = h->errinfo_pool + i;
940         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
941         err_dma_handle = h->errinfo_pool_dhandle
942             + i * sizeof(ErrorInfo_struct);
943         h->nr_allocs++;
944
945         c->cmdindex = i;
946
947         INIT_LIST_HEAD(&c->list);
948         c->busaddr = (__u32) cmd_dma_handle;
949         temp64.val = (__u64) err_dma_handle;
950         c->ErrDesc.Addr.lower = temp64.val32.lower;
951         c->ErrDesc.Addr.upper = temp64.val32.upper;
952         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
953
954         c->ctlr = h->ctlr;
955         return c;
956 }
957
958 /* allocate a command using pci_alloc_consistent, used for ioctls,
959  * etc., not for the main i/o path.
960  */
961 static CommandList_struct *cmd_special_alloc(ctlr_info_t *h)
962 {
963         CommandList_struct *c;
964         u64bit temp64;
965         dma_addr_t cmd_dma_handle, err_dma_handle;
966
967         c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
968                 sizeof(CommandList_struct), &cmd_dma_handle);
969         if (c == NULL)
970                 return NULL;
971         memset(c, 0, sizeof(CommandList_struct));
972
973         c->cmdindex = -1;
974
975         c->err_info = (ErrorInfo_struct *)
976             pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
977                     &err_dma_handle);
978
979         if (c->err_info == NULL) {
980                 pci_free_consistent(h->pdev,
981                         sizeof(CommandList_struct), c, cmd_dma_handle);
982                 return NULL;
983         }
984         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
985
986         INIT_LIST_HEAD(&c->list);
987         c->busaddr = (__u32) cmd_dma_handle;
988         temp64.val = (__u64) err_dma_handle;
989         c->ErrDesc.Addr.lower = temp64.val32.lower;
990         c->ErrDesc.Addr.upper = temp64.val32.upper;
991         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
992
993         c->ctlr = h->ctlr;
994         return c;
995 }
996
997 static void cmd_free(ctlr_info_t *h, CommandList_struct *c)
998 {
999         int i;
1000
1001         i = c - h->cmd_pool;
1002         clear_bit(i & (BITS_PER_LONG - 1),
1003                   h->cmd_pool_bits + (i / BITS_PER_LONG));
1004         h->nr_frees++;
1005 }
1006
1007 static void cmd_special_free(ctlr_info_t *h, CommandList_struct *c)
1008 {
1009         u64bit temp64;
1010
1011         temp64.val32.lower = c->ErrDesc.Addr.lower;
1012         temp64.val32.upper = c->ErrDesc.Addr.upper;
1013         pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
1014                             c->err_info, (dma_addr_t) temp64.val);
1015         pci_free_consistent(h->pdev, sizeof(CommandList_struct), c,
1016                 (dma_addr_t) cciss_tag_discard_error_bits(h, (u32) c->busaddr));
1017 }
1018
1019 static inline ctlr_info_t *get_host(struct gendisk *disk)
1020 {
1021         return disk->queue->queuedata;
1022 }
1023
1024 static inline drive_info_struct *get_drv(struct gendisk *disk)
1025 {
1026         return disk->private_data;
1027 }
1028
1029 /*
1030  * Open.  Make sure the device is really there.
1031  */
1032 static int cciss_open(struct block_device *bdev, fmode_t mode)
1033 {
1034         ctlr_info_t *h = get_host(bdev->bd_disk);
1035         drive_info_struct *drv = get_drv(bdev->bd_disk);
1036
1037         dev_dbg(&h->pdev->dev, "cciss_open %s\n", bdev->bd_disk->disk_name);
1038         if (drv->busy_configuring)
1039                 return -EBUSY;
1040         /*
1041          * Root is allowed to open raw volume zero even if it's not configured
1042          * so array config can still work. Root is also allowed to open any
1043          * volume that has a LUN ID, so it can issue IOCTL to reread the
1044          * disk information.  I don't think I really like this
1045          * but I'm already using way to many device nodes to claim another one
1046          * for "raw controller".
1047          */
1048         if (drv->heads == 0) {
1049                 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
1050                         /* if not node 0 make sure it is a partition = 0 */
1051                         if (MINOR(bdev->bd_dev) & 0x0f) {
1052                                 return -ENXIO;
1053                                 /* if it is, make sure we have a LUN ID */
1054                         } else if (memcmp(drv->LunID, CTLR_LUNID,
1055                                 sizeof(drv->LunID))) {
1056                                 return -ENXIO;
1057                         }
1058                 }
1059                 if (!capable(CAP_SYS_ADMIN))
1060                         return -EPERM;
1061         }
1062         drv->usage_count++;
1063         h->usage_count++;
1064         return 0;
1065 }
1066
1067 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode)
1068 {
1069         int ret;
1070
1071         mutex_lock(&cciss_mutex);
1072         ret = cciss_open(bdev, mode);
1073         mutex_unlock(&cciss_mutex);
1074
1075         return ret;
1076 }
1077
1078 /*
1079  * Close.  Sync first.
1080  */
1081 static int cciss_release(struct gendisk *disk, fmode_t mode)
1082 {
1083         ctlr_info_t *h;
1084         drive_info_struct *drv;
1085
1086         mutex_lock(&cciss_mutex);
1087         h = get_host(disk);
1088         drv = get_drv(disk);
1089         dev_dbg(&h->pdev->dev, "cciss_release %s\n", disk->disk_name);
1090         drv->usage_count--;
1091         h->usage_count--;
1092         mutex_unlock(&cciss_mutex);
1093         return 0;
1094 }
1095
1096 static int do_ioctl(struct block_device *bdev, fmode_t mode,
1097                     unsigned cmd, unsigned long arg)
1098 {
1099         int ret;
1100         mutex_lock(&cciss_mutex);
1101         ret = cciss_ioctl(bdev, mode, cmd, arg);
1102         mutex_unlock(&cciss_mutex);
1103         return ret;
1104 }
1105
1106 #ifdef CONFIG_COMPAT
1107
1108 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1109                                   unsigned cmd, unsigned long arg);
1110 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1111                                       unsigned cmd, unsigned long arg);
1112
1113 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
1114                               unsigned cmd, unsigned long arg)
1115 {
1116         switch (cmd) {
1117         case CCISS_GETPCIINFO:
1118         case CCISS_GETINTINFO:
1119         case CCISS_SETINTINFO:
1120         case CCISS_GETNODENAME:
1121         case CCISS_SETNODENAME:
1122         case CCISS_GETHEARTBEAT:
1123         case CCISS_GETBUSTYPES:
1124         case CCISS_GETFIRMVER:
1125         case CCISS_GETDRIVVER:
1126         case CCISS_REVALIDVOLS:
1127         case CCISS_DEREGDISK:
1128         case CCISS_REGNEWDISK:
1129         case CCISS_REGNEWD:
1130         case CCISS_RESCANDISK:
1131         case CCISS_GETLUNINFO:
1132                 return do_ioctl(bdev, mode, cmd, arg);
1133
1134         case CCISS_PASSTHRU32:
1135                 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
1136         case CCISS_BIG_PASSTHRU32:
1137                 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
1138
1139         default:
1140                 return -ENOIOCTLCMD;
1141         }
1142 }
1143
1144 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1145                                   unsigned cmd, unsigned long arg)
1146 {
1147         IOCTL32_Command_struct __user *arg32 =
1148             (IOCTL32_Command_struct __user *) arg;
1149         IOCTL_Command_struct arg64;
1150         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
1151         int err;
1152         u32 cp;
1153
1154         err = 0;
1155         err |=
1156             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1157                            sizeof(arg64.LUN_info));
1158         err |=
1159             copy_from_user(&arg64.Request, &arg32->Request,
1160                            sizeof(arg64.Request));
1161         err |=
1162             copy_from_user(&arg64.error_info, &arg32->error_info,
1163                            sizeof(arg64.error_info));
1164         err |= get_user(arg64.buf_size, &arg32->buf_size);
1165         err |= get_user(cp, &arg32->buf);
1166         arg64.buf = compat_ptr(cp);
1167         err |= copy_to_user(p, &arg64, sizeof(arg64));
1168
1169         if (err)
1170                 return -EFAULT;
1171
1172         err = do_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
1173         if (err)
1174                 return err;
1175         err |=
1176             copy_in_user(&arg32->error_info, &p->error_info,
1177                          sizeof(arg32->error_info));
1178         if (err)
1179                 return -EFAULT;
1180         return err;
1181 }
1182
1183 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1184                                       unsigned cmd, unsigned long arg)
1185 {
1186         BIG_IOCTL32_Command_struct __user *arg32 =
1187             (BIG_IOCTL32_Command_struct __user *) arg;
1188         BIG_IOCTL_Command_struct arg64;
1189         BIG_IOCTL_Command_struct __user *p =
1190             compat_alloc_user_space(sizeof(arg64));
1191         int err;
1192         u32 cp;
1193
1194         memset(&arg64, 0, sizeof(arg64));
1195         err = 0;
1196         err |=
1197             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1198                            sizeof(arg64.LUN_info));
1199         err |=
1200             copy_from_user(&arg64.Request, &arg32->Request,
1201                            sizeof(arg64.Request));
1202         err |=
1203             copy_from_user(&arg64.error_info, &arg32->error_info,
1204                            sizeof(arg64.error_info));
1205         err |= get_user(arg64.buf_size, &arg32->buf_size);
1206         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1207         err |= get_user(cp, &arg32->buf);
1208         arg64.buf = compat_ptr(cp);
1209         err |= copy_to_user(p, &arg64, sizeof(arg64));
1210
1211         if (err)
1212                 return -EFAULT;
1213
1214         err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1215         if (err)
1216                 return err;
1217         err |=
1218             copy_in_user(&arg32->error_info, &p->error_info,
1219                          sizeof(arg32->error_info));
1220         if (err)
1221                 return -EFAULT;
1222         return err;
1223 }
1224 #endif
1225
1226 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1227 {
1228         drive_info_struct *drv = get_drv(bdev->bd_disk);
1229
1230         if (!drv->cylinders)
1231                 return -ENXIO;
1232
1233         geo->heads = drv->heads;
1234         geo->sectors = drv->sectors;
1235         geo->cylinders = drv->cylinders;
1236         return 0;
1237 }
1238
1239 static void check_ioctl_unit_attention(ctlr_info_t *h, CommandList_struct *c)
1240 {
1241         if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1242                         c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1243                 (void)check_for_unit_attention(h, c);
1244 }
1245
1246 static int cciss_getpciinfo(ctlr_info_t *h, void __user *argp)
1247 {
1248         cciss_pci_info_struct pciinfo;
1249
1250         if (!argp)
1251                 return -EINVAL;
1252         pciinfo.domain = pci_domain_nr(h->pdev->bus);
1253         pciinfo.bus = h->pdev->bus->number;
1254         pciinfo.dev_fn = h->pdev->devfn;
1255         pciinfo.board_id = h->board_id;
1256         if (copy_to_user(argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1257                 return -EFAULT;
1258         return 0;
1259 }
1260
1261 static int cciss_getintinfo(ctlr_info_t *h, void __user *argp)
1262 {
1263         cciss_coalint_struct intinfo;
1264
1265         if (!argp)
1266                 return -EINVAL;
1267         intinfo.delay = readl(&h->cfgtable->HostWrite.CoalIntDelay);
1268         intinfo.count = readl(&h->cfgtable->HostWrite.CoalIntCount);
1269         if (copy_to_user
1270             (argp, &intinfo, sizeof(cciss_coalint_struct)))
1271                 return -EFAULT;
1272         return 0;
1273 }
1274
1275 static int cciss_setintinfo(ctlr_info_t *h, void __user *argp)
1276 {
1277         cciss_coalint_struct intinfo;
1278         unsigned long flags;
1279         int i;
1280
1281         if (!argp)
1282                 return -EINVAL;
1283         if (!capable(CAP_SYS_ADMIN))
1284                 return -EPERM;
1285         if (copy_from_user(&intinfo, argp, sizeof(intinfo)))
1286                 return -EFAULT;
1287         if ((intinfo.delay == 0) && (intinfo.count == 0))
1288                 return -EINVAL;
1289         spin_lock_irqsave(&h->lock, flags);
1290         /* Update the field, and then ring the doorbell */
1291         writel(intinfo.delay, &(h->cfgtable->HostWrite.CoalIntDelay));
1292         writel(intinfo.count, &(h->cfgtable->HostWrite.CoalIntCount));
1293         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1294
1295         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1296                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
1297                         break;
1298                 udelay(1000); /* delay and try again */
1299         }
1300         spin_unlock_irqrestore(&h->lock, flags);
1301         if (i >= MAX_IOCTL_CONFIG_WAIT)
1302                 return -EAGAIN;
1303         return 0;
1304 }
1305
1306 static int cciss_getnodename(ctlr_info_t *h, void __user *argp)
1307 {
1308         NodeName_type NodeName;
1309         int i;
1310
1311         if (!argp)
1312                 return -EINVAL;
1313         for (i = 0; i < 16; i++)
1314                 NodeName[i] = readb(&h->cfgtable->ServerName[i]);
1315         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1316                 return -EFAULT;
1317         return 0;
1318 }
1319
1320 static int cciss_setnodename(ctlr_info_t *h, void __user *argp)
1321 {
1322         NodeName_type NodeName;
1323         unsigned long flags;
1324         int i;
1325
1326         if (!argp)
1327                 return -EINVAL;
1328         if (!capable(CAP_SYS_ADMIN))
1329                 return -EPERM;
1330         if (copy_from_user(NodeName, argp, sizeof(NodeName_type)))
1331                 return -EFAULT;
1332         spin_lock_irqsave(&h->lock, flags);
1333         /* Update the field, and then ring the doorbell */
1334         for (i = 0; i < 16; i++)
1335                 writeb(NodeName[i], &h->cfgtable->ServerName[i]);
1336         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1337         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1338                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
1339                         break;
1340                 udelay(1000); /* delay and try again */
1341         }
1342         spin_unlock_irqrestore(&h->lock, flags);
1343         if (i >= MAX_IOCTL_CONFIG_WAIT)
1344                 return -EAGAIN;
1345         return 0;
1346 }
1347
1348 static int cciss_getheartbeat(ctlr_info_t *h, void __user *argp)
1349 {
1350         Heartbeat_type heartbeat;
1351
1352         if (!argp)
1353                 return -EINVAL;
1354         heartbeat = readl(&h->cfgtable->HeartBeat);
1355         if (copy_to_user(argp, &heartbeat, sizeof(Heartbeat_type)))
1356                 return -EFAULT;
1357         return 0;
1358 }
1359
1360 static int cciss_getbustypes(ctlr_info_t *h, void __user *argp)
1361 {
1362         BusTypes_type BusTypes;
1363
1364         if (!argp)
1365                 return -EINVAL;
1366         BusTypes = readl(&h->cfgtable->BusTypes);
1367         if (copy_to_user(argp, &BusTypes, sizeof(BusTypes_type)))
1368                 return -EFAULT;
1369         return 0;
1370 }
1371
1372 static int cciss_getfirmver(ctlr_info_t *h, void __user *argp)
1373 {
1374         FirmwareVer_type firmware;
1375
1376         if (!argp)
1377                 return -EINVAL;
1378         memcpy(firmware, h->firm_ver, 4);
1379
1380         if (copy_to_user
1381             (argp, firmware, sizeof(FirmwareVer_type)))
1382                 return -EFAULT;
1383         return 0;
1384 }
1385
1386 static int cciss_getdrivver(ctlr_info_t *h, void __user *argp)
1387 {
1388         DriverVer_type DriverVer = DRIVER_VERSION;
1389
1390         if (!argp)
1391                 return -EINVAL;
1392         if (copy_to_user(argp, &DriverVer, sizeof(DriverVer_type)))
1393                 return -EFAULT;
1394         return 0;
1395 }
1396
1397 static int cciss_getluninfo(ctlr_info_t *h,
1398         struct gendisk *disk, void __user *argp)
1399 {
1400         LogvolInfo_struct luninfo;
1401         drive_info_struct *drv = get_drv(disk);
1402
1403         if (!argp)
1404                 return -EINVAL;
1405         memcpy(&luninfo.LunID, drv->LunID, sizeof(luninfo.LunID));
1406         luninfo.num_opens = drv->usage_count;
1407         luninfo.num_parts = 0;
1408         if (copy_to_user(argp, &luninfo, sizeof(LogvolInfo_struct)))
1409                 return -EFAULT;
1410         return 0;
1411 }
1412
1413 static int cciss_passthru(ctlr_info_t *h, void __user *argp)
1414 {
1415         IOCTL_Command_struct iocommand;
1416         CommandList_struct *c;
1417         char *buff = NULL;
1418         u64bit temp64;
1419         DECLARE_COMPLETION_ONSTACK(wait);
1420
1421         if (!argp)
1422                 return -EINVAL;
1423
1424         if (!capable(CAP_SYS_RAWIO))
1425                 return -EPERM;
1426
1427         if (copy_from_user
1428             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1429                 return -EFAULT;
1430         if ((iocommand.buf_size < 1) &&
1431             (iocommand.Request.Type.Direction != XFER_NONE)) {
1432                 return -EINVAL;
1433         }
1434         if (iocommand.buf_size > 0) {
1435                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1436                 if (buff == NULL)
1437                         return -EFAULT;
1438         }
1439         if (iocommand.Request.Type.Direction == XFER_WRITE) {
1440                 /* Copy the data into the buffer we created */
1441                 if (copy_from_user(buff, iocommand.buf, iocommand.buf_size)) {
1442                         kfree(buff);
1443                         return -EFAULT;
1444                 }
1445         } else {
1446                 memset(buff, 0, iocommand.buf_size);
1447         }
1448         c = cmd_special_alloc(h);
1449         if (!c) {
1450                 kfree(buff);
1451                 return -ENOMEM;
1452         }
1453         /* Fill in the command type */
1454         c->cmd_type = CMD_IOCTL_PEND;
1455         /* Fill in Command Header */
1456         c->Header.ReplyQueue = 0;   /* unused in simple mode */
1457         if (iocommand.buf_size > 0) { /* buffer to fill */
1458                 c->Header.SGList = 1;
1459                 c->Header.SGTotal = 1;
1460         } else { /* no buffers to fill */
1461                 c->Header.SGList = 0;
1462                 c->Header.SGTotal = 0;
1463         }
1464         c->Header.LUN = iocommand.LUN_info;
1465         /* use the kernel address the cmd block for tag */
1466         c->Header.Tag.lower = c->busaddr;
1467
1468         /* Fill in Request block */
1469         c->Request = iocommand.Request;
1470
1471         /* Fill in the scatter gather information */
1472         if (iocommand.buf_size > 0) {
1473                 temp64.val = pci_map_single(h->pdev, buff,
1474                         iocommand.buf_size, PCI_DMA_BIDIRECTIONAL);
1475                 c->SG[0].Addr.lower = temp64.val32.lower;
1476                 c->SG[0].Addr.upper = temp64.val32.upper;
1477                 c->SG[0].Len = iocommand.buf_size;
1478                 c->SG[0].Ext = 0;  /* we are not chaining */
1479         }
1480         c->waiting = &wait;
1481
1482         enqueue_cmd_and_start_io(h, c);
1483         wait_for_completion(&wait);
1484
1485         /* unlock the buffers from DMA */
1486         temp64.val32.lower = c->SG[0].Addr.lower;
1487         temp64.val32.upper = c->SG[0].Addr.upper;
1488         pci_unmap_single(h->pdev, (dma_addr_t) temp64.val, iocommand.buf_size,
1489                          PCI_DMA_BIDIRECTIONAL);
1490         check_ioctl_unit_attention(h, c);
1491
1492         /* Copy the error information out */
1493         iocommand.error_info = *(c->err_info);
1494         if (copy_to_user(argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1495                 kfree(buff);
1496                 cmd_special_free(h, c);
1497                 return -EFAULT;
1498         }
1499
1500         if (iocommand.Request.Type.Direction == XFER_READ) {
1501                 /* Copy the data out of the buffer we created */
1502                 if (copy_to_user(iocommand.buf, buff, iocommand.buf_size)) {
1503                         kfree(buff);
1504                         cmd_special_free(h, c);
1505                         return -EFAULT;
1506                 }
1507         }
1508         kfree(buff);
1509         cmd_special_free(h, c);
1510         return 0;
1511 }
1512
1513 static int cciss_bigpassthru(ctlr_info_t *h, void __user *argp)
1514 {
1515         BIG_IOCTL_Command_struct *ioc;
1516         CommandList_struct *c;
1517         unsigned char **buff = NULL;
1518         int *buff_size = NULL;
1519         u64bit temp64;
1520         BYTE sg_used = 0;
1521         int status = 0;
1522         int i;
1523         DECLARE_COMPLETION_ONSTACK(wait);
1524         __u32 left;
1525         __u32 sz;
1526         BYTE __user *data_ptr;
1527
1528         if (!argp)
1529                 return -EINVAL;
1530         if (!capable(CAP_SYS_RAWIO))
1531                 return -EPERM;
1532         ioc = kmalloc(sizeof(*ioc), GFP_KERNEL);
1533         if (!ioc) {
1534                 status = -ENOMEM;
1535                 goto cleanup1;
1536         }
1537         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1538                 status = -EFAULT;
1539                 goto cleanup1;
1540         }
1541         if ((ioc->buf_size < 1) &&
1542             (ioc->Request.Type.Direction != XFER_NONE)) {
1543                 status = -EINVAL;
1544                 goto cleanup1;
1545         }
1546         /* Check kmalloc limits  using all SGs */
1547         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1548                 status = -EINVAL;
1549                 goto cleanup1;
1550         }
1551         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1552                 status = -EINVAL;
1553                 goto cleanup1;
1554         }
1555         buff = kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1556         if (!buff) {
1557                 status = -ENOMEM;
1558                 goto cleanup1;
1559         }
1560         buff_size = kmalloc(MAXSGENTRIES * sizeof(int), GFP_KERNEL);
1561         if (!buff_size) {
1562                 status = -ENOMEM;
1563                 goto cleanup1;
1564         }
1565         left = ioc->buf_size;
1566         data_ptr = ioc->buf;
1567         while (left) {
1568                 sz = (left > ioc->malloc_size) ? ioc->malloc_size : left;
1569                 buff_size[sg_used] = sz;
1570                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1571                 if (buff[sg_used] == NULL) {
1572                         status = -ENOMEM;
1573                         goto cleanup1;
1574                 }
1575                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1576                         if (copy_from_user(buff[sg_used], data_ptr, sz)) {
1577                                 status = -EFAULT;
1578                                 goto cleanup1;
1579                         }
1580                 } else {
1581                         memset(buff[sg_used], 0, sz);
1582                 }
1583                 left -= sz;
1584                 data_ptr += sz;
1585                 sg_used++;
1586         }
1587         c = cmd_special_alloc(h);
1588         if (!c) {
1589                 status = -ENOMEM;
1590                 goto cleanup1;
1591         }
1592         c->cmd_type = CMD_IOCTL_PEND;
1593         c->Header.ReplyQueue = 0;
1594         c->Header.SGList = sg_used;
1595         c->Header.SGTotal = sg_used;
1596         c->Header.LUN = ioc->LUN_info;
1597         c->Header.Tag.lower = c->busaddr;
1598
1599         c->Request = ioc->Request;
1600         for (i = 0; i < sg_used; i++) {
1601                 temp64.val = pci_map_single(h->pdev, buff[i], buff_size[i],
1602                                     PCI_DMA_BIDIRECTIONAL);
1603                 c->SG[i].Addr.lower = temp64.val32.lower;
1604                 c->SG[i].Addr.upper = temp64.val32.upper;
1605                 c->SG[i].Len = buff_size[i];
1606                 c->SG[i].Ext = 0;       /* we are not chaining */
1607         }
1608         c->waiting = &wait;
1609         enqueue_cmd_and_start_io(h, c);
1610         wait_for_completion(&wait);
1611         /* unlock the buffers from DMA */
1612         for (i = 0; i < sg_used; i++) {
1613                 temp64.val32.lower = c->SG[i].Addr.lower;
1614                 temp64.val32.upper = c->SG[i].Addr.upper;
1615                 pci_unmap_single(h->pdev,
1616                         (dma_addr_t) temp64.val, buff_size[i],
1617                         PCI_DMA_BIDIRECTIONAL);
1618         }
1619         check_ioctl_unit_attention(h, c);
1620         /* Copy the error information out */
1621         ioc->error_info = *(c->err_info);
1622         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1623                 cmd_special_free(h, c);
1624                 status = -EFAULT;
1625                 goto cleanup1;
1626         }
1627         if (ioc->Request.Type.Direction == XFER_READ) {
1628                 /* Copy the data out of the buffer we created */
1629                 BYTE __user *ptr = ioc->buf;
1630                 for (i = 0; i < sg_used; i++) {
1631                         if (copy_to_user(ptr, buff[i], buff_size[i])) {
1632                                 cmd_special_free(h, c);
1633                                 status = -EFAULT;
1634                                 goto cleanup1;
1635                         }
1636                         ptr += buff_size[i];
1637                 }
1638         }
1639         cmd_special_free(h, c);
1640         status = 0;
1641 cleanup1:
1642         if (buff) {
1643                 for (i = 0; i < sg_used; i++)
1644                         kfree(buff[i]);
1645                 kfree(buff);
1646         }
1647         kfree(buff_size);
1648         kfree(ioc);
1649         return status;
1650 }
1651
1652 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1653         unsigned int cmd, unsigned long arg)
1654 {
1655         struct gendisk *disk = bdev->bd_disk;
1656         ctlr_info_t *h = get_host(disk);
1657         void __user *argp = (void __user *)arg;
1658
1659         dev_dbg(&h->pdev->dev, "cciss_ioctl: Called with cmd=%x %lx\n",
1660                 cmd, arg);
1661         switch (cmd) {
1662         case CCISS_GETPCIINFO:
1663                 return cciss_getpciinfo(h, argp);
1664         case CCISS_GETINTINFO:
1665                 return cciss_getintinfo(h, argp);
1666         case CCISS_SETINTINFO:
1667                 return cciss_setintinfo(h, argp);
1668         case CCISS_GETNODENAME:
1669                 return cciss_getnodename(h, argp);
1670         case CCISS_SETNODENAME:
1671                 return cciss_setnodename(h, argp);
1672         case CCISS_GETHEARTBEAT:
1673                 return cciss_getheartbeat(h, argp);
1674         case CCISS_GETBUSTYPES:
1675                 return cciss_getbustypes(h, argp);
1676         case CCISS_GETFIRMVER:
1677                 return cciss_getfirmver(h, argp);
1678         case CCISS_GETDRIVVER:
1679                 return cciss_getdrivver(h, argp);
1680         case CCISS_DEREGDISK:
1681         case CCISS_REGNEWD:
1682         case CCISS_REVALIDVOLS:
1683                 return rebuild_lun_table(h, 0, 1);
1684         case CCISS_GETLUNINFO:
1685                 return cciss_getluninfo(h, disk, argp);
1686         case CCISS_PASSTHRU:
1687                 return cciss_passthru(h, argp);
1688         case CCISS_BIG_PASSTHRU:
1689                 return cciss_bigpassthru(h, argp);
1690
1691         /* scsi_cmd_ioctl handles these, below, though some are not */
1692         /* very meaningful for cciss.  SG_IO is the main one people want. */
1693
1694         case SG_GET_VERSION_NUM:
1695         case SG_SET_TIMEOUT:
1696         case SG_GET_TIMEOUT:
1697         case SG_GET_RESERVED_SIZE:
1698         case SG_SET_RESERVED_SIZE:
1699         case SG_EMULATED_HOST:
1700         case SG_IO:
1701         case SCSI_IOCTL_SEND_COMMAND:
1702                 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
1703
1704         /* scsi_cmd_ioctl would normally handle these, below, but */
1705         /* they aren't a good fit for cciss, as CD-ROMs are */
1706         /* not supported, and we don't have any bus/target/lun */
1707         /* which we present to the kernel. */
1708
1709         case CDROM_SEND_PACKET:
1710         case CDROMCLOSETRAY:
1711         case CDROMEJECT:
1712         case SCSI_IOCTL_GET_IDLUN:
1713         case SCSI_IOCTL_GET_BUS_NUMBER:
1714         default:
1715                 return -ENOTTY;
1716         }
1717 }
1718
1719 static void cciss_check_queues(ctlr_info_t *h)
1720 {
1721         int start_queue = h->next_to_run;
1722         int i;
1723
1724         /* check to see if we have maxed out the number of commands that can
1725          * be placed on the queue.  If so then exit.  We do this check here
1726          * in case the interrupt we serviced was from an ioctl and did not
1727          * free any new commands.
1728          */
1729         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1730                 return;
1731
1732         /* We have room on the queue for more commands.  Now we need to queue
1733          * them up.  We will also keep track of the next queue to run so
1734          * that every queue gets a chance to be started first.
1735          */
1736         for (i = 0; i < h->highest_lun + 1; i++) {
1737                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1738                 /* make sure the disk has been added and the drive is real
1739                  * because this can be called from the middle of init_one.
1740                  */
1741                 if (!h->drv[curr_queue])
1742                         continue;
1743                 if (!(h->drv[curr_queue]->queue) ||
1744                         !(h->drv[curr_queue]->heads))
1745                         continue;
1746                 blk_start_queue(h->gendisk[curr_queue]->queue);
1747
1748                 /* check to see if we have maxed out the number of commands
1749                  * that can be placed on the queue.
1750                  */
1751                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1752                         if (curr_queue == start_queue) {
1753                                 h->next_to_run =
1754                                     (start_queue + 1) % (h->highest_lun + 1);
1755                                 break;
1756                         } else {
1757                                 h->next_to_run = curr_queue;
1758                                 break;
1759                         }
1760                 }
1761         }
1762 }
1763
1764 static void cciss_softirq_done(struct request *rq)
1765 {
1766         CommandList_struct *c = rq->completion_data;
1767         ctlr_info_t *h = hba[c->ctlr];
1768         SGDescriptor_struct *curr_sg = c->SG;
1769         u64bit temp64;
1770         unsigned long flags;
1771         int i, ddir;
1772         int sg_index = 0;
1773
1774         if (c->Request.Type.Direction == XFER_READ)
1775                 ddir = PCI_DMA_FROMDEVICE;
1776         else
1777                 ddir = PCI_DMA_TODEVICE;
1778
1779         /* command did not need to be retried */
1780         /* unmap the DMA mapping for all the scatter gather elements */
1781         for (i = 0; i < c->Header.SGList; i++) {
1782                 if (curr_sg[sg_index].Ext == CCISS_SG_CHAIN) {
1783                         cciss_unmap_sg_chain_block(h, c);
1784                         /* Point to the next block */
1785                         curr_sg = h->cmd_sg_list[c->cmdindex];
1786                         sg_index = 0;
1787                 }
1788                 temp64.val32.lower = curr_sg[sg_index].Addr.lower;
1789                 temp64.val32.upper = curr_sg[sg_index].Addr.upper;
1790                 pci_unmap_page(h->pdev, temp64.val, curr_sg[sg_index].Len,
1791                                 ddir);
1792                 ++sg_index;
1793         }
1794
1795         dev_dbg(&h->pdev->dev, "Done with %p\n", rq);
1796
1797         /* set the residual count for pc requests */
1798         if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
1799                 rq->resid_len = c->err_info->ResidualCnt;
1800
1801         blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1802
1803         spin_lock_irqsave(&h->lock, flags);
1804         cmd_free(h, c);
1805         cciss_check_queues(h);
1806         spin_unlock_irqrestore(&h->lock, flags);
1807 }
1808
1809 static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1810         unsigned char scsi3addr[], uint32_t log_unit)
1811 {
1812         memcpy(scsi3addr, h->drv[log_unit]->LunID,
1813                 sizeof(h->drv[log_unit]->LunID));
1814 }
1815
1816 /* This function gets the SCSI vendor, model, and revision of a logical drive
1817  * via the inquiry page 0.  Model, vendor, and rev are set to empty strings if
1818  * they cannot be read.
1819  */
1820 static void cciss_get_device_descr(ctlr_info_t *h, int logvol,
1821                                    char *vendor, char *model, char *rev)
1822 {
1823         int rc;
1824         InquiryData_struct *inq_buf;
1825         unsigned char scsi3addr[8];
1826
1827         *vendor = '\0';
1828         *model = '\0';
1829         *rev = '\0';
1830
1831         inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1832         if (!inq_buf)
1833                 return;
1834
1835         log_unit_to_scsi3addr(h, scsi3addr, logvol);
1836         rc = sendcmd_withirq(h, CISS_INQUIRY, inq_buf, sizeof(*inq_buf), 0,
1837                         scsi3addr, TYPE_CMD);
1838         if (rc == IO_OK) {
1839                 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1840                 vendor[VENDOR_LEN] = '\0';
1841                 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1842                 model[MODEL_LEN] = '\0';
1843                 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1844                 rev[REV_LEN] = '\0';
1845         }
1846
1847         kfree(inq_buf);
1848         return;
1849 }
1850
1851 /* This function gets the serial number of a logical drive via
1852  * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1853  * number cannot be had, for whatever reason, 16 bytes of 0xff
1854  * are returned instead.
1855  */
1856 static void cciss_get_serial_no(ctlr_info_t *h, int logvol,
1857                                 unsigned char *serial_no, int buflen)
1858 {
1859 #define PAGE_83_INQ_BYTES 64
1860         int rc;
1861         unsigned char *buf;
1862         unsigned char scsi3addr[8];
1863
1864         if (buflen > 16)
1865                 buflen = 16;
1866         memset(serial_no, 0xff, buflen);
1867         buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1868         if (!buf)
1869                 return;
1870         memset(serial_no, 0, buflen);
1871         log_unit_to_scsi3addr(h, scsi3addr, logvol);
1872         rc = sendcmd_withirq(h, CISS_INQUIRY, buf,
1873                 PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1874         if (rc == IO_OK)
1875                 memcpy(serial_no, &buf[8], buflen);
1876         kfree(buf);
1877         return;
1878 }
1879
1880 /*
1881  * cciss_add_disk sets up the block device queue for a logical drive
1882  */
1883 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1884                                 int drv_index)
1885 {
1886         disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1887         if (!disk->queue)
1888                 goto init_queue_failure;
1889         sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1890         disk->major = h->major;
1891         disk->first_minor = drv_index << NWD_SHIFT;
1892         disk->fops = &cciss_fops;
1893         if (cciss_create_ld_sysfs_entry(h, drv_index))
1894                 goto cleanup_queue;
1895         disk->private_data = h->drv[drv_index];
1896         disk->driverfs_dev = &h->drv[drv_index]->dev;
1897
1898         /* Set up queue information */
1899         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1900
1901         /* This is a hardware imposed limit. */
1902         blk_queue_max_segments(disk->queue, h->maxsgentries);
1903
1904         blk_queue_max_hw_sectors(disk->queue, h->cciss_max_sectors);
1905
1906         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1907
1908         disk->queue->queuedata = h;
1909
1910         blk_queue_logical_block_size(disk->queue,
1911                                      h->drv[drv_index]->block_size);
1912
1913         /* Make sure all queue data is written out before */
1914         /* setting h->drv[drv_index]->queue, as setting this */
1915         /* allows the interrupt handler to start the queue */
1916         wmb();
1917         h->drv[drv_index]->queue = disk->queue;
1918         add_disk(disk);
1919         return 0;
1920
1921 cleanup_queue:
1922         blk_cleanup_queue(disk->queue);
1923         disk->queue = NULL;
1924 init_queue_failure:
1925         return -1;
1926 }
1927
1928 /* This function will check the usage_count of the drive to be updated/added.
1929  * If the usage_count is zero and it is a heretofore unknown drive, or,
1930  * the drive's capacity, geometry, or serial number has changed,
1931  * then the drive information will be updated and the disk will be
1932  * re-registered with the kernel.  If these conditions don't hold,
1933  * then it will be left alone for the next reboot.  The exception to this
1934  * is disk 0 which will always be left registered with the kernel since it
1935  * is also the controller node.  Any changes to disk 0 will show up on
1936  * the next reboot.
1937  */
1938 static void cciss_update_drive_info(ctlr_info_t *h, int drv_index,
1939         int first_time, int via_ioctl)
1940 {
1941         struct gendisk *disk;
1942         InquiryData_struct *inq_buff = NULL;
1943         unsigned int block_size;
1944         sector_t total_size;
1945         unsigned long flags = 0;
1946         int ret = 0;
1947         drive_info_struct *drvinfo;
1948
1949         /* Get information about the disk and modify the driver structure */
1950         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1951         drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
1952         if (inq_buff == NULL || drvinfo == NULL)
1953                 goto mem_msg;
1954
1955         /* testing to see if 16-byte CDBs are already being used */
1956         if (h->cciss_read == CCISS_READ_16) {
1957                 cciss_read_capacity_16(h, drv_index,
1958                         &total_size, &block_size);
1959
1960         } else {
1961                 cciss_read_capacity(h, drv_index, &total_size, &block_size);
1962                 /* if read_capacity returns all F's this volume is >2TB */
1963                 /* in size so we switch to 16-byte CDB's for all */
1964                 /* read/write ops */
1965                 if (total_size == 0xFFFFFFFFULL) {
1966                         cciss_read_capacity_16(h, drv_index,
1967                         &total_size, &block_size);
1968                         h->cciss_read = CCISS_READ_16;
1969                         h->cciss_write = CCISS_WRITE_16;
1970                 } else {
1971                         h->cciss_read = CCISS_READ_10;
1972                         h->cciss_write = CCISS_WRITE_10;
1973                 }
1974         }
1975
1976         cciss_geometry_inquiry(h, drv_index, total_size, block_size,
1977                                inq_buff, drvinfo);
1978         drvinfo->block_size = block_size;
1979         drvinfo->nr_blocks = total_size + 1;
1980
1981         cciss_get_device_descr(h, drv_index, drvinfo->vendor,
1982                                 drvinfo->model, drvinfo->rev);
1983         cciss_get_serial_no(h, drv_index, drvinfo->serial_no,
1984                         sizeof(drvinfo->serial_no));
1985         /* Save the lunid in case we deregister the disk, below. */
1986         memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
1987                 sizeof(drvinfo->LunID));
1988
1989         /* Is it the same disk we already know, and nothing's changed? */
1990         if (h->drv[drv_index]->raid_level != -1 &&
1991                 ((memcmp(drvinfo->serial_no,
1992                                 h->drv[drv_index]->serial_no, 16) == 0) &&
1993                 drvinfo->block_size == h->drv[drv_index]->block_size &&
1994                 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
1995                 drvinfo->heads == h->drv[drv_index]->heads &&
1996                 drvinfo->sectors == h->drv[drv_index]->sectors &&
1997                 drvinfo->cylinders == h->drv[drv_index]->cylinders))
1998                         /* The disk is unchanged, nothing to update */
1999                         goto freeret;
2000
2001         /* If we get here it's not the same disk, or something's changed,
2002          * so we need to * deregister it, and re-register it, if it's not
2003          * in use.
2004          * If the disk already exists then deregister it before proceeding
2005          * (unless it's the first disk (for the controller node).
2006          */
2007         if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
2008                 dev_warn(&h->pdev->dev, "disk %d has changed.\n", drv_index);
2009                 spin_lock_irqsave(&h->lock, flags);
2010                 h->drv[drv_index]->busy_configuring = 1;
2011                 spin_unlock_irqrestore(&h->lock, flags);
2012
2013                 /* deregister_disk sets h->drv[drv_index]->queue = NULL
2014                  * which keeps the interrupt handler from starting
2015                  * the queue.
2016                  */
2017                 ret = deregister_disk(h, drv_index, 0, via_ioctl);
2018         }
2019
2020         /* If the disk is in use return */
2021         if (ret)
2022                 goto freeret;
2023
2024         /* Save the new information from cciss_geometry_inquiry
2025          * and serial number inquiry.  If the disk was deregistered
2026          * above, then h->drv[drv_index] will be NULL.
2027          */
2028         if (h->drv[drv_index] == NULL) {
2029                 drvinfo->device_initialized = 0;
2030                 h->drv[drv_index] = drvinfo;
2031                 drvinfo = NULL; /* so it won't be freed below. */
2032         } else {
2033                 /* special case for cxd0 */
2034                 h->drv[drv_index]->block_size = drvinfo->block_size;
2035                 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
2036                 h->drv[drv_index]->heads = drvinfo->heads;
2037                 h->drv[drv_index]->sectors = drvinfo->sectors;
2038                 h->drv[drv_index]->cylinders = drvinfo->cylinders;
2039                 h->drv[drv_index]->raid_level = drvinfo->raid_level;
2040                 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
2041                 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
2042                         VENDOR_LEN + 1);
2043                 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
2044                 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
2045         }
2046
2047         ++h->num_luns;
2048         disk = h->gendisk[drv_index];
2049         set_capacity(disk, h->drv[drv_index]->nr_blocks);
2050
2051         /* If it's not disk 0 (drv_index != 0)
2052          * or if it was disk 0, but there was previously
2053          * no actual corresponding configured logical drive
2054          * (raid_leve == -1) then we want to update the
2055          * logical drive's information.
2056          */
2057         if (drv_index || first_time) {
2058                 if (cciss_add_disk(h, disk, drv_index) != 0) {
2059                         cciss_free_gendisk(h, drv_index);
2060                         cciss_free_drive_info(h, drv_index);
2061                         dev_warn(&h->pdev->dev, "could not update disk %d\n",
2062                                 drv_index);
2063                         --h->num_luns;
2064                 }
2065         }
2066
2067 freeret:
2068         kfree(inq_buff);
2069         kfree(drvinfo);
2070         return;
2071 mem_msg:
2072         dev_err(&h->pdev->dev, "out of memory\n");
2073         goto freeret;
2074 }
2075
2076 /* This function will find the first index of the controllers drive array
2077  * that has a null drv pointer and allocate the drive info struct and
2078  * will return that index   This is where new drives will be added.
2079  * If the index to be returned is greater than the highest_lun index for
2080  * the controller then highest_lun is set * to this new index.
2081  * If there are no available indexes or if tha allocation fails, then -1
2082  * is returned.  * "controller_node" is used to know if this is a real
2083  * logical drive, or just the controller node, which determines if this
2084  * counts towards highest_lun.
2085  */
2086 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
2087 {
2088         int i;
2089         drive_info_struct *drv;
2090
2091         /* Search for an empty slot for our drive info */
2092         for (i = 0; i < CISS_MAX_LUN; i++) {
2093
2094                 /* if not cxd0 case, and it's occupied, skip it. */
2095                 if (h->drv[i] && i != 0)
2096                         continue;
2097                 /*
2098                  * If it's cxd0 case, and drv is alloc'ed already, and a
2099                  * disk is configured there, skip it.
2100                  */
2101                 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2102                         continue;
2103
2104                 /*
2105                  * We've found an empty slot.  Update highest_lun
2106                  * provided this isn't just the fake cxd0 controller node.
2107                  */
2108                 if (i > h->highest_lun && !controller_node)
2109                         h->highest_lun = i;
2110
2111                 /* If adding a real disk at cxd0, and it's already alloc'ed */
2112                 if (i == 0 && h->drv[i] != NULL)
2113                         return i;
2114
2115                 /*
2116                  * Found an empty slot, not already alloc'ed.  Allocate it.
2117                  * Mark it with raid_level == -1, so we know it's new later on.
2118                  */
2119                 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2120                 if (!drv)
2121                         return -1;
2122                 drv->raid_level = -1; /* so we know it's new */
2123                 h->drv[i] = drv;
2124                 return i;
2125         }
2126         return -1;
2127 }
2128
2129 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2130 {
2131         kfree(h->drv[drv_index]);
2132         h->drv[drv_index] = NULL;
2133 }
2134
2135 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2136 {
2137         put_disk(h->gendisk[drv_index]);
2138         h->gendisk[drv_index] = NULL;
2139 }
2140
2141 /* cciss_add_gendisk finds a free hba[]->drv structure
2142  * and allocates a gendisk if needed, and sets the lunid
2143  * in the drvinfo structure.   It returns the index into
2144  * the ->drv[] array, or -1 if none are free.
2145  * is_controller_node indicates whether highest_lun should
2146  * count this disk, or if it's only being added to provide
2147  * a means to talk to the controller in case no logical
2148  * drives have yet been configured.
2149  */
2150 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2151         int controller_node)
2152 {
2153         int drv_index;
2154
2155         drv_index = cciss_alloc_drive_info(h, controller_node);
2156         if (drv_index == -1)
2157                 return -1;
2158
2159         /*Check if the gendisk needs to be allocated */
2160         if (!h->gendisk[drv_index]) {
2161                 h->gendisk[drv_index] =
2162                         alloc_disk(1 << NWD_SHIFT);
2163                 if (!h->gendisk[drv_index]) {
2164                         dev_err(&h->pdev->dev,
2165                                 "could not allocate a new disk %d\n",
2166                                 drv_index);
2167                         goto err_free_drive_info;
2168                 }
2169         }
2170         memcpy(h->drv[drv_index]->LunID, lunid,
2171                 sizeof(h->drv[drv_index]->LunID));
2172         if (cciss_create_ld_sysfs_entry(h, drv_index))
2173                 goto err_free_disk;
2174         /* Don't need to mark this busy because nobody */
2175         /* else knows about this disk yet to contend */
2176         /* for access to it. */
2177         h->drv[drv_index]->busy_configuring = 0;
2178         wmb();
2179         return drv_index;
2180
2181 err_free_disk:
2182         cciss_free_gendisk(h, drv_index);
2183 err_free_drive_info:
2184         cciss_free_drive_info(h, drv_index);
2185         return -1;
2186 }
2187
2188 /* This is for the special case of a controller which
2189  * has no logical drives.  In this case, we still need
2190  * to register a disk so the controller can be accessed
2191  * by the Array Config Utility.
2192  */
2193 static void cciss_add_controller_node(ctlr_info_t *h)
2194 {
2195         struct gendisk *disk;
2196         int drv_index;
2197
2198         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2199                 return;
2200
2201         drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2202         if (drv_index == -1)
2203                 goto error;
2204         h->drv[drv_index]->block_size = 512;
2205         h->drv[drv_index]->nr_blocks = 0;
2206         h->drv[drv_index]->heads = 0;
2207         h->drv[drv_index]->sectors = 0;
2208         h->drv[drv_index]->cylinders = 0;
2209         h->drv[drv_index]->raid_level = -1;
2210         memset(h->drv[drv_index]->serial_no, 0, 16);
2211         disk = h->gendisk[drv_index];
2212         if (cciss_add_disk(h, disk, drv_index) == 0)
2213                 return;
2214         cciss_free_gendisk(h, drv_index);
2215         cciss_free_drive_info(h, drv_index);
2216 error:
2217         dev_warn(&h->pdev->dev, "could not add disk 0.\n");
2218         return;
2219 }
2220
2221 /* This function will add and remove logical drives from the Logical
2222  * drive array of the controller and maintain persistency of ordering
2223  * so that mount points are preserved until the next reboot.  This allows
2224  * for the removal of logical drives in the middle of the drive array
2225  * without a re-ordering of those drives.
2226  * INPUT
2227  * h            = The controller to perform the operations on
2228  */
2229 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2230         int via_ioctl)
2231 {
2232         int num_luns;
2233         ReportLunData_struct *ld_buff = NULL;
2234         int return_code;
2235         int listlength = 0;
2236         int i;
2237         int drv_found;
2238         int drv_index = 0;
2239         unsigned char lunid[8] = CTLR_LUNID;
2240         unsigned long flags;
2241
2242         if (!capable(CAP_SYS_RAWIO))
2243                 return -EPERM;
2244
2245         /* Set busy_configuring flag for this operation */
2246         spin_lock_irqsave(&h->lock, flags);
2247         if (h->busy_configuring) {
2248                 spin_unlock_irqrestore(&h->lock, flags);
2249                 return -EBUSY;
2250         }
2251         h->busy_configuring = 1;
2252         spin_unlock_irqrestore(&h->lock, flags);
2253
2254         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2255         if (ld_buff == NULL)
2256                 goto mem_msg;
2257
2258         return_code = sendcmd_withirq(h, CISS_REPORT_LOG, ld_buff,
2259                                       sizeof(ReportLunData_struct),
2260                                       0, CTLR_LUNID, TYPE_CMD);
2261
2262         if (return_code == IO_OK)
2263                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2264         else {  /* reading number of logical volumes failed */
2265                 dev_warn(&h->pdev->dev,
2266                         "report logical volume command failed\n");
2267                 listlength = 0;
2268                 goto freeret;
2269         }
2270
2271         num_luns = listlength / 8;      /* 8 bytes per entry */
2272         if (num_luns > CISS_MAX_LUN) {
2273                 num_luns = CISS_MAX_LUN;
2274                 dev_warn(&h->pdev->dev, "more luns configured"
2275                        " on controller than can be handled by"
2276                        " this driver.\n");
2277         }
2278
2279         if (num_luns == 0)
2280                 cciss_add_controller_node(h);
2281
2282         /* Compare controller drive array to driver's drive array
2283          * to see if any drives are missing on the controller due
2284          * to action of Array Config Utility (user deletes drive)
2285          * and deregister logical drives which have disappeared.
2286          */
2287         for (i = 0; i <= h->highest_lun; i++) {
2288                 int j;
2289                 drv_found = 0;
2290
2291                 /* skip holes in the array from already deleted drives */
2292                 if (h->drv[i] == NULL)
2293                         continue;
2294
2295                 for (j = 0; j < num_luns; j++) {
2296                         memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2297                         if (memcmp(h->drv[i]->LunID, lunid,
2298                                 sizeof(lunid)) == 0) {
2299                                 drv_found = 1;
2300                                 break;
2301                         }
2302                 }
2303                 if (!drv_found) {
2304                         /* Deregister it from the OS, it's gone. */
2305                         spin_lock_irqsave(&h->lock, flags);
2306                         h->drv[i]->busy_configuring = 1;
2307                         spin_unlock_irqrestore(&h->lock, flags);
2308                         return_code = deregister_disk(h, i, 1, via_ioctl);
2309                         if (h->drv[i] != NULL)
2310                                 h->drv[i]->busy_configuring = 0;
2311                 }
2312         }
2313
2314         /* Compare controller drive array to driver's drive array.
2315          * Check for updates in the drive information and any new drives
2316          * on the controller due to ACU adding logical drives, or changing
2317          * a logical drive's size, etc.  Reregister any new/changed drives
2318          */
2319         for (i = 0; i < num_luns; i++) {
2320                 int j;
2321
2322                 drv_found = 0;
2323
2324                 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2325                 /* Find if the LUN is already in the drive array
2326                  * of the driver.  If so then update its info
2327                  * if not in use.  If it does not exist then find
2328                  * the first free index and add it.
2329                  */
2330                 for (j = 0; j <= h->highest_lun; j++) {
2331                         if (h->drv[j] != NULL &&
2332                                 memcmp(h->drv[j]->LunID, lunid,
2333                                         sizeof(h->drv[j]->LunID)) == 0) {
2334                                 drv_index = j;
2335                                 drv_found = 1;
2336                                 break;
2337                         }
2338                 }
2339
2340                 /* check if the drive was found already in the array */
2341                 if (!drv_found) {
2342                         drv_index = cciss_add_gendisk(h, lunid, 0);
2343                         if (drv_index == -1)
2344                                 goto freeret;
2345                 }
2346                 cciss_update_drive_info(h, drv_index, first_time, via_ioctl);
2347         }               /* end for */
2348
2349 freeret:
2350         kfree(ld_buff);
2351         h->busy_configuring = 0;
2352         /* We return -1 here to tell the ACU that we have registered/updated
2353          * all of the drives that we can and to keep it from calling us
2354          * additional times.
2355          */
2356         return -1;
2357 mem_msg:
2358         dev_err(&h->pdev->dev, "out of memory\n");
2359         h->busy_configuring = 0;
2360         goto freeret;
2361 }
2362
2363 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2364 {
2365         /* zero out the disk size info */
2366         drive_info->nr_blocks = 0;
2367         drive_info->block_size = 0;
2368         drive_info->heads = 0;
2369         drive_info->sectors = 0;
2370         drive_info->cylinders = 0;
2371         drive_info->raid_level = -1;
2372         memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2373         memset(drive_info->model, 0, sizeof(drive_info->model));
2374         memset(drive_info->rev, 0, sizeof(drive_info->rev));
2375         memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2376         /*
2377          * don't clear the LUNID though, we need to remember which
2378          * one this one is.
2379          */
2380 }
2381
2382 /* This function will deregister the disk and it's queue from the
2383  * kernel.  It must be called with the controller lock held and the
2384  * drv structures busy_configuring flag set.  It's parameters are:
2385  *
2386  * disk = This is the disk to be deregistered
2387  * drv  = This is the drive_info_struct associated with the disk to be
2388  *        deregistered.  It contains information about the disk used
2389  *        by the driver.
2390  * clear_all = This flag determines whether or not the disk information
2391  *             is going to be completely cleared out and the highest_lun
2392  *             reset.  Sometimes we want to clear out information about
2393  *             the disk in preparation for re-adding it.  In this case
2394  *             the highest_lun should be left unchanged and the LunID
2395  *             should not be cleared.
2396  * via_ioctl
2397  *    This indicates whether we've reached this path via ioctl.
2398  *    This affects the maximum usage count allowed for c0d0 to be messed with.
2399  *    If this path is reached via ioctl(), then the max_usage_count will
2400  *    be 1, as the process calling ioctl() has got to have the device open.
2401  *    If we get here via sysfs, then the max usage count will be zero.
2402 */
2403 static int deregister_disk(ctlr_info_t *h, int drv_index,
2404                            int clear_all, int via_ioctl)
2405 {
2406         int i;
2407         struct gendisk *disk;
2408         drive_info_struct *drv;
2409         int recalculate_highest_lun;
2410
2411         if (!capable(CAP_SYS_RAWIO))
2412                 return -EPERM;
2413
2414         drv = h->drv[drv_index];
2415         disk = h->gendisk[drv_index];
2416
2417         /* make sure logical volume is NOT is use */
2418         if (clear_all || (h->gendisk[0] == disk)) {
2419                 if (drv->usage_count > via_ioctl)
2420                         return -EBUSY;
2421         } else if (drv->usage_count > 0)
2422                 return -EBUSY;
2423
2424         recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2425
2426         /* invalidate the devices and deregister the disk.  If it is disk
2427          * zero do not deregister it but just zero out it's values.  This
2428          * allows us to delete disk zero but keep the controller registered.
2429          */
2430         if (h->gendisk[0] != disk) {
2431                 struct request_queue *q = disk->queue;
2432                 if (disk->flags & GENHD_FL_UP) {
2433                         cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2434                         del_gendisk(disk);
2435                 }
2436                 if (q)
2437                         blk_cleanup_queue(q);
2438                 /* If clear_all is set then we are deleting the logical
2439                  * drive, not just refreshing its info.  For drives
2440                  * other than disk 0 we will call put_disk.  We do not
2441                  * do this for disk 0 as we need it to be able to
2442                  * configure the controller.
2443                  */
2444                 if (clear_all){
2445                         /* This isn't pretty, but we need to find the
2446                          * disk in our array and NULL our the pointer.
2447                          * This is so that we will call alloc_disk if
2448                          * this index is used again later.
2449                          */
2450                         for (i=0; i < CISS_MAX_LUN; i++){
2451                                 if (h->gendisk[i] == disk) {
2452                                         h->gendisk[i] = NULL;
2453                                         break;
2454                                 }
2455                         }
2456                         put_disk(disk);
2457                 }
2458         } else {
2459                 set_capacity(disk, 0);
2460                 cciss_clear_drive_info(drv);
2461         }
2462
2463         --h->num_luns;
2464
2465         /* if it was the last disk, find the new hightest lun */
2466         if (clear_all && recalculate_highest_lun) {
2467                 int newhighest = -1;
2468                 for (i = 0; i <= h->highest_lun; i++) {
2469                         /* if the disk has size > 0, it is available */
2470                         if (h->drv[i] && h->drv[i]->heads)
2471                                 newhighest = i;
2472                 }
2473                 h->highest_lun = newhighest;
2474         }
2475         return 0;
2476 }
2477
2478 static int fill_cmd(ctlr_info_t *h, CommandList_struct *c, __u8 cmd, void *buff,
2479                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2480                 int cmd_type)
2481 {
2482         u64bit buff_dma_handle;
2483         int status = IO_OK;
2484
2485         c->cmd_type = CMD_IOCTL_PEND;
2486         c->Header.ReplyQueue = 0;
2487         if (buff != NULL) {
2488                 c->Header.SGList = 1;
2489                 c->Header.SGTotal = 1;
2490         } else {
2491                 c->Header.SGList = 0;
2492                 c->Header.SGTotal = 0;
2493         }
2494         c->Header.Tag.lower = c->busaddr;
2495         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2496
2497         c->Request.Type.Type = cmd_type;
2498         if (cmd_type == TYPE_CMD) {
2499                 switch (cmd) {
2500                 case CISS_INQUIRY:
2501                         /* are we trying to read a vital product page */
2502                         if (page_code != 0) {
2503                                 c->Request.CDB[1] = 0x01;
2504                                 c->Request.CDB[2] = page_code;
2505                         }
2506                         c->Request.CDBLen = 6;
2507                         c->Request.Type.Attribute = ATTR_SIMPLE;
2508                         c->Request.Type.Direction = XFER_READ;
2509                         c->Request.Timeout = 0;
2510                         c->Request.CDB[0] = CISS_INQUIRY;
2511                         c->Request.CDB[4] = size & 0xFF;
2512                         break;
2513                 case CISS_REPORT_LOG:
2514                 case CISS_REPORT_PHYS:
2515                         /* Talking to controller so It's a physical command
2516                            mode = 00 target = 0.  Nothing to write.
2517                          */
2518                         c->Request.CDBLen = 12;
2519                         c->Request.Type.Attribute = ATTR_SIMPLE;
2520                         c->Request.Type.Direction = XFER_READ;
2521                         c->Request.Timeout = 0;
2522                         c->Request.CDB[0] = cmd;
2523                         c->Request.CDB[6] = (size >> 24) & 0xFF; /* MSB */
2524                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2525                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2526                         c->Request.CDB[9] = size & 0xFF;
2527                         break;
2528
2529                 case CCISS_READ_CAPACITY:
2530                         c->Request.CDBLen = 10;
2531                         c->Request.Type.Attribute = ATTR_SIMPLE;
2532                         c->Request.Type.Direction = XFER_READ;
2533                         c->Request.Timeout = 0;
2534                         c->Request.CDB[0] = cmd;
2535                         break;
2536                 case CCISS_READ_CAPACITY_16:
2537                         c->Request.CDBLen = 16;
2538                         c->Request.Type.Attribute = ATTR_SIMPLE;
2539                         c->Request.Type.Direction = XFER_READ;
2540                         c->Request.Timeout = 0;
2541                         c->Request.CDB[0] = cmd;
2542                         c->Request.CDB[1] = 0x10;
2543                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2544                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2545                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2546                         c->Request.CDB[13] = size & 0xFF;
2547                         c->Request.Timeout = 0;
2548                         c->Request.CDB[0] = cmd;
2549                         break;
2550                 case CCISS_CACHE_FLUSH:
2551                         c->Request.CDBLen = 12;
2552                         c->Request.Type.Attribute = ATTR_SIMPLE;
2553                         c->Request.Type.Direction = XFER_WRITE;
2554                         c->Request.Timeout = 0;
2555                         c->Request.CDB[0] = BMIC_WRITE;
2556                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2557                         break;
2558                 case TEST_UNIT_READY:
2559                         c->Request.CDBLen = 6;
2560                         c->Request.Type.Attribute = ATTR_SIMPLE;
2561                         c->Request.Type.Direction = XFER_NONE;
2562                         c->Request.Timeout = 0;
2563                         break;
2564                 default:
2565                         dev_warn(&h->pdev->dev, "Unknown Command 0x%c\n", cmd);
2566                         return IO_ERROR;
2567                 }
2568         } else if (cmd_type == TYPE_MSG) {
2569                 switch (cmd) {
2570                 case 0: /* ABORT message */
2571                         c->Request.CDBLen = 12;
2572                         c->Request.Type.Attribute = ATTR_SIMPLE;
2573                         c->Request.Type.Direction = XFER_WRITE;
2574                         c->Request.Timeout = 0;
2575                         c->Request.CDB[0] = cmd;        /* abort */
2576                         c->Request.CDB[1] = 0;  /* abort a command */
2577                         /* buff contains the tag of the command to abort */
2578                         memcpy(&c->Request.CDB[4], buff, 8);
2579                         break;
2580                 case 1: /* RESET message */
2581                         c->Request.CDBLen = 16;
2582                         c->Request.Type.Attribute = ATTR_SIMPLE;
2583                         c->Request.Type.Direction = XFER_NONE;
2584                         c->Request.Timeout = 0;
2585                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2586                         c->Request.CDB[0] = cmd;        /* reset */
2587                         c->Request.CDB[1] = 0x03;       /* reset a target */
2588                         break;
2589                 case 3: /* No-Op message */
2590                         c->Request.CDBLen = 1;
2591                         c->Request.Type.Attribute = ATTR_SIMPLE;
2592                         c->Request.Type.Direction = XFER_WRITE;
2593                         c->Request.Timeout = 0;
2594                         c->Request.CDB[0] = cmd;
2595                         break;
2596                 default:
2597                         dev_warn(&h->pdev->dev,
2598                                 "unknown message type %d\n", cmd);
2599                         return IO_ERROR;
2600                 }
2601         } else {
2602                 dev_warn(&h->pdev->dev, "unknown command type %d\n", cmd_type);
2603                 return IO_ERROR;
2604         }
2605         /* Fill in the scatter gather information */
2606         if (size > 0) {
2607                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2608                                                              buff, size,
2609                                                              PCI_DMA_BIDIRECTIONAL);
2610                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2611                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2612                 c->SG[0].Len = size;
2613                 c->SG[0].Ext = 0;       /* we are not chaining */
2614         }
2615         return status;
2616 }
2617
2618 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2619 {
2620         switch (c->err_info->ScsiStatus) {
2621         case SAM_STAT_GOOD:
2622                 return IO_OK;
2623         case SAM_STAT_CHECK_CONDITION:
2624                 switch (0xf & c->err_info->SenseInfo[2]) {
2625                 case 0: return IO_OK; /* no sense */
2626                 case 1: return IO_OK; /* recovered error */
2627                 default:
2628                         if (check_for_unit_attention(h, c))
2629                                 return IO_NEEDS_RETRY;
2630                         dev_warn(&h->pdev->dev, "cmd 0x%02x "
2631                                 "check condition, sense key = 0x%02x\n",
2632                                 c->Request.CDB[0], c->err_info->SenseInfo[2]);
2633                 }
2634                 break;
2635         default:
2636                 dev_warn(&h->pdev->dev, "cmd 0x%02x"
2637                         "scsi status = 0x%02x\n",
2638                         c->Request.CDB[0], c->err_info->ScsiStatus);
2639                 break;
2640         }
2641         return IO_ERROR;
2642 }
2643
2644 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2645 {
2646         int return_status = IO_OK;
2647
2648         if (c->err_info->CommandStatus == CMD_SUCCESS)
2649                 return IO_OK;
2650
2651         switch (c->err_info->CommandStatus) {
2652         case CMD_TARGET_STATUS:
2653                 return_status = check_target_status(h, c);
2654                 break;
2655         case CMD_DATA_UNDERRUN:
2656         case CMD_DATA_OVERRUN:
2657                 /* expected for inquiry and report lun commands */
2658                 break;
2659         case CMD_INVALID:
2660                 dev_warn(&h->pdev->dev, "cmd 0x%02x is "
2661                        "reported invalid\n", c->Request.CDB[0]);
2662                 return_status = IO_ERROR;
2663                 break;
2664         case CMD_PROTOCOL_ERR:
2665                 dev_warn(&h->pdev->dev, "cmd 0x%02x has "
2666                        "protocol error\n", c->Request.CDB[0]);
2667                 return_status = IO_ERROR;
2668                 break;
2669         case CMD_HARDWARE_ERR:
2670                 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2671                        " hardware error\n", c->Request.CDB[0]);
2672                 return_status = IO_ERROR;
2673                 break;
2674         case CMD_CONNECTION_LOST:
2675                 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2676                        "connection lost\n", c->Request.CDB[0]);
2677                 return_status = IO_ERROR;
2678                 break;
2679         case CMD_ABORTED:
2680                 dev_warn(&h->pdev->dev, "cmd 0x%02x was "
2681                        "aborted\n", c->Request.CDB[0]);
2682                 return_status = IO_ERROR;
2683                 break;
2684         case CMD_ABORT_FAILED:
2685                 dev_warn(&h->pdev->dev, "cmd 0x%02x reports "
2686                        "abort failed\n", c->Request.CDB[0]);
2687                 return_status = IO_ERROR;
2688                 break;
2689         case CMD_UNSOLICITED_ABORT:
2690                 dev_warn(&h->pdev->dev, "unsolicited abort 0x%02x\n",
2691                         c->Request.CDB[0]);
2692                 return_status = IO_NEEDS_RETRY;
2693                 break;
2694         case CMD_UNABORTABLE:
2695                 dev_warn(&h->pdev->dev, "cmd unabortable\n");
2696                 return_status = IO_ERROR;
2697                 break;
2698         default:
2699                 dev_warn(&h->pdev->dev, "cmd 0x%02x returned "
2700                        "unknown status %x\n", c->Request.CDB[0],
2701                        c->err_info->CommandStatus);
2702                 return_status = IO_ERROR;
2703         }
2704         return return_status;
2705 }
2706
2707 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2708         int attempt_retry)
2709 {
2710         DECLARE_COMPLETION_ONSTACK(wait);
2711         u64bit buff_dma_handle;
2712         int return_status = IO_OK;
2713
2714 resend_cmd2:
2715         c->waiting = &wait;
2716         enqueue_cmd_and_start_io(h, c);
2717
2718         wait_for_completion(&wait);
2719
2720         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2721                 goto command_done;
2722
2723         return_status = process_sendcmd_error(h, c);
2724
2725         if (return_status == IO_NEEDS_RETRY &&
2726                 c->retry_count < MAX_CMD_RETRIES) {
2727                 dev_warn(&h->pdev->dev, "retrying 0x%02x\n",
2728                         c->Request.CDB[0]);
2729                 c->retry_count++;
2730                 /* erase the old error information */
2731                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2732                 return_status = IO_OK;
2733                 INIT_COMPLETION(wait);
2734                 goto resend_cmd2;
2735         }
2736
2737 command_done:
2738         /* unlock the buffers from DMA */
2739         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2740         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2741         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2742                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2743         return return_status;
2744 }
2745
2746 static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
2747                            __u8 page_code, unsigned char scsi3addr[],
2748                         int cmd_type)
2749 {
2750         CommandList_struct *c;
2751         int return_status;
2752
2753         c = cmd_special_alloc(h);
2754         if (!c)
2755                 return -ENOMEM;
2756         return_status = fill_cmd(h, c, cmd, buff, size, page_code,
2757                 scsi3addr, cmd_type);
2758         if (return_status == IO_OK)
2759                 return_status = sendcmd_withirq_core(h, c, 1);
2760
2761         cmd_special_free(h, c);
2762         return return_status;
2763 }
2764
2765 static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
2766                                    sector_t total_size,
2767                                    unsigned int block_size,
2768                                    InquiryData_struct *inq_buff,
2769                                    drive_info_struct *drv)
2770 {
2771         int return_code;
2772         unsigned long t;
2773         unsigned char scsi3addr[8];
2774
2775         memset(inq_buff, 0, sizeof(InquiryData_struct));
2776         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2777         return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
2778                         sizeof(*inq_buff), 0xC1, scsi3addr, TYPE_CMD);
2779         if (return_code == IO_OK) {
2780                 if (inq_buff->data_byte[8] == 0xFF) {
2781                         dev_warn(&h->pdev->dev,
2782                                "reading geometry failed, volume "
2783                                "does not support reading geometry\n");
2784                         drv->heads = 255;
2785                         drv->sectors = 32;      /* Sectors per track */
2786                         drv->cylinders = total_size + 1;
2787                         drv->raid_level = RAID_UNKNOWN;
2788                 } else {
2789                         drv->heads = inq_buff->data_byte[6];
2790                         drv->sectors = inq_buff->data_byte[7];
2791                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2792                         drv->cylinders += inq_buff->data_byte[5];
2793                         drv->raid_level = inq_buff->data_byte[8];
2794                 }
2795                 drv->block_size = block_size;
2796                 drv->nr_blocks = total_size + 1;
2797                 t = drv->heads * drv->sectors;
2798                 if (t > 1) {
2799                         sector_t real_size = total_size + 1;
2800                         unsigned long rem = sector_div(real_size, t);
2801                         if (rem)
2802                                 real_size++;
2803                         drv->cylinders = real_size;
2804                 }
2805         } else {                /* Get geometry failed */
2806                 dev_warn(&h->pdev->dev, "reading geometry failed\n");
2807         }
2808 }
2809
2810 static void
2811 cciss_read_capacity(ctlr_info_t *h, int logvol, sector_t *total_size,
2812                     unsigned int *block_size)
2813 {
2814         ReadCapdata_struct *buf;
2815         int return_code;
2816         unsigned char scsi3addr[8];
2817
2818         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2819         if (!buf) {
2820                 dev_warn(&h->pdev->dev, "out of memory\n");
2821                 return;
2822         }
2823
2824         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2825         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY, buf,
2826                 sizeof(ReadCapdata_struct), 0, scsi3addr, TYPE_CMD);
2827         if (return_code == IO_OK) {
2828                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2829                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2830         } else {                /* read capacity command failed */
2831                 dev_warn(&h->pdev->dev, "read capacity failed\n");
2832                 *total_size = 0;
2833                 *block_size = BLOCK_SIZE;
2834         }
2835         kfree(buf);
2836 }
2837
2838 static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
2839         sector_t *total_size, unsigned int *block_size)
2840 {
2841         ReadCapdata_struct_16 *buf;
2842         int return_code;
2843         unsigned char scsi3addr[8];
2844
2845         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2846         if (!buf) {
2847                 dev_warn(&h->pdev->dev, "out of memory\n");
2848                 return;
2849         }
2850
2851         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2852         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY_16,
2853                 buf, sizeof(ReadCapdata_struct_16),
2854                         0, scsi3addr, TYPE_CMD);
2855         if (return_code == IO_OK) {
2856                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2857                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2858         } else {                /* read capacity command failed */
2859                 dev_warn(&h->pdev->dev, "read capacity failed\n");
2860                 *total_size = 0;
2861                 *block_size = BLOCK_SIZE;
2862         }
2863         dev_info(&h->pdev->dev, "      blocks= %llu block_size= %d\n",
2864                (unsigned long long)*total_size+1, *block_size);
2865         kfree(buf);
2866 }
2867
2868 static int cciss_revalidate(struct gendisk *disk)
2869 {
2870         ctlr_info_t *h = get_host(disk);
2871         drive_info_struct *drv = get_drv(disk);
2872         int logvol;
2873         int FOUND = 0;
2874         unsigned int block_size;
2875         sector_t total_size;
2876         InquiryData_struct *inq_buff = NULL;
2877
2878         for (logvol = 0; logvol <= h->highest_lun; logvol++) {
2879                 if (!h->drv[logvol])
2880                         continue;
2881                 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2882                         sizeof(drv->LunID)) == 0) {
2883                         FOUND = 1;
2884                         break;
2885                 }
2886         }
2887
2888         if (!FOUND)
2889                 return 1;
2890
2891         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2892         if (inq_buff == NULL) {
2893                 dev_warn(&h->pdev->dev, "out of memory\n");
2894                 return 1;
2895         }
2896         if (h->cciss_read == CCISS_READ_10) {
2897                 cciss_read_capacity(h, logvol,
2898                                         &total_size, &block_size);
2899         } else {
2900                 cciss_read_capacity_16(h, logvol,
2901                                         &total_size, &block_size);
2902         }
2903         cciss_geometry_inquiry(h, logvol, total_size, block_size,
2904                                inq_buff, drv);
2905
2906         blk_queue_logical_block_size(drv->queue, drv->block_size);
2907         set_capacity(disk, drv->nr_blocks);
2908
2909         kfree(inq_buff);
2910         return 0;
2911 }
2912
2913 /*
2914  * Map (physical) PCI mem into (virtual) kernel space
2915  */
2916 static void __iomem *remap_pci_mem(ulong base, ulong size)
2917 {
2918         ulong page_base = ((ulong) base) & PAGE_MASK;
2919         ulong page_offs = ((ulong) base) - page_base;
2920         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2921
2922         return page_remapped ? (page_remapped + page_offs) : NULL;
2923 }
2924
2925 /*
2926  * Takes jobs of the Q and sends them to the hardware, then puts it on
2927  * the Q to wait for completion.
2928  */
2929 static void start_io(ctlr_info_t *h)
2930 {
2931         CommandList_struct *c;
2932
2933         while (!list_empty(&h->reqQ)) {
2934                 c = list_entry(h->reqQ.next, CommandList_struct, list);
2935                 /* can't do anything if fifo is full */
2936                 if ((h->access.fifo_full(h))) {
2937                         dev_warn(&h->pdev->dev, "fifo full\n");
2938                         break;
2939                 }
2940
2941                 /* Get the first entry from the Request Q */
2942                 removeQ(c);
2943                 h->Qdepth--;
2944
2945                 /* Tell the controller execute command */
2946                 h->access.submit_command(h, c);
2947
2948                 /* Put job onto the completed Q */
2949                 addQ(&h->cmpQ, c);
2950         }
2951 }
2952
2953 /* Assumes that h->lock is held. */
2954 /* Zeros out the error record and then resends the command back */
2955 /* to the controller */
2956 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2957 {
2958         /* erase the old error information */
2959         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2960
2961         /* add it to software queue and then send it to the controller */
2962         addQ(&h->reqQ, c);
2963         h->Qdepth++;
2964         if (h->Qdepth > h->maxQsinceinit)
2965                 h->maxQsinceinit = h->Qdepth;
2966
2967         start_io(h);
2968 }
2969
2970 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
2971         unsigned int msg_byte, unsigned int host_byte,
2972         unsigned int driver_byte)
2973 {
2974         /* inverse of macros in scsi.h */
2975         return (scsi_status_byte & 0xff) |
2976                 ((msg_byte & 0xff) << 8) |
2977                 ((host_byte & 0xff) << 16) |
2978                 ((driver_byte & 0xff) << 24);
2979 }
2980
2981 static inline int evaluate_target_status(ctlr_info_t *h,
2982                         CommandList_struct *cmd, int *retry_cmd)
2983 {
2984         unsigned char sense_key;
2985         unsigned char status_byte, msg_byte, host_byte, driver_byte;
2986         int error_value;
2987
2988         *retry_cmd = 0;
2989         /* If we get in here, it means we got "target status", that is, scsi status */
2990         status_byte = cmd->err_info->ScsiStatus;
2991         driver_byte = DRIVER_OK;
2992         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
2993
2994         if (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC)
2995                 host_byte = DID_PASSTHROUGH;
2996         else
2997                 host_byte = DID_OK;
2998
2999         error_value = make_status_bytes(status_byte, msg_byte,
3000                 host_byte, driver_byte);
3001
3002         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
3003                 if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC)
3004                         dev_warn(&h->pdev->dev, "cmd %p "
3005                                "has SCSI Status 0x%x\n",
3006                                cmd, cmd->err_info->ScsiStatus);
3007                 return error_value;
3008         }
3009
3010         /* check the sense key */
3011         sense_key = 0xf & cmd->err_info->SenseInfo[2];
3012         /* no status or recovered error */
3013         if (((sense_key == 0x0) || (sense_key == 0x1)) &&
3014             (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC))
3015                 error_value = 0;
3016
3017         if (check_for_unit_attention(h, cmd)) {
3018                 *retry_cmd = !(cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC);
3019                 return 0;
3020         }
3021
3022         /* Not SG_IO or similar? */
3023         if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC) {
3024                 if (error_value != 0)
3025                         dev_warn(&h->pdev->dev, "cmd %p has CHECK CONDITION"
3026                                " sense key = 0x%x\n", cmd, sense_key);
3027                 return error_value;
3028         }
3029
3030         /* SG_IO or similar, copy sense data back */
3031         if (cmd->rq->sense) {
3032                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
3033                         cmd->rq->sense_len = cmd->err_info->SenseLen;
3034                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
3035                         cmd->rq->sense_len);
3036         } else
3037                 cmd->rq->sense_len = 0;
3038
3039         return error_value;
3040 }
3041
3042 /* checks the status of the job and calls complete buffers to mark all
3043  * buffers for the completed job. Note that this function does not need
3044  * to hold the hba/queue lock.
3045  */
3046 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
3047                                     int timeout)
3048 {
3049         int retry_cmd = 0;
3050         struct request *rq = cmd->rq;
3051
3052         rq->errors = 0;
3053
3054         if (timeout)
3055                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
3056
3057         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
3058                 goto after_error_processing;
3059
3060         switch (cmd->err_info->CommandStatus) {
3061         case CMD_TARGET_STATUS:
3062                 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
3063                 break;
3064         case CMD_DATA_UNDERRUN:
3065                 if (cmd->rq->cmd_type == REQ_TYPE_FS) {
3066                         dev_warn(&h->pdev->dev, "cmd %p has"
3067                                " completed with data underrun "
3068                                "reported\n", cmd);
3069                         cmd->rq->resid_len = cmd->err_info->ResidualCnt;
3070                 }
3071                 break;
3072         case CMD_DATA_OVERRUN:
3073                 if (cmd->rq->cmd_type == REQ_TYPE_FS)
3074                         dev_warn(&h->pdev->dev, "cciss: cmd %p has"
3075                                " completed with data overrun "
3076                                "reported\n", cmd);
3077                 break;
3078         case CMD_INVALID:
3079                 dev_warn(&h->pdev->dev, "cciss: cmd %p is "
3080                        "reported invalid\n", cmd);
3081                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3082                         cmd->err_info->CommandStatus, DRIVER_OK,
3083                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3084                                 DID_PASSTHROUGH : DID_ERROR);
3085                 break;
3086         case CMD_PROTOCOL_ERR:
3087                 dev_warn(&h->pdev->dev, "cciss: cmd %p has "
3088                        "protocol error\n", cmd);
3089                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3090                         cmd->err_info->CommandStatus, DRIVER_OK,
3091                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3092                                 DID_PASSTHROUGH : DID_ERROR);
3093                 break;
3094         case CMD_HARDWARE_ERR:
3095                 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3096                        " hardware error\n", cmd);
3097                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3098                         cmd->err_info->CommandStatus, DRIVER_OK,
3099                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3100                                 DID_PASSTHROUGH : DID_ERROR);
3101                 break;
3102         case CMD_CONNECTION_LOST:
3103                 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3104                        "connection lost\n", cmd);
3105                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3106                         cmd->err_info->CommandStatus, DRIVER_OK,
3107                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3108                                 DID_PASSTHROUGH : DID_ERROR);
3109                 break;
3110         case CMD_ABORTED:
3111                 dev_warn(&h->pdev->dev, "cciss: cmd %p was "
3112                        "aborted\n", cmd);
3113                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3114                         cmd->err_info->CommandStatus, DRIVER_OK,
3115                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3116                                 DID_PASSTHROUGH : DID_ABORT);
3117                 break;
3118         case CMD_ABORT_FAILED:
3119                 dev_warn(&h->pdev->dev, "cciss: cmd %p reports "
3120                        "abort failed\n", cmd);
3121                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3122                         cmd->err_info->CommandStatus, DRIVER_OK,
3123                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3124                                 DID_PASSTHROUGH : DID_ERROR);
3125                 break;
3126         case CMD_UNSOLICITED_ABORT:
3127                 dev_warn(&h->pdev->dev, "cciss%d: unsolicited "
3128                        "abort %p\n", h->ctlr, cmd);
3129                 if (cmd->retry_count < MAX_CMD_RETRIES) {
3130                         retry_cmd = 1;
3131                         dev_warn(&h->pdev->dev, "retrying %p\n", cmd);
3132                         cmd->retry_count++;
3133                 } else
3134                         dev_warn(&h->pdev->dev,
3135                                 "%p retried too many times\n", cmd);
3136                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3137                         cmd->err_info->CommandStatus, DRIVER_OK,
3138                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3139                                 DID_PASSTHROUGH : DID_ABORT);
3140                 break;
3141         case CMD_TIMEOUT:
3142                 dev_warn(&h->pdev->dev, "cmd %p timedout\n", cmd);
3143                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3144                         cmd->err_info->CommandStatus, DRIVER_OK,
3145                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3146                                 DID_PASSTHROUGH : DID_ERROR);
3147                 break;
3148         case CMD_UNABORTABLE:
3149                 dev_warn(&h->pdev->dev, "cmd %p unabortable\n", cmd);
3150                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3151                         cmd->err_info->CommandStatus, DRIVER_OK,
3152                         cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC ?
3153                                 DID_PASSTHROUGH : DID_ERROR);
3154                 break;
3155         default:
3156                 dev_warn(&h->pdev->dev, "cmd %p returned "
3157                        "unknown status %x\n", cmd,
3158                        cmd->err_info->CommandStatus);
3159                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3160                         cmd->err_info->CommandStatus, DRIVER_OK,
3161                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3162                                 DID_PASSTHROUGH : DID_ERROR);
3163         }
3164
3165 after_error_processing:
3166
3167         /* We need to return this command */
3168         if (retry_cmd) {
3169                 resend_cciss_cmd(h, cmd);
3170                 return;
3171         }
3172         cmd->rq->completion_data = cmd;
3173         blk_complete_request(cmd->rq);
3174 }
3175
3176 static inline u32 cciss_tag_contains_index(u32 tag)
3177 {
3178 #define DIRECT_LOOKUP_BIT 0x10
3179         return tag & DIRECT_LOOKUP_BIT;
3180 }
3181
3182 static inline u32 cciss_tag_to_index(u32 tag)
3183 {
3184 #define DIRECT_LOOKUP_SHIFT 5
3185         return tag >> DIRECT_LOOKUP_SHIFT;
3186 }
3187
3188 static inline u32 cciss_tag_discard_error_bits(ctlr_info_t *h, u32 tag)
3189 {
3190 #define CCISS_PERF_ERROR_BITS ((1 << DIRECT_LOOKUP_SHIFT) - 1)
3191 #define CCISS_SIMPLE_ERROR_BITS 0x03
3192         if (likely(h->transMethod & CFGTBL_Trans_Performant))
3193                 return tag & ~CCISS_PERF_ERROR_BITS;
3194         return tag & ~CCISS_SIMPLE_ERROR_BITS;
3195 }
3196
3197 static inline void cciss_mark_tag_indexed(u32 *tag)
3198 {
3199         *tag |= DIRECT_LOOKUP_BIT;
3200 }
3201
3202 static inline void cciss_set_tag_index(u32 *tag, u32 index)
3203 {
3204         *tag |= (index << DIRECT_LOOKUP_SHIFT);
3205 }
3206
3207 /*
3208  * Get a request and submit it to the controller.
3209  */
3210 static void do_cciss_request(struct request_queue *q)
3211 {
3212         ctlr_info_t *h = q->queuedata;
3213         CommandList_struct *c;
3214         sector_t start_blk;
3215         int seg;
3216         struct request *creq;
3217         u64bit temp64;
3218         struct scatterlist *tmp_sg;
3219         SGDescriptor_struct *curr_sg;
3220         drive_info_struct *drv;
3221         int i, dir;
3222         int sg_index = 0;
3223         int chained = 0;
3224
3225       queue:
3226         creq = blk_peek_request(q);
3227         if (!creq)
3228                 goto startio;
3229
3230         BUG_ON(creq->nr_phys_segments > h->maxsgentries);
3231
3232         c = cmd_alloc(h);
3233         if (!c)
3234                 goto full;
3235
3236         blk_start_request(creq);
3237
3238         tmp_sg = h->scatter_list[c->cmdindex];
3239         spin_unlock_irq(q->queue_lock);
3240
3241         c->cmd_type = CMD_RWREQ;
3242         c->rq = creq;
3243
3244         /* fill in the request */
3245         drv = creq->rq_disk->private_data;
3246         c->Header.ReplyQueue = 0;       /* unused in simple mode */
3247         /* got command from pool, so use the command block index instead */
3248         /* for direct lookups. */
3249         /* The first 2 bits are reserved for controller error reporting. */
3250         cciss_set_tag_index(&c->Header.Tag.lower, c->cmdindex);
3251         cciss_mark_tag_indexed(&c->Header.Tag.lower);
3252         memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3253         c->Request.CDBLen = 10; /* 12 byte commands not in FW yet; */
3254         c->Request.Type.Type = TYPE_CMD;        /* It is a command. */
3255         c->Request.Type.Attribute = ATTR_SIMPLE;
3256         c->Request.Type.Direction =
3257             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3258         c->Request.Timeout = 0; /* Don't time out */
3259         c->Request.CDB[0] =
3260             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3261         start_blk = blk_rq_pos(creq);
3262         dev_dbg(&h->pdev->dev, "sector =%d nr_sectors=%d\n",
3263                (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3264         sg_init_table(tmp_sg, h->maxsgentries);
3265         seg = blk_rq_map_sg(q, creq, tmp_sg);
3266
3267         /* get the DMA records for the setup */
3268         if (c->Request.Type.Direction == XFER_READ)
3269                 dir = PCI_DMA_FROMDEVICE;
3270         else
3271                 dir = PCI_DMA_TODEVICE;
3272
3273         curr_sg = c->SG;
3274         sg_index = 0;
3275         chained = 0;
3276
3277         for (i = 0; i < seg; i++) {
3278                 if (((sg_index+1) == (h->max_cmd_sgentries)) &&
3279                         !chained && ((seg - i) > 1)) {
3280                         /* Point to next chain block. */
3281                         curr_sg = h->cmd_sg_list[c->cmdindex];
3282                         sg_index = 0;
3283                         chained = 1;
3284                 }
3285                 curr_sg[sg_index].Len = tmp_sg[i].length;
3286                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3287                                                 tmp_sg[i].offset,
3288                                                 tmp_sg[i].length, dir);
3289                 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3290                 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3291                 curr_sg[sg_index].Ext = 0;  /* we are not chaining */
3292                 ++sg_index;
3293         }
3294         if (chained)
3295                 cciss_map_sg_chain_block(h, c, h->cmd_sg_list[c->cmdindex],
3296                         (seg - (h->max_cmd_sgentries - 1)) *
3297                                 sizeof(SGDescriptor_struct));
3298
3299         /* track how many SG entries we are using */
3300         if (seg > h->maxSG)
3301                 h->maxSG = seg;
3302
3303         dev_dbg(&h->pdev->dev, "Submitting %u sectors in %d segments "
3304                         "chained[%d]\n",
3305                         blk_rq_sectors(creq), seg, chained);
3306
3307         c->Header.SGTotal = seg + chained;
3308         if (seg <= h->max_cmd_sgentries)
3309                 c->Header.SGList = c->Header.SGTotal;
3310         else
3311                 c->Header.SGList = h->max_cmd_sgentries;
3312         set_performant_mode(h, c);
3313
3314         if (likely(creq->cmd_type == REQ_TYPE_FS)) {
3315                 if(h->cciss_read == CCISS_READ_10) {
3316                         c->Request.CDB[1] = 0;
3317                         c->Request.CDB[2] = (start_blk >> 24) & 0xff; /* MSB */
3318                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3319                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3320                         c->Request.CDB[5] = start_blk & 0xff;
3321                         c->Request.CDB[6] = 0; /* (sect >> 24) & 0xff; MSB */
3322                         c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3323                         c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3324                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3325                 } else {
3326                         u32 upper32 = upper_32_bits(start_blk);
3327
3328                         c->Request.CDBLen = 16;
3329                         c->Request.CDB[1]= 0;
3330                         c->Request.CDB[2]= (upper32 >> 24) & 0xff; /* MSB */
3331                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3332                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3333                         c->Request.CDB[5]= upper32 & 0xff;
3334                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3335                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3336                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3337                         c->Request.CDB[9]= start_blk & 0xff;
3338                         c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3339                         c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3340                         c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3341                         c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3342                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
3343                 }
3344         } else if (creq->cmd_type == REQ_TYPE_BLOCK_PC) {
3345                 c->Request.CDBLen = creq->cmd_len;
3346                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3347         } else {
3348                 dev_warn(&h->pdev->dev, "bad request type %d\n",
3349                         creq->cmd_type);
3350                 BUG();
3351         }
3352
3353         spin_lock_irq(q->queue_lock);
3354
3355         addQ(&h->reqQ, c);
3356         h->Qdepth++;
3357         if (h->Qdepth > h->maxQsinceinit)
3358                 h->maxQsinceinit = h->Qdepth;
3359
3360         goto queue;
3361 full:
3362         blk_stop_queue(q);
3363 startio:
3364         /* We will already have the driver lock here so not need
3365          * to lock it.
3366          */
3367         start_io(h);
3368 }
3369
3370 static inline unsigned long get_next_completion(ctlr_info_t *h)
3371 {
3372         return h->access.command_completed(h);
3373 }
3374
3375 static inline int interrupt_pending(ctlr_info_t *h)
3376 {
3377         return h->access.intr_pending(h);
3378 }
3379
3380 static inline long interrupt_not_for_us(ctlr_info_t *h)
3381 {
3382         return ((h->access.intr_pending(h) == 0) ||
3383                 (h->interrupts_enabled == 0));
3384 }
3385
3386 static inline int bad_tag(ctlr_info_t *h, u32 tag_index,
3387                         u32 raw_tag)
3388 {
3389         if (unlikely(tag_index >= h->nr_cmds)) {
3390                 dev_warn(&h->pdev->dev, "bad tag 0x%08x ignored.\n", raw_tag);
3391                 return 1;
3392         }
3393         return 0;
3394 }
3395
3396 static inline void finish_cmd(ctlr_info_t *h, CommandList_struct *c,
3397                                 u32 raw_tag)
3398 {
3399         removeQ(c);
3400         if (likely(c->cmd_type == CMD_RWREQ))
3401                 complete_command(h, c, 0);
3402         else if (c->cmd_type == CMD_IOCTL_PEND)
3403                 complete(c->waiting);
3404 #ifdef CONFIG_CISS_SCSI_TAPE
3405         else if (c->cmd_type == CMD_SCSI)
3406                 complete_scsi_command(c, 0, raw_tag);
3407 #endif
3408 }
3409
3410 static inline u32 next_command(ctlr_info_t *h)
3411 {
3412         u32 a;
3413
3414         if (unlikely(!(h->transMethod & CFGTBL_Trans_Performant)))
3415                 return h->access.command_completed(h);
3416
3417         if ((*(h->reply_pool_head) & 1) == (h->reply_pool_wraparound)) {
3418                 a = *(h->reply_pool_head); /* Next cmd in ring buffer */
3419                 (h->reply_pool_head)++;
3420                 h->commands_outstanding--;
3421         } else {
3422                 a = FIFO_EMPTY;
3423         }
3424         /* Check for wraparound */
3425         if (h->reply_pool_head == (h->reply_pool + h->max_commands)) {
3426                 h->reply_pool_head = h->reply_pool;
3427                 h->reply_pool_wraparound ^= 1;
3428         }
3429         return a;
3430 }
3431
3432 /* process completion of an indexed ("direct lookup") command */
3433 static inline u32 process_indexed_cmd(ctlr_info_t *h, u32 raw_tag)
3434 {
3435         u32 tag_index;
3436         CommandList_struct *c;
3437
3438         tag_index = cciss_tag_to_index(raw_tag);
3439         if (bad_tag(h, tag_index, raw_tag))
3440                 return next_command(h);
3441         c = h->cmd_pool + tag_index;
3442         finish_cmd(h, c, raw_tag);
3443         return next_command(h);
3444 }
3445
3446 /* process completion of a non-indexed command */
3447 static inline u32 process_nonindexed_cmd(ctlr_info_t *h, u32 raw_tag)
3448 {
3449         CommandList_struct *c = NULL;
3450         __u32 busaddr_masked, tag_masked;
3451
3452         tag_masked = cciss_tag_discard_error_bits(h, raw_tag);
3453         list_for_each_entry(c, &h->cmpQ, list) {
3454                 busaddr_masked = cciss_tag_discard_error_bits(h, c->busaddr);
3455                 if (busaddr_masked == tag_masked) {
3456                         finish_cmd(h, c, raw_tag);
3457                         return next_command(h);
3458                 }
3459         }
3460         bad_tag(h, h->nr_cmds + 1, raw_tag);
3461         return next_command(h);
3462 }
3463
3464 static irqreturn_t do_cciss_intx(int irq, void *dev_id)
3465 {
3466         ctlr_info_t *h = dev_id;
3467         unsigned long flags;
3468         u32 raw_tag;
3469
3470         if (interrupt_not_for_us(h))
3471                 return IRQ_NONE;
3472         spin_lock_irqsave(&h->lock, flags);
3473         while (interrupt_pending(h)) {
3474                 raw_tag = get_next_completion(h);
3475                 while (raw_tag != FIFO_EMPTY) {
3476                         if (cciss_tag_contains_index(raw_tag))
3477                                 raw_tag = process_indexed_cmd(h, raw_tag);
3478                         else
3479                                 raw_tag = process_nonindexed_cmd(h, raw_tag);
3480                 }
3481         }
3482         spin_unlock_irqrestore(&h->lock, flags);
3483         return IRQ_HANDLED;
3484 }
3485
3486 /* Add a second interrupt handler for MSI/MSI-X mode. In this mode we never
3487  * check the interrupt pending register because it is not set.
3488  */
3489 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id)
3490 {
3491         ctlr_info_t *h = dev_id;
3492         unsigned long flags;
3493         u32 raw_tag;
3494
3495         spin_lock_irqsave(&h->lock, flags);
3496         raw_tag = get_next_completion(h);
3497         while (raw_tag != FIFO_EMPTY) {
3498                 if (cciss_tag_contains_index(raw_tag))
3499                         raw_tag = process_indexed_cmd(h, raw_tag);
3500                 else
3501                         raw_tag = process_nonindexed_cmd(h, raw_tag);
3502         }
3503         spin_unlock_irqrestore(&h->lock, flags);
3504         return IRQ_HANDLED;
3505 }
3506
3507 /**
3508  * add_to_scan_list() - add controller to rescan queue
3509  * @h:                Pointer to the controller.
3510  *
3511  * Adds the controller to the rescan queue if not already on the queue.
3512  *
3513  * returns 1 if added to the queue, 0 if skipped (could be on the
3514  * queue already, or the controller could be initializing or shutting
3515  * down).
3516  **/
3517 static int add_to_scan_list(struct ctlr_info *h)
3518 {
3519         struct ctlr_info *test_h;
3520         int found = 0;
3521         int ret = 0;
3522
3523         if (h->busy_initializing)
3524                 return 0;
3525
3526         if (!mutex_trylock(&h->busy_shutting_down))
3527                 return 0;
3528
3529         mutex_lock(&scan_mutex);
3530         list_for_each_entry(test_h, &scan_q, scan_list) {
3531                 if (test_h == h) {
3532                         found = 1;
3533                         break;
3534                 }
3535         }
3536         if (!found && !h->busy_scanning) {
3537                 INIT_COMPLETION(h->scan_wait);
3538                 list_add_tail(&h->scan_list, &scan_q);
3539                 ret = 1;
3540         }
3541         mutex_unlock(&scan_mutex);
3542         mutex_unlock(&h->busy_shutting_down);
3543
3544         return ret;
3545 }
3546
3547 /**
3548  * remove_from_scan_list() - remove controller from rescan queue
3549  * @h:                     Pointer to the controller.
3550  *
3551  * Removes the controller from the rescan queue if present. Blocks if
3552  * the controller is currently conducting a rescan.  The controller
3553  * can be in one of three states:
3554  * 1. Doesn't need a scan
3555  * 2. On the scan list, but not scanning yet (we remove it)
3556  * 3. Busy scanning (and not on the list). In this case we want to wait for
3557  *    the scan to complete to make sure the scanning thread for this
3558  *    controller is completely idle.
3559  **/
3560 static void remove_from_scan_list(struct ctlr_info *h)
3561 {
3562         struct ctlr_info *test_h, *tmp_h;
3563
3564         mutex_lock(&scan_mutex);
3565         list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3566                 if (test_h == h) { /* state 2. */
3567                         list_del(&h->scan_list);
3568                         complete_all(&h->scan_wait);
3569                         mutex_unlock(&scan_mutex);
3570                         return;
3571                 }
3572         }
3573         if (h->busy_scanning) { /* state 3. */
3574                 mutex_unlock(&scan_mutex);
3575                 wait_for_completion(&h->scan_wait);
3576         } else { /* state 1, nothing to do. */
3577                 mutex_unlock(&scan_mutex);
3578         }
3579 }
3580
3581 /**
3582  * scan_thread() - kernel thread used to rescan controllers
3583  * @data:        Ignored.
3584  *
3585  * A kernel thread used scan for drive topology changes on
3586  * controllers. The thread processes only one controller at a time
3587  * using a queue.  Controllers are added to the queue using
3588  * add_to_scan_list() and removed from the queue either after done
3589  * processing or using remove_from_scan_list().
3590  *
3591  * returns 0.
3592  **/
3593 static int scan_thread(void *data)
3594 {
3595         struct ctlr_info *h;
3596
3597         while (1) {
3598                 set_current_state(TASK_INTERRUPTIBLE);
3599                 schedule();
3600                 if (kthread_should_stop())
3601                         break;
3602
3603                 while (1) {
3604                         mutex_lock(&scan_mutex);
3605                         if (list_empty(&scan_q)) {
3606                                 mutex_unlock(&scan_mutex);
3607                                 break;
3608                         }
3609
3610                         h = list_entry(scan_q.next,
3611                                        struct ctlr_info,
3612                                        scan_list);
3613                         list_del(&h->scan_list);
3614                         h->busy_scanning = 1;
3615                         mutex_unlock(&scan_mutex);
3616
3617                         rebuild_lun_table(h, 0, 0);
3618                         complete_all(&h->scan_wait);
3619                         mutex_lock(&scan_mutex);
3620                         h->busy_scanning = 0;
3621                         mutex_unlock(&scan_mutex);
3622                 }
3623         }
3624
3625         return 0;
3626 }
3627
3628 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3629 {
3630         if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3631                 return 0;
3632
3633         switch (c->err_info->SenseInfo[12]) {
3634         case STATE_CHANGED:
3635                 dev_warn(&h->pdev->dev, "a state change "
3636                         "detected, command retried\n");
3637                 return 1;
3638         break;
3639         case LUN_FAILED:
3640                 dev_warn(&h->pdev->dev, "LUN failure "
3641                         "detected, action required\n");
3642                 return 1;
3643         break;
3644         case REPORT_LUNS_CHANGED:
3645                 dev_warn(&h->pdev->dev, "report LUN data changed\n");
3646         /*
3647          * Here, we could call add_to_scan_list and wake up the scan thread,
3648          * except that it's quite likely that we will get more than one
3649          * REPORT_LUNS_CHANGED condition in quick succession, which means
3650          * that those which occur after the first one will likely happen
3651          * *during* the scan_thread's rescan.  And the rescan code is not
3652          * robust enough to restart in the middle, undoing what it has already
3653          * done, and it's not clear that it's even possible to do this, since
3654          * part of what it does is notify the block layer, which starts
3655          * doing it's own i/o to read partition tables and so on, and the
3656          * driver doesn't have visibility to know what might need undoing.
3657          * In any event, if possible, it is horribly complicated to get right
3658          * so we just don't do it for now.
3659          *
3660          * Note: this REPORT_LUNS_CHANGED condition only occurs on the MSA2012.
3661          */
3662                 return 1;
3663         break;
3664         case POWER_OR_RESET:
3665                 dev_warn(&h->pdev->dev,
3666                         "a power on or device reset detected\n");
3667                 return 1;
3668         break;
3669         case UNIT_ATTENTION_CLEARED:
3670                 dev_warn(&h->pdev->dev,
3671                         "unit attention cleared by another initiator\n");
3672                 return 1;
3673         break;
3674         default:
3675                 dev_warn(&h->pdev->dev, "unknown unit attention detected\n");
3676                 return 1;
3677         }
3678 }
3679
3680 /*
3681  *  We cannot read the structure directly, for portability we must use
3682  *   the io functions.
3683  *   This is for debug only.
3684  */
3685 static void print_cfg_table(ctlr_info_t *h)
3686 {
3687         int i;
3688         char temp_name[17];
3689         CfgTable_struct *tb = h->cfgtable;
3690
3691         dev_dbg(&h->pdev->dev, "Controller Configuration information\n");
3692         dev_dbg(&h->pdev->dev, "------------------------------------\n");
3693         for (i = 0; i < 4; i++)
3694                 temp_name[i] = readb(&(tb->Signature[i]));
3695         temp_name[4] = '\0';
3696         dev_dbg(&h->pdev->dev, "   Signature = %s\n", temp_name);
3697         dev_dbg(&h->pdev->dev, "   Spec Number = %d\n",
3698                 readl(&(tb->SpecValence)));
3699         dev_dbg(&h->pdev->dev, "   Transport methods supported = 0x%x\n",
3700                readl(&(tb->TransportSupport)));
3701         dev_dbg(&h->pdev->dev, "   Transport methods active = 0x%x\n",
3702                readl(&(tb->TransportActive)));
3703         dev_dbg(&h->pdev->dev, "   Requested transport Method = 0x%x\n",
3704                readl(&(tb->HostWrite.TransportRequest)));
3705         dev_dbg(&h->pdev->dev, "   Coalesce Interrupt Delay = 0x%x\n",
3706                readl(&(tb->HostWrite.CoalIntDelay)));
3707         dev_dbg(&h->pdev->dev, "   Coalesce Interrupt Count = 0x%x\n",
3708                readl(&(tb->HostWrite.CoalIntCount)));
3709         dev_dbg(&h->pdev->dev, "   Max outstanding commands = 0x%d\n",
3710                readl(&(tb->CmdsOutMax)));
3711         dev_dbg(&h->pdev->dev, "   Bus Types = 0x%x\n",
3712                 readl(&(tb->BusTypes)));
3713         for (i = 0; i < 16; i++)
3714                 temp_name[i] = readb(&(tb->ServerName[i]));
3715         temp_name[16] = '\0';
3716         dev_dbg(&h->pdev->dev, "   Server Name = %s\n", temp_name);
3717         dev_dbg(&h->pdev->dev, "   Heartbeat Counter = 0x%x\n\n\n",
3718                 readl(&(tb->HeartBeat)));
3719 }
3720
3721 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3722 {
3723         int i, offset, mem_type, bar_type;
3724         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3725                 return 0;
3726         offset = 0;
3727         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3728                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3729                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3730                         offset += 4;
3731                 else {
3732                         mem_type = pci_resource_flags(pdev, i) &
3733                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3734                         switch (mem_type) {
3735                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
3736                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3737                                 offset += 4;    /* 32 bit */
3738                                 break;
3739                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
3740                                 offset += 8;
3741                                 break;
3742                         default:        /* reserved in PCI 2.2 */
3743                                 dev_warn(&pdev->dev,
3744                                        "Base address is invalid\n");
3745                                 return -1;
3746                                 break;
3747                         }
3748                 }
3749                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3750                         return i + 1;
3751         }
3752         return -1;
3753 }
3754
3755 /* Fill in bucket_map[], given nsgs (the max number of
3756  * scatter gather elements supported) and bucket[],
3757  * which is an array of 8 integers.  The bucket[] array
3758  * contains 8 different DMA transfer sizes (in 16
3759  * byte increments) which the controller uses to fetch
3760  * commands.  This function fills in bucket_map[], which
3761  * maps a given number of scatter gather elements to one of
3762  * the 8 DMA transfer sizes.  The point of it is to allow the
3763  * controller to only do as much DMA as needed to fetch the
3764  * command, with the DMA transfer size encoded in the lower
3765  * bits of the command address.
3766  */
3767 static void  calc_bucket_map(int bucket[], int num_buckets,
3768         int nsgs, int *bucket_map)
3769 {
3770         int i, j, b, size;
3771
3772         /* even a command with 0 SGs requires 4 blocks */
3773 #define MINIMUM_TRANSFER_BLOCKS 4
3774 #define NUM_BUCKETS 8
3775         /* Note, bucket_map must have nsgs+1 entries. */
3776         for (i = 0; i <= nsgs; i++) {
3777                 /* Compute size of a command with i SG entries */
3778                 size = i + MINIMUM_TRANSFER_BLOCKS;
3779                 b = num_buckets; /* Assume the biggest bucket */
3780                 /* Find the bucket that is just big enough */
3781                 for (j = 0; j < 8; j++) {
3782                         if (bucket[j] >= size) {
3783                                 b = j;
3784                                 break;
3785                         }
3786                 }
3787                 /* for a command with i SG entries, use bucket b. */
3788                 bucket_map[i] = b;
3789         }
3790 }
3791
3792 static void __devinit cciss_wait_for_mode_change_ack(ctlr_info_t *h)
3793 {
3794         int i;
3795
3796         /* under certain very rare conditions, this can take awhile.
3797          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3798          * as we enter this code.) */
3799         for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3800                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3801                         break;
3802                 usleep_range(10000, 20000);
3803         }
3804 }
3805
3806 static __devinit void cciss_enter_performant_mode(ctlr_info_t *h,
3807         u32 use_short_tags)
3808 {
3809         /* This is a bit complicated.  There are 8 registers on
3810          * the controller which we write to to tell it 8 different
3811          * sizes of commands which there may be.  It's a way of
3812          * reducing the DMA done to fetch each command.  Encoded into
3813          * each command's tag are 3 bits which communicate to the controller
3814          * which of the eight sizes that command fits within.  The size of
3815          * each command depends on how many scatter gather entries there are.
3816          * Each SG entry requires 16 bytes.  The eight registers are programmed
3817          * with the number of 16-byte blocks a command of that size requires.
3818          * The smallest command possible requires 5 such 16 byte blocks.
3819          * the largest command possible requires MAXSGENTRIES + 4 16-byte
3820          * blocks.  Note, this only extends to the SG entries contained
3821          * within the command block, and does not extend to chained blocks
3822          * of SG elements.   bft[] contains the eight values we write to
3823          * the registers.  They are not evenly distributed, but have more
3824          * sizes for small commands, and fewer sizes for larger commands.
3825          */
3826         __u32 trans_offset;
3827         int bft[8] = { 5, 6, 8, 10, 12, 20, 28, MAXSGENTRIES + 4};
3828                         /*
3829                          *  5 = 1 s/g entry or 4k
3830                          *  6 = 2 s/g entry or 8k
3831                          *  8 = 4 s/g entry or 16k
3832                          * 10 = 6 s/g entry or 24k
3833                          */
3834         unsigned long register_value;
3835         BUILD_BUG_ON(28 > MAXSGENTRIES + 4);
3836
3837         h->reply_pool_wraparound = 1; /* spec: init to 1 */
3838
3839         /* Controller spec: zero out this buffer. */
3840         memset(h->reply_pool, 0, h->max_commands * sizeof(__u64));
3841         h->reply_pool_head = h->reply_pool;
3842
3843         trans_offset = readl(&(h->cfgtable->TransMethodOffset));
3844         calc_bucket_map(bft, ARRAY_SIZE(bft), h->maxsgentries,
3845                                 h->blockFetchTable);
3846         writel(bft[0], &h->transtable->BlockFetch0);
3847         writel(bft[1], &h->transtable->BlockFetch1);
3848         writel(bft[2], &h->transtable->BlockFetch2);
3849         writel(bft[3], &h->transtable->BlockFetch3);
3850         writel(bft[4], &h->transtable->BlockFetch4);
3851         writel(bft[5], &h->transtable->BlockFetch5);
3852         writel(bft[6], &h->transtable->BlockFetch6);
3853         writel(bft[7], &h->transtable->BlockFetch7);
3854
3855         /* size of controller ring buffer */
3856         writel(h->max_commands, &h->transtable->RepQSize);
3857         writel(1, &h->transtable->RepQCount);
3858         writel(0, &h->transtable->RepQCtrAddrLow32);
3859         writel(0, &h->transtable->RepQCtrAddrHigh32);
3860         writel(h->reply_pool_dhandle, &h->transtable->RepQAddr0Low32);
3861         writel(0, &h->transtable->RepQAddr0High32);
3862         writel(CFGTBL_Trans_Performant | use_short_tags,
3863                         &(h->cfgtable->HostWrite.TransportRequest));
3864
3865         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
3866         cciss_wait_for_mode_change_ack(h);
3867         register_value = readl(&(h->cfgtable->TransportActive));
3868         if (!(register_value & CFGTBL_Trans_Performant))
3869                 dev_warn(&h->pdev->dev, "cciss: unable to get board into"
3870                                         " performant mode\n");
3871 }
3872
3873 static void __devinit cciss_put_controller_into_performant_mode(ctlr_info_t *h)
3874 {
3875         __u32 trans_support;
3876
3877         dev_dbg(&h->pdev->dev, "Trying to put board into Performant mode\n");
3878         /* Attempt to put controller into performant mode if supported */
3879         /* Does board support performant mode? */
3880         trans_support = readl(&(h->cfgtable->TransportSupport));
3881         if (!(trans_support & PERFORMANT_MODE))
3882                 return;
3883
3884         dev_dbg(&h->pdev->dev, "Placing controller into performant mode\n");
3885         /* Performant mode demands commands on a 32 byte boundary
3886          * pci_alloc_consistent aligns on page boundarys already.
3887          * Just need to check if divisible by 32
3888          */
3889         if ((sizeof(CommandList_struct) % 32) != 0) {
3890                 dev_warn(&h->pdev->dev, "%s %d %s\n",
3891                         "cciss info: command size[",
3892                         (int)sizeof(CommandList_struct),
3893                         "] not divisible by 32, no performant mode..\n");
3894                 return;
3895         }
3896
3897         /* Performant mode ring buffer and supporting data structures */
3898         h->reply_pool = (__u64 *)pci_alloc_consistent(
3899                 h->pdev, h->max_commands * sizeof(__u64),
3900                 &(h->reply_pool_dhandle));
3901
3902         /* Need a block fetch table for performant mode */
3903         h->blockFetchTable = kmalloc(((h->maxsgentries+1) *
3904                 sizeof(__u32)), GFP_KERNEL);
3905
3906         if ((h->reply_pool == NULL) || (h->blockFetchTable == NULL))
3907                 goto clean_up;
3908
3909         cciss_enter_performant_mode(h,
3910                 trans_support & CFGTBL_Trans_use_short_tags);
3911
3912         /* Change the access methods to the performant access methods */
3913         h->access = SA5_performant_access;
3914         h->transMethod = CFGTBL_Trans_Performant;
3915
3916         return;
3917 clean_up:
3918         kfree(h->blockFetchTable);
3919         if (h->reply_pool)
3920                 pci_free_consistent(h->pdev,
3921                                 h->max_commands * sizeof(__u64),
3922                                 h->reply_pool,
3923                                 h->reply_pool_dhandle);
3924         return;
3925
3926 } /* cciss_put_controller_into_performant_mode */
3927
3928 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
3929  * controllers that are capable. If not, we use IO-APIC mode.
3930  */
3931
3932 static void __devinit cciss_interrupt_mode(ctlr_info_t *h)
3933 {
3934 #ifdef CONFIG_PCI_MSI
3935         int err;
3936         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
3937         {0, 2}, {0, 3}
3938         };
3939
3940         /* Some boards advertise MSI but don't really support it */
3941         if ((h->board_id == 0x40700E11) || (h->board_id == 0x40800E11) ||
3942             (h->board_id == 0x40820E11) || (h->board_id == 0x40830E11))
3943                 goto default_int_mode;
3944
3945         if (pci_find_capability(h->pdev, PCI_CAP_ID_MSIX)) {
3946                 err = pci_enable_msix(h->pdev, cciss_msix_entries, 4);
3947                 if (!err) {
3948                         h->intr[0] = cciss_msix_entries[0].vector;
3949                         h->intr[1] = cciss_msix_entries[1].vector;
3950                         h->intr[2] = cciss_msix_entries[2].vector;
3951                         h->intr[3] = cciss_msix_entries[3].vector;
3952                         h->msix_vector = 1;
3953                         return;
3954                 }
3955                 if (err > 0) {
3956                         dev_warn(&h->pdev->dev,
3957                                 "only %d MSI-X vectors available\n", err);
3958                         goto default_int_mode;
3959                 } else {
3960                         dev_warn(&h->pdev->dev,
3961                                 "MSI-X init failed %d\n", err);
3962                         goto default_int_mode;
3963                 }
3964         }
3965         if (pci_find_capability(h->pdev, PCI_CAP_ID_MSI)) {
3966                 if (!pci_enable_msi(h->pdev))
3967                         h->msi_vector = 1;
3968                 else
3969                         dev_warn(&h->pdev->dev, "MSI init failed\n");
3970         }
3971 default_int_mode:
3972 #endif                          /* CONFIG_PCI_MSI */
3973         /* if we get here we're going to use the default interrupt mode */
3974         h->intr[PERF_MODE_INT] = h->pdev->irq;
3975         return;
3976 }
3977
3978 static int __devinit cciss_lookup_board_id(struct pci_dev *pdev, u32 *board_id)
3979 {
3980         int i;
3981         u32 subsystem_vendor_id, subsystem_device_id;
3982
3983         subsystem_vendor_id = pdev->subsystem_vendor;
3984         subsystem_device_id = pdev->subsystem_device;
3985         *board_id = ((subsystem_device_id << 16) & 0xffff0000) |
3986                         subsystem_vendor_id;
3987
3988         for (i = 0; i < ARRAY_SIZE(products); i++)
3989                 if (*board_id == products[i].board_id)
3990                         return i;
3991         dev_warn(&pdev->dev, "unrecognized board ID: 0x%08x, ignoring.\n",
3992                 *board_id);
3993         return -ENODEV;
3994 }
3995
3996 static inline bool cciss_board_disabled(ctlr_info_t *h)
3997 {
3998         u16 command;
3999
4000         (void) pci_read_config_word(h->pdev, PCI_COMMAND, &command);
4001         return ((command & PCI_COMMAND_MEMORY) == 0);
4002 }
4003
4004 static int __devinit cciss_pci_find_memory_BAR(struct pci_dev *pdev,
4005         unsigned long *memory_bar)
4006 {
4007         int i;
4008
4009         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
4010                 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM) {
4011                         /* addressing mode bits already removed */
4012                         *memory_bar = pci_resource_start(pdev, i);
4013                         dev_dbg(&pdev->dev, "memory BAR = %lx\n",
4014                                 *memory_bar);
4015                         return 0;
4016                 }
4017         dev_warn(&pdev->dev, "no memory BAR found\n");
4018         return -ENODEV;
4019 }
4020
4021 static int __devinit cciss_wait_for_board_state(struct pci_dev *pdev,
4022         void __iomem *vaddr, int wait_for_ready)
4023 #define BOARD_READY 1
4024 #define BOARD_NOT_READY 0
4025 {
4026         int i, iterations;
4027         u32 scratchpad;
4028
4029         if (wait_for_ready)
4030                 iterations = CCISS_BOARD_READY_ITERATIONS;
4031         else
4032                 iterations = CCISS_BOARD_NOT_READY_ITERATIONS;
4033
4034         for (i = 0; i < iterations; i++) {
4035                 scratchpad = readl(vaddr + SA5_SCRATCHPAD_OFFSET);
4036                 if (wait_for_ready) {
4037                         if (scratchpad == CCISS_FIRMWARE_READY)
4038                                 return 0;
4039                 } else {
4040                         if (scratchpad != CCISS_FIRMWARE_READY)
4041                                 return 0;
4042                 }
4043                 msleep(CCISS_BOARD_READY_POLL_INTERVAL_MSECS);
4044         }
4045         dev_warn(&pdev->dev, "board not ready, timed out.\n");
4046         return -ENODEV;
4047 }
4048
4049 static int __devinit cciss_find_cfg_addrs(struct pci_dev *pdev,
4050         void __iomem *vaddr, u32 *cfg_base_addr, u64 *cfg_base_addr_index,
4051         u64 *cfg_offset)
4052 {
4053         *cfg_base_addr = readl(vaddr + SA5_CTCFG_OFFSET);
4054         *cfg_offset = readl(vaddr + SA5_CTMEM_OFFSET);
4055         *cfg_base_addr &= (u32) 0x0000ffff;
4056         *cfg_base_addr_index = find_PCI_BAR_index(pdev, *cfg_base_addr);
4057         if (*cfg_base_addr_index == -1) {
4058                 dev_warn(&pdev->dev, "cannot find cfg_base_addr_index, "
4059                         "*cfg_base_addr = 0x%08x\n", *cfg_base_addr);
4060                 return -ENODEV;
4061         }
4062         return 0;
4063 }
4064
4065 static int __devinit cciss_find_cfgtables(ctlr_info_t *h)
4066 {
4067         u64 cfg_offset;
4068         u32 cfg_base_addr;
4069         u64 cfg_base_addr_index;
4070         u32 trans_offset;
4071         int rc;
4072
4073         rc = cciss_find_cfg_addrs(h->pdev, h->vaddr, &cfg_base_addr,
4074                 &cfg_base_addr_index, &cfg_offset);
4075         if (rc)
4076                 return rc;
4077         h->cfgtable = remap_pci_mem(pci_resource_start(h->pdev,
4078                 cfg_base_addr_index) + cfg_offset, sizeof(h->cfgtable));
4079         if (!h->cfgtable)
4080                 return -ENOMEM;
4081         /* Find performant mode table. */
4082         trans_offset = readl(&h->cfgtable->TransMethodOffset);
4083         h->transtable = remap_pci_mem(pci_resource_start(h->pdev,
4084                                 cfg_base_addr_index)+cfg_offset+trans_offset,
4085                                 sizeof(*h->transtable));
4086         if (!h->transtable)
4087                 return -ENOMEM;
4088         return 0;
4089 }
4090
4091 static void __devinit cciss_get_max_perf_mode_cmds(struct ctlr_info *h)
4092 {
4093         h->max_commands = readl(&(h->cfgtable->MaxPerformantModeCommands));
4094
4095         /* Limit commands in memory limited kdump scenario. */
4096         if (reset_devices && h->max_commands > 32)
4097                 h->max_commands = 32;
4098
4099         if (h->max_commands < 16) {
4100                 dev_warn(&h->pdev->dev, "Controller reports "
4101                         "max supported commands of %d, an obvious lie. "
4102                         "Using 16.  Ensure that firmware is up to date.\n",
4103                         h->max_commands);
4104                 h->max_commands = 16;
4105         }
4106 }
4107
4108 /* Interrogate the hardware for some limits:
4109  * max commands, max SG elements without chaining, and with chaining,
4110  * SG chain block size, etc.
4111  */
4112 static void __devinit cciss_find_board_params(ctlr_info_t *h)
4113 {
4114         cciss_get_max_perf_mode_cmds(h);
4115         h->nr_cmds = h->max_commands - 4; /* Allow room for some ioctls */
4116         h->maxsgentries = readl(&(h->cfgtable->MaxSGElements));
4117         /*
4118          * Limit in-command s/g elements to 32 save dma'able memory.
4119          * Howvever spec says if 0, use 31
4120          */
4121         h->max_cmd_sgentries = 31;
4122         if (h->maxsgentries > 512) {
4123                 h->max_cmd_sgentries = 32;
4124                 h->chainsize = h->maxsgentries - h->max_cmd_sgentries + 1;
4125                 h->maxsgentries--; /* save one for chain pointer */
4126         } else {
4127                 h->maxsgentries = 31; /* default to traditional values */
4128                 h->chainsize = 0;
4129         }
4130 }
4131
4132 static inline bool CISS_signature_present(ctlr_info_t *h)
4133 {
4134         if ((readb(&h->cfgtable->Signature[0]) != 'C') ||
4135             (readb(&h->cfgtable->Signature[1]) != 'I') ||
4136             (readb(&h->cfgtable->Signature[2]) != 'S') ||
4137             (readb(&h->cfgtable->Signature[3]) != 'S')) {
4138                 dev_warn(&h->pdev->dev, "not a valid CISS config table\n");
4139                 return false;
4140         }
4141         return true;
4142 }
4143
4144 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
4145 static inline void cciss_enable_scsi_prefetch(ctlr_info_t *h)
4146 {
4147 #ifdef CONFIG_X86
4148         u32 prefetch;
4149
4150         prefetch = readl(&(h->cfgtable->SCSI_Prefetch));
4151         prefetch |= 0x100;
4152         writel(prefetch, &(h->cfgtable->SCSI_Prefetch));
4153 #endif
4154 }
4155
4156 /* Disable DMA prefetch for the P600.  Otherwise an ASIC bug may result
4157  * in a prefetch beyond physical memory.
4158  */
4159 static inline void cciss_p600_dma_prefetch_quirk(ctlr_info_t *h)
4160 {
4161         u32 dma_prefetch;
4162         __u32 dma_refetch;
4163
4164         if (h->board_id != 0x3225103C)
4165                 return;
4166         dma_prefetch = readl(h->vaddr + I2O_DMA1_CFG);
4167         dma_prefetch |= 0x8000;
4168         writel(dma_prefetch, h->vaddr + I2O_DMA1_CFG);
4169         pci_read_config_dword(h->pdev, PCI_COMMAND_PARITY, &dma_refetch);
4170         dma_refetch |= 0x1;
4171         pci_write_config_dword(h->pdev, PCI_COMMAND_PARITY, dma_refetch);
4172 }
4173
4174 static int __devinit cciss_pci_init(ctlr_info_t *h)
4175 {
4176         int prod_index, err;
4177
4178         prod_index = cciss_lookup_board_id(h->pdev, &h->board_id);
4179         if (prod_index < 0)
4180                 return -ENODEV;
4181         h->product_name = products[prod_index].product_name;
4182         h->access = *(products[prod_index].access);
4183
4184         if (cciss_board_disabled(h)) {
4185                 dev_warn(&h->pdev->dev, "controller appears to be disabled\n");
4186                 return -ENODEV;
4187         }
4188         err = pci_enable_device(h->pdev);
4189         if (err) {
4190                 dev_warn(&h->pdev->dev, "Unable to Enable PCI device\n");
4191                 return err;
4192         }
4193
4194         err = pci_request_regions(h->pdev, "cciss");
4195         if (err) {
4196                 dev_warn(&h->pdev->dev,
4197                         "Cannot obtain PCI resources, aborting\n");
4198                 return err;
4199         }
4200
4201         dev_dbg(&h->pdev->dev, "irq = %x\n", h->pdev->irq);
4202         dev_dbg(&h->pdev->dev, "board_id = %x\n", h->board_id);
4203
4204 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
4205  * else we use the IO-APIC interrupt assigned to us by system ROM.
4206  */
4207         cciss_interrupt_mode(h);
4208         err = cciss_pci_find_memory_BAR(h->pdev, &h->paddr);
4209         if (err)
4210                 goto err_out_free_res;
4211         h->vaddr = remap_pci_mem(h->paddr, 0x250);
4212         if (!h->vaddr) {
4213                 err = -ENOMEM;
4214                 goto err_out_free_res;
4215         }
4216         err = cciss_wait_for_board_state(h->pdev, h->vaddr, BOARD_READY);
4217         if (err)
4218                 goto err_out_free_res;
4219         err = cciss_find_cfgtables(h);
4220         if (err)
4221                 goto err_out_free_res;
4222         print_cfg_table(h);
4223         cciss_find_board_params(h);
4224
4225         if (!CISS_signature_present(h)) {
4226                 err = -ENODEV;
4227                 goto err_out_free_res;
4228         }
4229         cciss_enable_scsi_prefetch(h);
4230         cciss_p600_dma_prefetch_quirk(h);
4231         cciss_put_controller_into_performant_mode(h);
4232         return 0;
4233
4234 err_out_free_res:
4235         /*
4236          * Deliberately omit pci_disable_device(): it does something nasty to
4237          * Smart Array controllers that pci_enable_device does not undo
4238          */
4239         if (h->transtable)
4240                 iounmap(h->transtable);
4241         if (h->cfgtable)
4242                 iounmap(h->cfgtable);
4243         if (h->vaddr)
4244                 iounmap(h->vaddr);
4245         pci_release_regions(h->pdev);
4246         return err;
4247 }
4248
4249 /* Function to find the first free pointer into our hba[] array
4250  * Returns -1 if no free entries are left.
4251  */
4252 static int alloc_cciss_hba(struct pci_dev *pdev)
4253 {
4254         int i;
4255
4256         for (i = 0; i < MAX_CTLR; i++) {
4257                 if (!hba[i]) {
4258                         ctlr_info_t *h;
4259
4260                         h = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
4261                         if (!h)
4262                                 goto Enomem;
4263                         hba[i] = h;
4264                         return i;
4265                 }
4266         }
4267         dev_warn(&pdev->dev, "This driver supports a maximum"
4268                " of %d controllers.\n", MAX_CTLR);
4269         return -1;
4270 Enomem:
4271         dev_warn(&pdev->dev, "out of memory.\n");
4272         return -1;
4273 }
4274
4275 static void free_hba(ctlr_info_t *h)
4276 {
4277         int i;
4278
4279         hba[h->ctlr] = NULL;
4280         for (i = 0; i < h->highest_lun + 1; i++)
4281                 if (h->gendisk[i] != NULL)
4282                         put_disk(h->gendisk[i]);
4283         kfree(h);
4284 }
4285
4286 /* Send a message CDB to the firmware. */
4287 static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
4288 {
4289         typedef struct {
4290                 CommandListHeader_struct CommandHeader;
4291                 RequestBlock_struct Request;
4292                 ErrDescriptor_struct ErrorDescriptor;
4293         } Command;
4294         static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
4295         Command *cmd;
4296         dma_addr_t paddr64;
4297         uint32_t paddr32, tag;
4298         void __iomem *vaddr;
4299         int i, err;
4300
4301         vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
4302         if (vaddr == NULL)
4303                 return -ENOMEM;
4304
4305         /* The Inbound Post Queue only accepts 32-bit physical addresses for the
4306            CCISS commands, so they must be allocated from the lower 4GiB of
4307            memory. */
4308         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
4309         if (err) {
4310                 iounmap(vaddr);
4311                 return -ENOMEM;
4312         }
4313
4314         cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
4315         if (cmd == NULL) {
4316                 iounmap(vaddr);
4317                 return -ENOMEM;
4318         }
4319
4320         /* This must fit, because of the 32-bit consistent DMA mask.  Also,
4321            although there's no guarantee, we assume that the address is at
4322            least 4-byte aligned (most likely, it's page-aligned). */
4323         paddr32 = paddr64;
4324
4325         cmd->CommandHeader.ReplyQueue = 0;
4326         cmd->CommandHeader.SGList = 0;
4327         cmd->CommandHeader.SGTotal = 0;
4328         cmd->CommandHeader.Tag.lower = paddr32;
4329         cmd->CommandHeader.Tag.upper = 0;
4330         memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
4331
4332         cmd->Request.CDBLen = 16;
4333         cmd->Request.Type.Type = TYPE_MSG;
4334         cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
4335         cmd->Request.Type.Direction = XFER_NONE;
4336         cmd->Request.Timeout = 0; /* Don't time out */
4337         cmd->Request.CDB[0] = opcode;
4338         cmd->Request.CDB[1] = type;
4339         memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
4340
4341         cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
4342         cmd->ErrorDescriptor.Addr.upper = 0;
4343         cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
4344
4345         writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
4346
4347         for (i = 0; i < 10; i++) {
4348                 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
4349                 if ((tag & ~3) == paddr32)
4350                         break;
4351                 schedule_timeout_uninterruptible(HZ);
4352         }
4353
4354         iounmap(vaddr);
4355
4356         /* we leak the DMA buffer here ... no choice since the controller could
4357            still complete the command. */
4358         if (i == 10) {
4359                 dev_err(&pdev->dev,
4360                         "controller message %02x:%02x timed out\n",
4361                         opcode, type);
4362                 return -ETIMEDOUT;
4363         }
4364
4365         pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
4366
4367         if (tag & 2) {
4368                 dev_err(&pdev->dev, "controller message %02x:%02x failed\n",
4369                         opcode, type);
4370                 return -EIO;
4371         }
4372
4373         dev_info(&pdev->dev, "controller message %02x:%02x succeeded\n",
4374                 opcode, type);
4375         return 0;
4376 }
4377
4378 #define cciss_soft_reset_controller(p) cciss_message(p, 1, 0)
4379 #define cciss_noop(p) cciss_message(p, 3, 0)
4380
4381 static int cciss_controller_hard_reset(struct pci_dev *pdev,
4382         void * __iomem vaddr, bool use_doorbell)
4383 {
4384         u16 pmcsr;
4385         int pos;
4386
4387         if (use_doorbell) {
4388                 /* For everything after the P600, the PCI power state method
4389                  * of resetting the controller doesn't work, so we have this
4390                  * other way using the doorbell register.
4391                  */
4392                 dev_info(&pdev->dev, "using doorbell to reset controller\n");
4393                 writel(DOORBELL_CTLR_RESET, vaddr + SA5_DOORBELL);
4394                 msleep(1000);
4395         } else { /* Try to do it the PCI power state way */
4396
4397                 /* Quoting from the Open CISS Specification: "The Power
4398                  * Management Control/Status Register (CSR) controls the power
4399                  * state of the device.  The normal operating state is D0,
4400                  * CSR=00h.  The software off state is D3, CSR=03h.  To reset
4401                  * the controller, place the interface device in D3 then to D0,
4402                  * this causes a secondary PCI reset which will reset the
4403                  * controller." */
4404
4405                 pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4406                 if (pos == 0) {
4407                         dev_err(&pdev->dev,
4408                                 "cciss_controller_hard_reset: "
4409                                 "PCI PM not supported\n");
4410                         return -ENODEV;
4411                 }
4412                 dev_info(&pdev->dev, "using PCI PM to reset controller\n");
4413                 /* enter the D3hot power management state */
4414                 pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4415                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4416                 pmcsr |= PCI_D3hot;
4417                 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4418
4419                 msleep(500);
4420
4421                 /* enter the D0 power management state */
4422                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4423                 pmcsr |= PCI_D0;
4424                 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4425
4426                 msleep(500);
4427         }
4428         return 0;
4429 }
4430
4431 /* This does a hard reset of the controller using PCI power management
4432  * states or using the doorbell register. */
4433 static __devinit int cciss_kdump_hard_reset_controller(struct pci_dev *pdev)
4434 {
4435         u64 cfg_offset;
4436         u32 cfg_base_addr;
4437         u64 cfg_base_addr_index;
4438         void __iomem *vaddr;
4439         unsigned long paddr;
4440         u32 misc_fw_support, active_transport;
4441         int rc;
4442         CfgTable_struct __iomem *cfgtable;
4443         bool use_doorbell;
4444         u32 board_id;
4445         u16 command_register;
4446
4447         /* For controllers as old a the p600, this is very nearly
4448          * the same thing as
4449          *
4450          * pci_save_state(pci_dev);
4451          * pci_set_power_state(pci_dev, PCI_D3hot);
4452          * pci_set_power_state(pci_dev, PCI_D0);
4453          * pci_restore_state(pci_dev);
4454          *
4455          * For controllers newer than the P600, the pci power state
4456          * method of resetting doesn't work so we have another way
4457          * using the doorbell register.
4458          */
4459
4460         /* Exclude 640x boards.  These are two pci devices in one slot
4461          * which share a battery backed cache module.  One controls the
4462          * cache, the other accesses the cache through the one that controls
4463          * it.  If we reset the one controlling the cache, the other will
4464          * likely not be happy.  Just forbid resetting this conjoined mess.
4465          */
4466         cciss_lookup_board_id(pdev, &board_id);
4467         if (board_id == 0x409C0E11 || board_id == 0x409D0E11) {
4468                 dev_warn(&pdev->dev, "Cannot reset Smart Array 640x "
4469                                 "due to shared cache module.");
4470                 return -ENODEV;
4471         }
4472
4473         /* Save the PCI command register */
4474         pci_read_config_word(pdev, 4, &command_register);
4475         /* Turn the board off.  This is so that later pci_restore_state()
4476          * won't turn the board on before the rest of config space is ready.
4477          */
4478         pci_disable_device(pdev);
4479         pci_save_state(pdev);
4480
4481         /* find the first memory BAR, so we can find the cfg table */
4482         rc = cciss_pci_find_memory_BAR(pdev, &paddr);
4483         if (rc)
4484                 return rc;
4485         vaddr = remap_pci_mem(paddr, 0x250);
4486         if (!vaddr)
4487                 return -ENOMEM;
4488
4489         /* find cfgtable in order to check if reset via doorbell is supported */
4490         rc = cciss_find_cfg_addrs(pdev, vaddr, &cfg_base_addr,
4491                                         &cfg_base_addr_index, &cfg_offset);
4492         if (rc)
4493                 goto unmap_vaddr;
4494         cfgtable = remap_pci_mem(pci_resource_start(pdev,
4495                        cfg_base_addr_index) + cfg_offset, sizeof(*cfgtable));
4496         if (!cfgtable) {
4497                 rc = -ENOMEM;
4498                 goto unmap_vaddr;
4499         }
4500
4501         /* If reset via doorbell register is supported, use that. */
4502         misc_fw_support = readl(&cfgtable->misc_fw_support);
4503         use_doorbell = misc_fw_support & MISC_FW_DOORBELL_RESET;
4504
4505         /* The doorbell reset seems to cause lockups on some Smart
4506          * Arrays (e.g. P410, P410i, maybe others).  Until this is
4507          * fixed or at least isolated, avoid the doorbell reset.
4508          */
4509         use_doorbell = 0;
4510
4511         rc = cciss_controller_hard_reset(pdev, vaddr, use_doorbell);
4512         if (rc)
4513                 goto unmap_cfgtable;
4514         pci_restore_state(pdev);
4515         rc = pci_enable_device(pdev);
4516         if (rc) {
4517                 dev_warn(&pdev->dev, "failed to enable device.\n");
4518                 goto unmap_cfgtable;
4519         }
4520         pci_write_config_word(pdev, 4, command_register);
4521
4522         /* Some devices (notably the HP Smart Array 5i Controller)
4523            need a little pause here */
4524         msleep(CCISS_POST_RESET_PAUSE_MSECS);
4525
4526         /* Wait for board to become not ready, then ready. */
4527         dev_info(&pdev->dev, "Waiting for board to become ready.\n");
4528         rc = cciss_wait_for_board_state(pdev, vaddr, BOARD_NOT_READY);
4529         if (rc) /* Don't bail, might be E500, etc. which can't be reset */
4530                 dev_warn(&pdev->dev,
4531                         "failed waiting for board to become not ready\n");
4532         rc = cciss_wait_for_board_state(pdev, vaddr, BOARD_READY);
4533         if (rc) {
4534                 dev_warn(&pdev->dev,
4535                         "failed waiting for board to become ready\n");
4536                 goto unmap_cfgtable;
4537         }
4538         dev_info(&pdev->dev, "board ready.\n");
4539
4540         /* Controller should be in simple mode at this point.  If it's not,
4541          * It means we're on one of those controllers which doesn't support
4542          * the doorbell reset method and on which the PCI power management reset
4543          * method doesn't work (P800, for example.)
4544          * In those cases, don't try to proceed, as it generally doesn't work.
4545          */
4546         active_transport = readl(&cfgtable->TransportActive);
4547         if (active_transport & PERFORMANT_MODE) {
4548                 dev_warn(&pdev->dev, "Unable to successfully reset controller,"
4549                         " Ignoring controller.\n");
4550                 rc = -ENODEV;
4551         }
4552
4553 unmap_cfgtable:
4554         iounmap(cfgtable);
4555
4556 unmap_vaddr:
4557         iounmap(vaddr);
4558         return rc;
4559 }
4560
4561 static __devinit int cciss_init_reset_devices(struct pci_dev *pdev)
4562 {
4563         int rc, i;
4564
4565         if (!reset_devices)
4566                 return 0;
4567
4568         /* Reset the controller with a PCI power-cycle or via doorbell */
4569         rc = cciss_kdump_hard_reset_controller(pdev);
4570
4571         /* -ENOTSUPP here means we cannot reset the controller
4572          * but it's already (and still) up and running in
4573          * "performant mode".  Or, it might be 640x, which can't reset
4574          * due to concerns about shared bbwc between 6402/6404 pair.
4575          */
4576         if (rc == -ENOTSUPP)
4577                 return 0; /* just try to do the kdump anyhow. */
4578         if (rc)
4579                 return -ENODEV;
4580
4581         /* Now try to get the controller to respond to a no-op */
4582         for (i = 0; i < CCISS_POST_RESET_NOOP_RETRIES; i++) {
4583                 if (cciss_noop(pdev) == 0)
4584                         break;
4585                 else
4586                         dev_warn(&pdev->dev, "no-op failed%s\n",
4587                                 (i < CCISS_POST_RESET_NOOP_RETRIES - 1 ?
4588                                         "; re-trying" : ""));
4589                 msleep(CCISS_POST_RESET_NOOP_INTERVAL_MSECS);
4590         }
4591         return 0;
4592 }
4593
4594 /*
4595  *  This is it.  Find all the controllers and register them.  I really hate
4596  *  stealing all these major device numbers.
4597  *  returns the number of block devices registered.
4598  */
4599 static int __devinit cciss_init_one(struct pci_dev *pdev,
4600                                     const struct pci_device_id *ent)
4601 {
4602         int i;
4603         int j = 0;
4604         int k = 0;
4605         int rc;
4606         int dac, return_code;
4607         InquiryData_struct *inq_buff;
4608         ctlr_info_t *h;
4609
4610         rc = cciss_init_reset_devices(pdev);
4611         if (rc)
4612                 return rc;
4613         i = alloc_cciss_hba(pdev);
4614         if (i < 0)
4615                 return -1;
4616
4617         h = hba[i];
4618         h->pdev = pdev;
4619         h->busy_initializing = 1;
4620         INIT_LIST_HEAD(&h->cmpQ);
4621         INIT_LIST_HEAD(&h->reqQ);
4622         mutex_init(&h->busy_shutting_down);
4623
4624         if (cciss_pci_init(h) != 0)
4625                 goto clean_no_release_regions;
4626
4627         sprintf(h->devname, "cciss%d", i);
4628         h->ctlr = i;
4629
4630         init_completion(&h->scan_wait);
4631
4632         if (cciss_create_hba_sysfs_entry(h))
4633                 goto clean0;
4634
4635         /* configure PCI DMA stuff */
4636         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
4637                 dac = 1;
4638         else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
4639                 dac = 0;
4640         else {
4641                 dev_err(&h->pdev->dev, "no suitable DMA available\n");
4642                 goto clean1;
4643         }
4644
4645         /*
4646          * register with the major number, or get a dynamic major number
4647          * by passing 0 as argument.  This is done for greater than
4648          * 8 controller support.
4649          */
4650         if (i < MAX_CTLR_ORIG)
4651                 h->major = COMPAQ_CISS_MAJOR + i;
4652         rc = register_blkdev(h->major, h->devname);
4653         if (rc == -EBUSY || rc == -EINVAL) {
4654                 dev_err(&h->pdev->dev,
4655                        "Unable to get major number %d for %s "
4656                        "on hba %d\n", h->major, h->devname, i);
4657                 goto clean1;
4658         } else {
4659                 if (i >= MAX_CTLR_ORIG)
4660                         h->major = rc;
4661         }
4662
4663         /* make sure the board interrupts are off */
4664         h->access.set_intr_mask(h, CCISS_INTR_OFF);
4665         if (h->msi_vector || h->msix_vector) {
4666                 if (request_irq(h->intr[PERF_MODE_INT],
4667                                 do_cciss_msix_intr,
4668                                 IRQF_DISABLED, h->devname, h)) {
4669                         dev_err(&h->pdev->dev, "Unable to get irq %d for %s\n",
4670                                h->intr[PERF_MODE_INT], h->devname);
4671                         goto clean2;
4672                 }
4673         } else {
4674                 if (request_irq(h->intr[PERF_MODE_INT], do_cciss_intx,
4675                                 IRQF_DISABLED, h->devname, h)) {
4676                         dev_err(&h->pdev->dev, "Unable to get irq %d for %s\n",
4677                                h->intr[PERF_MODE_INT], h->devname);
4678                         goto clean2;
4679                 }
4680         }
4681
4682         dev_info(&h->pdev->dev, "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
4683                h->devname, pdev->device, pci_name(pdev),
4684                h->intr[PERF_MODE_INT], dac ? "" : " not");
4685
4686         h->cmd_pool_bits =
4687             kmalloc(DIV_ROUND_UP(h->nr_cmds, BITS_PER_LONG)
4688                         * sizeof(unsigned long), GFP_KERNEL);
4689         h->cmd_pool = (CommandList_struct *)
4690             pci_alloc_consistent(h->pdev,
4691                     h->nr_cmds * sizeof(CommandList_struct),
4692                     &(h->cmd_pool_dhandle));
4693         h->errinfo_pool = (ErrorInfo_struct *)
4694             pci_alloc_consistent(h->pdev,
4695                     h->nr_cmds * sizeof(ErrorInfo_struct),
4696                     &(h->errinfo_pool_dhandle));
4697         if ((h->cmd_pool_bits == NULL)
4698             || (h->cmd_pool == NULL)
4699             || (h->errinfo_pool == NULL)) {
4700                 dev_err(&h->pdev->dev, "out of memory");
4701                 goto clean4;
4702         }
4703
4704         /* Need space for temp scatter list */
4705         h->scatter_list = kmalloc(h->max_commands *
4706                                                 sizeof(struct scatterlist *),
4707                                                 GFP_KERNEL);
4708         if (!h->scatter_list)
4709                 goto clean4;
4710
4711         for (k = 0; k < h->nr_cmds; k++) {
4712                 h->scatter_list[k] = kmalloc(sizeof(struct scatterlist) *
4713                                                         h->maxsgentries,
4714                                                         GFP_KERNEL);
4715                 if (h->scatter_list[k] == NULL) {
4716                         dev_err(&h->pdev->dev,
4717                                 "could not allocate s/g lists\n");
4718                         goto clean4;
4719                 }
4720         }
4721         h->cmd_sg_list = cciss_allocate_sg_chain_blocks(h,
4722                 h->chainsize, h->nr_cmds);
4723         if (!h->cmd_sg_list && h->chainsize > 0)
4724                 goto clean4;
4725
4726         spin_lock_init(&h->lock);
4727
4728         /* Initialize the pdev driver private data.
4729            have it point to h.  */
4730         pci_set_drvdata(pdev, h);
4731         /* command and error info recs zeroed out before
4732            they are used */
4733         memset(h->cmd_pool_bits, 0,
4734                DIV_ROUND_UP(h->nr_cmds, BITS_PER_LONG)
4735                         * sizeof(unsigned long));
4736
4737         h->num_luns = 0;
4738         h->highest_lun = -1;
4739         for (j = 0; j < CISS_MAX_LUN; j++) {
4740                 h->drv[j] = NULL;
4741                 h->gendisk[j] = NULL;
4742         }
4743
4744         cciss_scsi_setup(h);
4745
4746         /* Turn the interrupts on so we can service requests */
4747         h->access.set_intr_mask(h, CCISS_INTR_ON);
4748
4749         /* Get the firmware version */
4750         inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
4751         if (inq_buff == NULL) {
4752                 dev_err(&h->pdev->dev, "out of memory\n");
4753                 goto clean4;
4754         }
4755
4756         return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
4757                 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
4758         if (return_code == IO_OK) {
4759                 h->firm_ver[0] = inq_buff->data_byte[32];
4760                 h->firm_ver[1] = inq_buff->data_byte[33];
4761                 h->firm_ver[2] = inq_buff->data_byte[34];
4762                 h->firm_ver[3] = inq_buff->data_byte[35];
4763         } else {         /* send command failed */
4764                 dev_warn(&h->pdev->dev, "unable to determine firmware"
4765                         " version of controller\n");
4766         }
4767         kfree(inq_buff);
4768
4769         cciss_procinit(h);
4770
4771         h->cciss_max_sectors = 8192;
4772
4773         rebuild_lun_table(h, 1, 0);
4774         h->busy_initializing = 0;
4775         return 1;
4776
4777 clean4:
4778         kfree(h->cmd_pool_bits);
4779         /* Free up sg elements */
4780         for (k-- ; k >= 0; k--)
4781                 kfree(h->scatter_list[k]);
4782         kfree(h->scatter_list);
4783         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
4784         if (h->cmd_pool)
4785                 pci_free_consistent(h->pdev,
4786                                     h->nr_cmds * sizeof(CommandList_struct),
4787                                     h->cmd_pool, h->cmd_pool_dhandle);
4788         if (h->errinfo_pool)
4789                 pci_free_consistent(h->pdev,
4790                                     h->nr_cmds * sizeof(ErrorInfo_struct),
4791                                     h->errinfo_pool,
4792                                     h->errinfo_pool_dhandle);
4793         free_irq(h->intr[PERF_MODE_INT], h);
4794 clean2:
4795         unregister_blkdev(h->major, h->devname);
4796 clean1:
4797         cciss_destroy_hba_sysfs_entry(h);
4798 clean0:
4799         pci_release_regions(pdev);
4800 clean_no_release_regions:
4801         h->busy_initializing = 0;
4802
4803         /*
4804          * Deliberately omit pci_disable_device(): it does something nasty to
4805          * Smart Array controllers that pci_enable_device does not undo
4806          */
4807         pci_set_drvdata(pdev, NULL);
4808         free_hba(h);
4809         return -1;
4810 }
4811
4812 static void cciss_shutdown(struct pci_dev *pdev)
4813 {
4814         ctlr_info_t *h;
4815         char *flush_buf;
4816         int return_code;
4817
4818         h = pci_get_drvdata(pdev);
4819         flush_buf = kzalloc(4, GFP_KERNEL);
4820         if (!flush_buf) {
4821                 dev_warn(&h->pdev->dev, "cache not flushed, out of memory.\n");
4822                 return;
4823         }
4824         /* write all data in the battery backed cache to disk */
4825         memset(flush_buf, 0, 4);
4826         return_code = sendcmd_withirq(h, CCISS_CACHE_FLUSH, flush_buf,
4827                 4, 0, CTLR_LUNID, TYPE_CMD);
4828         kfree(flush_buf);
4829         if (return_code != IO_OK)
4830                 dev_warn(&h->pdev->dev, "Error flushing cache\n");
4831         h->access.set_intr_mask(h, CCISS_INTR_OFF);
4832         free_irq(h->intr[PERF_MODE_INT], h);
4833 }
4834
4835 static void __devexit cciss_remove_one(struct pci_dev *pdev)
4836 {
4837         ctlr_info_t *h;
4838         int i, j;
4839
4840         if (pci_get_drvdata(pdev) == NULL) {
4841                 dev_err(&pdev->dev, "Unable to remove device\n");
4842                 return;
4843         }
4844
4845         h = pci_get_drvdata(pdev);
4846         i = h->ctlr;
4847         if (hba[i] == NULL) {
4848                 dev_err(&pdev->dev, "device appears to already be removed\n");
4849                 return;
4850         }
4851
4852         mutex_lock(&h->busy_shutting_down);
4853
4854         remove_from_scan_list(h);
4855         remove_proc_entry(h->devname, proc_cciss);
4856         unregister_blkdev(h->major, h->devname);
4857
4858         /* remove it from the disk list */
4859         for (j = 0; j < CISS_MAX_LUN; j++) {
4860                 struct gendisk *disk = h->gendisk[j];
4861                 if (disk) {
4862                         struct request_queue *q = disk->queue;
4863
4864                         if (disk->flags & GENHD_FL_UP) {
4865                                 cciss_destroy_ld_sysfs_entry(h, j, 1);
4866                                 del_gendisk(disk);
4867                         }
4868                         if (q)
4869                                 blk_cleanup_queue(q);
4870                 }
4871         }
4872
4873 #ifdef CONFIG_CISS_SCSI_TAPE
4874         cciss_unregister_scsi(h);       /* unhook from SCSI subsystem */
4875 #endif
4876
4877         cciss_shutdown(pdev);
4878
4879 #ifdef CONFIG_PCI_MSI
4880         if (h->msix_vector)
4881                 pci_disable_msix(h->pdev);
4882         else if (h->msi_vector)
4883                 pci_disable_msi(h->pdev);
4884 #endif                          /* CONFIG_PCI_MSI */
4885
4886         iounmap(h->transtable);
4887         iounmap(h->cfgtable);
4888         iounmap(h->vaddr);
4889
4890         pci_free_consistent(h->pdev, h->nr_cmds * sizeof(CommandList_struct),
4891                             h->cmd_pool, h->cmd_pool_dhandle);
4892         pci_free_consistent(h->pdev, h->nr_cmds * sizeof(ErrorInfo_struct),
4893                             h->errinfo_pool, h->errinfo_pool_dhandle);
4894         kfree(h->cmd_pool_bits);
4895         /* Free up sg elements */
4896         for (j = 0; j < h->nr_cmds; j++)
4897                 kfree(h->scatter_list[j]);
4898         kfree(h->scatter_list);
4899         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
4900         /*
4901          * Deliberately omit pci_disable_device(): it does something nasty to
4902          * Smart Array controllers that pci_enable_device does not undo
4903          */
4904         pci_release_regions(pdev);
4905         pci_set_drvdata(pdev, NULL);
4906         cciss_destroy_hba_sysfs_entry(h);
4907         mutex_unlock(&h->busy_shutting_down);
4908         free_hba(h);
4909 }
4910
4911 static struct pci_driver cciss_pci_driver = {
4912         .name = "cciss",
4913         .probe = cciss_init_one,
4914         .remove = __devexit_p(cciss_remove_one),
4915         .id_table = cciss_pci_device_id,        /* id_table */
4916         .shutdown = cciss_shutdown,
4917 };
4918
4919 /*
4920  *  This is it.  Register the PCI driver information for the cards we control
4921  *  the OS will call our registered routines when it finds one of our cards.
4922  */
4923 static int __init cciss_init(void)
4924 {
4925         int err;
4926
4927         /*
4928          * The hardware requires that commands are aligned on a 64-bit
4929          * boundary. Given that we use pci_alloc_consistent() to allocate an
4930          * array of them, the size must be a multiple of 8 bytes.
4931          */
4932         BUILD_BUG_ON(sizeof(CommandList_struct) % COMMANDLIST_ALIGNMENT);
4933         printk(KERN_INFO DRIVER_NAME "\n");
4934
4935         err = bus_register(&cciss_bus_type);
4936         if (err)
4937                 return err;
4938
4939         /* Start the scan thread */
4940         cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
4941         if (IS_ERR(cciss_scan_thread)) {
4942                 err = PTR_ERR(cciss_scan_thread);
4943                 goto err_bus_unregister;
4944         }
4945
4946         /* Register for our PCI devices */
4947         err = pci_register_driver(&cciss_pci_driver);
4948         if (err)
4949                 goto err_thread_stop;
4950
4951         return err;
4952
4953 err_thread_stop:
4954         kthread_stop(cciss_scan_thread);
4955 err_bus_unregister:
4956         bus_unregister(&cciss_bus_type);
4957
4958         return err;
4959 }
4960
4961 static void __exit cciss_cleanup(void)
4962 {
4963         int i;
4964
4965         pci_unregister_driver(&cciss_pci_driver);
4966         /* double check that all controller entrys have been removed */
4967         for (i = 0; i < MAX_CTLR; i++) {
4968                 if (hba[i] != NULL) {
4969                         dev_warn(&hba[i]->pdev->dev,
4970                                 "had to remove controller\n");
4971                         cciss_remove_one(hba[i]->pdev);
4972                 }
4973         }
4974         kthread_stop(cciss_scan_thread);
4975         if (proc_cciss)
4976                 remove_proc_entry("driver/cciss", NULL);
4977         bus_unregister(&cciss_bus_type);
4978 }
4979
4980 module_init(cciss_init);
4981 module_exit(cciss_cleanup);