a6c0494dd0544b67e454ffc9beac6592146fdcc7
[linux-2.6.git] / drivers / block / cciss.c
1 /*
2  *    Disk Array driver for HP Smart Array controllers.
3  *    (C) Copyright 2000, 2007 Hewlett-Packard Development Company, L.P.
4  *
5  *    This program is free software; you can redistribute it and/or modify
6  *    it under the terms of the GNU General Public License as published by
7  *    the Free Software Foundation; version 2 of the License.
8  *
9  *    This program is distributed in the hope that it will be useful,
10  *    but WITHOUT ANY WARRANTY; without even the implied warranty of
11  *    MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12  *    General Public License for more details.
13  *
14  *    You should have received a copy of the GNU General Public License
15  *    along with this program; if not, write to the Free Software
16  *    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
17  *    02111-1307, USA.
18  *
19  *    Questions/Comments/Bugfixes to iss_storagedev@hp.com
20  *
21  */
22
23 #include <linux/module.h>
24 #include <linux/interrupt.h>
25 #include <linux/types.h>
26 #include <linux/pci.h>
27 #include <linux/kernel.h>
28 #include <linux/slab.h>
29 #include <linux/smp_lock.h>
30 #include <linux/delay.h>
31 #include <linux/major.h>
32 #include <linux/fs.h>
33 #include <linux/bio.h>
34 #include <linux/blkpg.h>
35 #include <linux/timer.h>
36 #include <linux/proc_fs.h>
37 #include <linux/seq_file.h>
38 #include <linux/init.h>
39 #include <linux/jiffies.h>
40 #include <linux/hdreg.h>
41 #include <linux/spinlock.h>
42 #include <linux/compat.h>
43 #include <linux/mutex.h>
44 #include <asm/uaccess.h>
45 #include <asm/io.h>
46
47 #include <linux/dma-mapping.h>
48 #include <linux/blkdev.h>
49 #include <linux/genhd.h>
50 #include <linux/completion.h>
51 #include <scsi/scsi.h>
52 #include <scsi/sg.h>
53 #include <scsi/scsi_ioctl.h>
54 #include <linux/cdrom.h>
55 #include <linux/scatterlist.h>
56 #include <linux/kthread.h>
57
58 #define CCISS_DRIVER_VERSION(maj,min,submin) ((maj<<16)|(min<<8)|(submin))
59 #define DRIVER_NAME "HP CISS Driver (v 3.6.26)"
60 #define DRIVER_VERSION CCISS_DRIVER_VERSION(3, 6, 26)
61
62 /* Embedded module documentation macros - see modules.h */
63 MODULE_AUTHOR("Hewlett-Packard Company");
64 MODULE_DESCRIPTION("Driver for HP Smart Array Controllers");
65 MODULE_SUPPORTED_DEVICE("HP Smart Array Controllers");
66 MODULE_VERSION("3.6.26");
67 MODULE_LICENSE("GPL");
68
69 static int cciss_allow_hpsa;
70 module_param(cciss_allow_hpsa, int, S_IRUGO|S_IWUSR);
71 MODULE_PARM_DESC(cciss_allow_hpsa,
72         "Prevent cciss driver from accessing hardware known to be "
73         " supported by the hpsa driver");
74
75 #include "cciss_cmd.h"
76 #include "cciss.h"
77 #include <linux/cciss_ioctl.h>
78
79 /* define the PCI info for the cards we can control */
80 static const struct pci_device_id cciss_pci_device_id[] = {
81         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISS,  0x0E11, 0x4070},
82         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4080},
83         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4082},
84         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4083},
85         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x4091},
86         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409A},
87         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409B},
88         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409C},
89         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409D},
90         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSA,     0x103C, 0x3225},
91         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3223},
92         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3234},
93         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3235},
94         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3211},
95         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3212},
96         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3213},
97         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3214},
98         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3215},
99         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3237},
100         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x323D},
101         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3241},
102         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3243},
103         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3245},
104         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3247},
105         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3249},
106         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x324A},
107         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x324B},
108         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3250},
109         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3251},
110         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3252},
111         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3253},
112         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3254},
113         {0,}
114 };
115
116 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
117
118 /*  board_id = Subsystem Device ID & Vendor ID
119  *  product = Marketing Name for the board
120  *  access = Address of the struct of function pointers
121  */
122 static struct board_type products[] = {
123         {0x40700E11, "Smart Array 5300", &SA5_access},
124         {0x40800E11, "Smart Array 5i", &SA5B_access},
125         {0x40820E11, "Smart Array 532", &SA5B_access},
126         {0x40830E11, "Smart Array 5312", &SA5B_access},
127         {0x409A0E11, "Smart Array 641", &SA5_access},
128         {0x409B0E11, "Smart Array 642", &SA5_access},
129         {0x409C0E11, "Smart Array 6400", &SA5_access},
130         {0x409D0E11, "Smart Array 6400 EM", &SA5_access},
131         {0x40910E11, "Smart Array 6i", &SA5_access},
132         {0x3225103C, "Smart Array P600", &SA5_access},
133         {0x3235103C, "Smart Array P400i", &SA5_access},
134         {0x3211103C, "Smart Array E200i", &SA5_access},
135         {0x3212103C, "Smart Array E200", &SA5_access},
136         {0x3213103C, "Smart Array E200i", &SA5_access},
137         {0x3214103C, "Smart Array E200i", &SA5_access},
138         {0x3215103C, "Smart Array E200i", &SA5_access},
139         {0x3237103C, "Smart Array E500", &SA5_access},
140 /* controllers below this line are also supported by the hpsa driver. */
141 #define HPSA_BOUNDARY 0x3223103C
142         {0x3223103C, "Smart Array P800", &SA5_access},
143         {0x3234103C, "Smart Array P400", &SA5_access},
144         {0x323D103C, "Smart Array P700m", &SA5_access},
145         {0x3241103C, "Smart Array P212", &SA5_access},
146         {0x3243103C, "Smart Array P410", &SA5_access},
147         {0x3245103C, "Smart Array P410i", &SA5_access},
148         {0x3247103C, "Smart Array P411", &SA5_access},
149         {0x3249103C, "Smart Array P812", &SA5_access},
150         {0x324A103C, "Smart Array P712m", &SA5_access},
151         {0x324B103C, "Smart Array P711m", &SA5_access},
152         {0x3250103C, "Smart Array", &SA5_access},
153         {0x3251103C, "Smart Array", &SA5_access},
154         {0x3252103C, "Smart Array", &SA5_access},
155         {0x3253103C, "Smart Array", &SA5_access},
156         {0x3254103C, "Smart Array", &SA5_access},
157 };
158
159 /* How long to wait (in milliseconds) for board to go into simple mode */
160 #define MAX_CONFIG_WAIT 30000
161 #define MAX_IOCTL_CONFIG_WAIT 1000
162
163 /*define how many times we will try a command because of bus resets */
164 #define MAX_CMD_RETRIES 3
165
166 #define MAX_CTLR        32
167
168 /* Originally cciss driver only supports 8 major numbers */
169 #define MAX_CTLR_ORIG   8
170
171 static ctlr_info_t *hba[MAX_CTLR];
172
173 static struct task_struct *cciss_scan_thread;
174 static DEFINE_MUTEX(scan_mutex);
175 static LIST_HEAD(scan_q);
176
177 static void do_cciss_request(struct request_queue *q);
178 static irqreturn_t do_cciss_intx(int irq, void *dev_id);
179 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id);
180 static int cciss_open(struct block_device *bdev, fmode_t mode);
181 static int cciss_release(struct gendisk *disk, fmode_t mode);
182 static int do_ioctl(struct block_device *bdev, fmode_t mode,
183                     unsigned int cmd, unsigned long arg);
184 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
185                        unsigned int cmd, unsigned long arg);
186 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
187
188 static int cciss_revalidate(struct gendisk *disk);
189 static int rebuild_lun_table(ctlr_info_t *h, int first_time, int via_ioctl);
190 static int deregister_disk(ctlr_info_t *h, int drv_index,
191                            int clear_all, int via_ioctl);
192
193 static void cciss_read_capacity(int ctlr, int logvol,
194                         sector_t *total_size, unsigned int *block_size);
195 static void cciss_read_capacity_16(int ctlr, int logvol,
196                         sector_t *total_size, unsigned int *block_size);
197 static void cciss_geometry_inquiry(int ctlr, int logvol,
198                         sector_t total_size,
199                         unsigned int block_size, InquiryData_struct *inq_buff,
200                                    drive_info_struct *drv);
201 static void __devinit cciss_interrupt_mode(ctlr_info_t *, struct pci_dev *,
202                                            __u32);
203 static void start_io(ctlr_info_t *h);
204 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
205                         __u8 page_code, unsigned char scsi3addr[],
206                         int cmd_type);
207 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
208         int attempt_retry);
209 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c);
210
211 static int add_to_scan_list(struct ctlr_info *h);
212 static int scan_thread(void *data);
213 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c);
214 static void cciss_hba_release(struct device *dev);
215 static void cciss_device_release(struct device *dev);
216 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index);
217 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index);
218 static inline u32 next_command(ctlr_info_t *h);
219
220 /* performant mode helper functions */
221 static void  calc_bucket_map(int *bucket, int num_buckets, int nsgs,
222                                 int *bucket_map);
223 static void cciss_put_controller_into_performant_mode(ctlr_info_t *h);
224
225 #ifdef CONFIG_PROC_FS
226 static void cciss_procinit(int i);
227 #else
228 static void cciss_procinit(int i)
229 {
230 }
231 #endif                          /* CONFIG_PROC_FS */
232
233 #ifdef CONFIG_COMPAT
234 static int cciss_compat_ioctl(struct block_device *, fmode_t,
235                               unsigned, unsigned long);
236 #endif
237
238 static const struct block_device_operations cciss_fops = {
239         .owner = THIS_MODULE,
240         .open = cciss_open,
241         .release = cciss_release,
242         .ioctl = do_ioctl,
243         .getgeo = cciss_getgeo,
244 #ifdef CONFIG_COMPAT
245         .compat_ioctl = cciss_compat_ioctl,
246 #endif
247         .revalidate_disk = cciss_revalidate,
248 };
249
250 /* set_performant_mode: Modify the tag for cciss performant
251  * set bit 0 for pull model, bits 3-1 for block fetch
252  * register number
253  */
254 static void set_performant_mode(ctlr_info_t *h, CommandList_struct *c)
255 {
256         if (likely(h->transMethod == CFGTBL_Trans_Performant))
257                 c->busaddr |= 1 | (h->blockFetchTable[c->Header.SGList] << 1);
258 }
259
260 /*
261  * Enqueuing and dequeuing functions for cmdlists.
262  */
263 static inline void addQ(struct hlist_head *list, CommandList_struct *c)
264 {
265         hlist_add_head(&c->list, list);
266 }
267
268 static inline void removeQ(CommandList_struct *c)
269 {
270         /*
271          * After kexec/dump some commands might still
272          * be in flight, which the firmware will try
273          * to complete. Resetting the firmware doesn't work
274          * with old fw revisions, so we have to mark
275          * them off as 'stale' to prevent the driver from
276          * falling over.
277          */
278         if (WARN_ON(hlist_unhashed(&c->list))) {
279                 c->cmd_type = CMD_MSG_STALE;
280                 return;
281         }
282
283         hlist_del_init(&c->list);
284 }
285
286 static void enqueue_cmd_and_start_io(ctlr_info_t *h,
287         CommandList_struct *c)
288 {
289         unsigned long flags;
290         set_performant_mode(h, c);
291         spin_lock_irqsave(&h->lock, flags);
292         addQ(&h->reqQ, c);
293         h->Qdepth++;
294         start_io(h);
295         spin_unlock_irqrestore(&h->lock, flags);
296 }
297
298 static void cciss_free_sg_chain_blocks(SGDescriptor_struct **cmd_sg_list,
299         int nr_cmds)
300 {
301         int i;
302
303         if (!cmd_sg_list)
304                 return;
305         for (i = 0; i < nr_cmds; i++) {
306                 kfree(cmd_sg_list[i]);
307                 cmd_sg_list[i] = NULL;
308         }
309         kfree(cmd_sg_list);
310 }
311
312 static SGDescriptor_struct **cciss_allocate_sg_chain_blocks(
313         ctlr_info_t *h, int chainsize, int nr_cmds)
314 {
315         int j;
316         SGDescriptor_struct **cmd_sg_list;
317
318         if (chainsize <= 0)
319                 return NULL;
320
321         cmd_sg_list = kmalloc(sizeof(*cmd_sg_list) * nr_cmds, GFP_KERNEL);
322         if (!cmd_sg_list)
323                 return NULL;
324
325         /* Build up chain blocks for each command */
326         for (j = 0; j < nr_cmds; j++) {
327                 /* Need a block of chainsized s/g elements. */
328                 cmd_sg_list[j] = kmalloc((chainsize *
329                         sizeof(*cmd_sg_list[j])), GFP_KERNEL);
330                 if (!cmd_sg_list[j]) {
331                         dev_err(&h->pdev->dev, "Cannot get memory "
332                                 "for s/g chains.\n");
333                         goto clean;
334                 }
335         }
336         return cmd_sg_list;
337 clean:
338         cciss_free_sg_chain_blocks(cmd_sg_list, nr_cmds);
339         return NULL;
340 }
341
342 static void cciss_unmap_sg_chain_block(ctlr_info_t *h, CommandList_struct *c)
343 {
344         SGDescriptor_struct *chain_sg;
345         u64bit temp64;
346
347         if (c->Header.SGTotal <= h->max_cmd_sgentries)
348                 return;
349
350         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
351         temp64.val32.lower = chain_sg->Addr.lower;
352         temp64.val32.upper = chain_sg->Addr.upper;
353         pci_unmap_single(h->pdev, temp64.val, chain_sg->Len, PCI_DMA_TODEVICE);
354 }
355
356 static void cciss_map_sg_chain_block(ctlr_info_t *h, CommandList_struct *c,
357         SGDescriptor_struct *chain_block, int len)
358 {
359         SGDescriptor_struct *chain_sg;
360         u64bit temp64;
361
362         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
363         chain_sg->Ext = CCISS_SG_CHAIN;
364         chain_sg->Len = len;
365         temp64.val = pci_map_single(h->pdev, chain_block, len,
366                                 PCI_DMA_TODEVICE);
367         chain_sg->Addr.lower = temp64.val32.lower;
368         chain_sg->Addr.upper = temp64.val32.upper;
369 }
370
371 #include "cciss_scsi.c"         /* For SCSI tape support */
372
373 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
374         "UNKNOWN"
375 };
376 #define RAID_UNKNOWN (sizeof(raid_label) / sizeof(raid_label[0])-1)
377
378 #ifdef CONFIG_PROC_FS
379
380 /*
381  * Report information about this controller.
382  */
383 #define ENG_GIG 1000000000
384 #define ENG_GIG_FACTOR (ENG_GIG/512)
385 #define ENGAGE_SCSI     "engage scsi"
386
387 static struct proc_dir_entry *proc_cciss;
388
389 static void cciss_seq_show_header(struct seq_file *seq)
390 {
391         ctlr_info_t *h = seq->private;
392
393         seq_printf(seq, "%s: HP %s Controller\n"
394                 "Board ID: 0x%08lx\n"
395                 "Firmware Version: %c%c%c%c\n"
396                 "IRQ: %d\n"
397                 "Logical drives: %d\n"
398                 "Current Q depth: %d\n"
399                 "Current # commands on controller: %d\n"
400                 "Max Q depth since init: %d\n"
401                 "Max # commands on controller since init: %d\n"
402                 "Max SG entries since init: %d\n",
403                 h->devname,
404                 h->product_name,
405                 (unsigned long)h->board_id,
406                 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
407                 h->firm_ver[3], (unsigned int)h->intr[PERF_MODE_INT],
408                 h->num_luns,
409                 h->Qdepth, h->commands_outstanding,
410                 h->maxQsinceinit, h->max_outstanding, h->maxSG);
411
412 #ifdef CONFIG_CISS_SCSI_TAPE
413         cciss_seq_tape_report(seq, h->ctlr);
414 #endif /* CONFIG_CISS_SCSI_TAPE */
415 }
416
417 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
418 {
419         ctlr_info_t *h = seq->private;
420         unsigned ctlr = h->ctlr;
421         unsigned long flags;
422
423         /* prevent displaying bogus info during configuration
424          * or deconfiguration of a logical volume
425          */
426         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
427         if (h->busy_configuring) {
428                 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
429                 return ERR_PTR(-EBUSY);
430         }
431         h->busy_configuring = 1;
432         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
433
434         if (*pos == 0)
435                 cciss_seq_show_header(seq);
436
437         return pos;
438 }
439
440 static int cciss_seq_show(struct seq_file *seq, void *v)
441 {
442         sector_t vol_sz, vol_sz_frac;
443         ctlr_info_t *h = seq->private;
444         unsigned ctlr = h->ctlr;
445         loff_t *pos = v;
446         drive_info_struct *drv = h->drv[*pos];
447
448         if (*pos > h->highest_lun)
449                 return 0;
450
451         if (drv == NULL) /* it's possible for h->drv[] to have holes. */
452                 return 0;
453
454         if (drv->heads == 0)
455                 return 0;
456
457         vol_sz = drv->nr_blocks;
458         vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
459         vol_sz_frac *= 100;
460         sector_div(vol_sz_frac, ENG_GIG_FACTOR);
461
462         if (drv->raid_level < 0 || drv->raid_level > RAID_UNKNOWN)
463                 drv->raid_level = RAID_UNKNOWN;
464         seq_printf(seq, "cciss/c%dd%d:"
465                         "\t%4u.%02uGB\tRAID %s\n",
466                         ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
467                         raid_label[drv->raid_level]);
468         return 0;
469 }
470
471 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
472 {
473         ctlr_info_t *h = seq->private;
474
475         if (*pos > h->highest_lun)
476                 return NULL;
477         *pos += 1;
478
479         return pos;
480 }
481
482 static void cciss_seq_stop(struct seq_file *seq, void *v)
483 {
484         ctlr_info_t *h = seq->private;
485
486         /* Only reset h->busy_configuring if we succeeded in setting
487          * it during cciss_seq_start. */
488         if (v == ERR_PTR(-EBUSY))
489                 return;
490
491         h->busy_configuring = 0;
492 }
493
494 static const struct seq_operations cciss_seq_ops = {
495         .start = cciss_seq_start,
496         .show  = cciss_seq_show,
497         .next  = cciss_seq_next,
498         .stop  = cciss_seq_stop,
499 };
500
501 static int cciss_seq_open(struct inode *inode, struct file *file)
502 {
503         int ret = seq_open(file, &cciss_seq_ops);
504         struct seq_file *seq = file->private_data;
505
506         if (!ret)
507                 seq->private = PDE(inode)->data;
508
509         return ret;
510 }
511
512 static ssize_t
513 cciss_proc_write(struct file *file, const char __user *buf,
514                  size_t length, loff_t *ppos)
515 {
516         int err;
517         char *buffer;
518
519 #ifndef CONFIG_CISS_SCSI_TAPE
520         return -EINVAL;
521 #endif
522
523         if (!buf || length > PAGE_SIZE - 1)
524                 return -EINVAL;
525
526         buffer = (char *)__get_free_page(GFP_KERNEL);
527         if (!buffer)
528                 return -ENOMEM;
529
530         err = -EFAULT;
531         if (copy_from_user(buffer, buf, length))
532                 goto out;
533         buffer[length] = '\0';
534
535 #ifdef CONFIG_CISS_SCSI_TAPE
536         if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
537                 struct seq_file *seq = file->private_data;
538                 ctlr_info_t *h = seq->private;
539
540                 err = cciss_engage_scsi(h->ctlr);
541                 if (err == 0)
542                         err = length;
543         } else
544 #endif /* CONFIG_CISS_SCSI_TAPE */
545                 err = -EINVAL;
546         /* might be nice to have "disengage" too, but it's not
547            safely possible. (only 1 module use count, lock issues.) */
548
549 out:
550         free_page((unsigned long)buffer);
551         return err;
552 }
553
554 static const struct file_operations cciss_proc_fops = {
555         .owner   = THIS_MODULE,
556         .open    = cciss_seq_open,
557         .read    = seq_read,
558         .llseek  = seq_lseek,
559         .release = seq_release,
560         .write   = cciss_proc_write,
561 };
562
563 static void __devinit cciss_procinit(int i)
564 {
565         struct proc_dir_entry *pde;
566
567         if (proc_cciss == NULL)
568                 proc_cciss = proc_mkdir("driver/cciss", NULL);
569         if (!proc_cciss)
570                 return;
571         pde = proc_create_data(hba[i]->devname, S_IWUSR | S_IRUSR | S_IRGRP |
572                                         S_IROTH, proc_cciss,
573                                         &cciss_proc_fops, hba[i]);
574 }
575 #endif                          /* CONFIG_PROC_FS */
576
577 #define MAX_PRODUCT_NAME_LEN 19
578
579 #define to_hba(n) container_of(n, struct ctlr_info, dev)
580 #define to_drv(n) container_of(n, drive_info_struct, dev)
581
582 static ssize_t host_store_rescan(struct device *dev,
583                                  struct device_attribute *attr,
584                                  const char *buf, size_t count)
585 {
586         struct ctlr_info *h = to_hba(dev);
587
588         add_to_scan_list(h);
589         wake_up_process(cciss_scan_thread);
590         wait_for_completion_interruptible(&h->scan_wait);
591
592         return count;
593 }
594 static DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
595
596 static ssize_t dev_show_unique_id(struct device *dev,
597                                  struct device_attribute *attr,
598                                  char *buf)
599 {
600         drive_info_struct *drv = to_drv(dev);
601         struct ctlr_info *h = to_hba(drv->dev.parent);
602         __u8 sn[16];
603         unsigned long flags;
604         int ret = 0;
605
606         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
607         if (h->busy_configuring)
608                 ret = -EBUSY;
609         else
610                 memcpy(sn, drv->serial_no, sizeof(sn));
611         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
612
613         if (ret)
614                 return ret;
615         else
616                 return snprintf(buf, 16 * 2 + 2,
617                                 "%02X%02X%02X%02X%02X%02X%02X%02X"
618                                 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
619                                 sn[0], sn[1], sn[2], sn[3],
620                                 sn[4], sn[5], sn[6], sn[7],
621                                 sn[8], sn[9], sn[10], sn[11],
622                                 sn[12], sn[13], sn[14], sn[15]);
623 }
624 static DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
625
626 static ssize_t dev_show_vendor(struct device *dev,
627                                struct device_attribute *attr,
628                                char *buf)
629 {
630         drive_info_struct *drv = to_drv(dev);
631         struct ctlr_info *h = to_hba(drv->dev.parent);
632         char vendor[VENDOR_LEN + 1];
633         unsigned long flags;
634         int ret = 0;
635
636         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
637         if (h->busy_configuring)
638                 ret = -EBUSY;
639         else
640                 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
641         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
642
643         if (ret)
644                 return ret;
645         else
646                 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
647 }
648 static DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
649
650 static ssize_t dev_show_model(struct device *dev,
651                               struct device_attribute *attr,
652                               char *buf)
653 {
654         drive_info_struct *drv = to_drv(dev);
655         struct ctlr_info *h = to_hba(drv->dev.parent);
656         char model[MODEL_LEN + 1];
657         unsigned long flags;
658         int ret = 0;
659
660         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
661         if (h->busy_configuring)
662                 ret = -EBUSY;
663         else
664                 memcpy(model, drv->model, MODEL_LEN + 1);
665         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
666
667         if (ret)
668                 return ret;
669         else
670                 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
671 }
672 static DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
673
674 static ssize_t dev_show_rev(struct device *dev,
675                             struct device_attribute *attr,
676                             char *buf)
677 {
678         drive_info_struct *drv = to_drv(dev);
679         struct ctlr_info *h = to_hba(drv->dev.parent);
680         char rev[REV_LEN + 1];
681         unsigned long flags;
682         int ret = 0;
683
684         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
685         if (h->busy_configuring)
686                 ret = -EBUSY;
687         else
688                 memcpy(rev, drv->rev, REV_LEN + 1);
689         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
690
691         if (ret)
692                 return ret;
693         else
694                 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
695 }
696 static DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
697
698 static ssize_t cciss_show_lunid(struct device *dev,
699                                 struct device_attribute *attr, char *buf)
700 {
701         drive_info_struct *drv = to_drv(dev);
702         struct ctlr_info *h = to_hba(drv->dev.parent);
703         unsigned long flags;
704         unsigned char lunid[8];
705
706         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
707         if (h->busy_configuring) {
708                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
709                 return -EBUSY;
710         }
711         if (!drv->heads) {
712                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
713                 return -ENOTTY;
714         }
715         memcpy(lunid, drv->LunID, sizeof(lunid));
716         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
717         return snprintf(buf, 20, "0x%02x%02x%02x%02x%02x%02x%02x%02x\n",
718                 lunid[0], lunid[1], lunid[2], lunid[3],
719                 lunid[4], lunid[5], lunid[6], lunid[7]);
720 }
721 static DEVICE_ATTR(lunid, S_IRUGO, cciss_show_lunid, NULL);
722
723 static ssize_t cciss_show_raid_level(struct device *dev,
724                                      struct device_attribute *attr, char *buf)
725 {
726         drive_info_struct *drv = to_drv(dev);
727         struct ctlr_info *h = to_hba(drv->dev.parent);
728         int raid;
729         unsigned long flags;
730
731         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
732         if (h->busy_configuring) {
733                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
734                 return -EBUSY;
735         }
736         raid = drv->raid_level;
737         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
738         if (raid < 0 || raid > RAID_UNKNOWN)
739                 raid = RAID_UNKNOWN;
740
741         return snprintf(buf, strlen(raid_label[raid]) + 7, "RAID %s\n",
742                         raid_label[raid]);
743 }
744 static DEVICE_ATTR(raid_level, S_IRUGO, cciss_show_raid_level, NULL);
745
746 static ssize_t cciss_show_usage_count(struct device *dev,
747                                       struct device_attribute *attr, char *buf)
748 {
749         drive_info_struct *drv = to_drv(dev);
750         struct ctlr_info *h = to_hba(drv->dev.parent);
751         unsigned long flags;
752         int count;
753
754         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
755         if (h->busy_configuring) {
756                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
757                 return -EBUSY;
758         }
759         count = drv->usage_count;
760         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
761         return snprintf(buf, 20, "%d\n", count);
762 }
763 static DEVICE_ATTR(usage_count, S_IRUGO, cciss_show_usage_count, NULL);
764
765 static struct attribute *cciss_host_attrs[] = {
766         &dev_attr_rescan.attr,
767         NULL
768 };
769
770 static struct attribute_group cciss_host_attr_group = {
771         .attrs = cciss_host_attrs,
772 };
773
774 static const struct attribute_group *cciss_host_attr_groups[] = {
775         &cciss_host_attr_group,
776         NULL
777 };
778
779 static struct device_type cciss_host_type = {
780         .name           = "cciss_host",
781         .groups         = cciss_host_attr_groups,
782         .release        = cciss_hba_release,
783 };
784
785 static struct attribute *cciss_dev_attrs[] = {
786         &dev_attr_unique_id.attr,
787         &dev_attr_model.attr,
788         &dev_attr_vendor.attr,
789         &dev_attr_rev.attr,
790         &dev_attr_lunid.attr,
791         &dev_attr_raid_level.attr,
792         &dev_attr_usage_count.attr,
793         NULL
794 };
795
796 static struct attribute_group cciss_dev_attr_group = {
797         .attrs = cciss_dev_attrs,
798 };
799
800 static const struct attribute_group *cciss_dev_attr_groups[] = {
801         &cciss_dev_attr_group,
802         NULL
803 };
804
805 static struct device_type cciss_dev_type = {
806         .name           = "cciss_device",
807         .groups         = cciss_dev_attr_groups,
808         .release        = cciss_device_release,
809 };
810
811 static struct bus_type cciss_bus_type = {
812         .name           = "cciss",
813 };
814
815 /*
816  * cciss_hba_release is called when the reference count
817  * of h->dev goes to zero.
818  */
819 static void cciss_hba_release(struct device *dev)
820 {
821         /*
822          * nothing to do, but need this to avoid a warning
823          * about not having a release handler from lib/kref.c.
824          */
825 }
826
827 /*
828  * Initialize sysfs entry for each controller.  This sets up and registers
829  * the 'cciss#' directory for each individual controller under
830  * /sys/bus/pci/devices/<dev>/.
831  */
832 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
833 {
834         device_initialize(&h->dev);
835         h->dev.type = &cciss_host_type;
836         h->dev.bus = &cciss_bus_type;
837         dev_set_name(&h->dev, "%s", h->devname);
838         h->dev.parent = &h->pdev->dev;
839
840         return device_add(&h->dev);
841 }
842
843 /*
844  * Remove sysfs entries for an hba.
845  */
846 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
847 {
848         device_del(&h->dev);
849         put_device(&h->dev); /* final put. */
850 }
851
852 /* cciss_device_release is called when the reference count
853  * of h->drv[x]dev goes to zero.
854  */
855 static void cciss_device_release(struct device *dev)
856 {
857         drive_info_struct *drv = to_drv(dev);
858         kfree(drv);
859 }
860
861 /*
862  * Initialize sysfs for each logical drive.  This sets up and registers
863  * the 'c#d#' directory for each individual logical drive under
864  * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
865  * /sys/block/cciss!c#d# to this entry.
866  */
867 static long cciss_create_ld_sysfs_entry(struct ctlr_info *h,
868                                        int drv_index)
869 {
870         struct device *dev;
871
872         if (h->drv[drv_index]->device_initialized)
873                 return 0;
874
875         dev = &h->drv[drv_index]->dev;
876         device_initialize(dev);
877         dev->type = &cciss_dev_type;
878         dev->bus = &cciss_bus_type;
879         dev_set_name(dev, "c%dd%d", h->ctlr, drv_index);
880         dev->parent = &h->dev;
881         h->drv[drv_index]->device_initialized = 1;
882         return device_add(dev);
883 }
884
885 /*
886  * Remove sysfs entries for a logical drive.
887  */
888 static void cciss_destroy_ld_sysfs_entry(struct ctlr_info *h, int drv_index,
889         int ctlr_exiting)
890 {
891         struct device *dev = &h->drv[drv_index]->dev;
892
893         /* special case for c*d0, we only destroy it on controller exit */
894         if (drv_index == 0 && !ctlr_exiting)
895                 return;
896
897         device_del(dev);
898         put_device(dev); /* the "final" put. */
899         h->drv[drv_index] = NULL;
900 }
901
902 /*
903  * For operations that cannot sleep, a command block is allocated at init,
904  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
905  * which ones are free or in use.  For operations that can wait for kmalloc
906  * to possible sleep, this routine can be called with get_from_pool set to 0.
907  * cmd_free() MUST be called with a got_from_pool set to 0 if cmd_alloc was.
908  */
909 static CommandList_struct *cmd_alloc(ctlr_info_t *h, int get_from_pool)
910 {
911         CommandList_struct *c;
912         int i;
913         u64bit temp64;
914         dma_addr_t cmd_dma_handle, err_dma_handle;
915
916         if (!get_from_pool) {
917                 c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
918                         sizeof(CommandList_struct), &cmd_dma_handle);
919                 if (c == NULL)
920                         return NULL;
921                 memset(c, 0, sizeof(CommandList_struct));
922
923                 c->cmdindex = -1;
924
925                 c->err_info = (ErrorInfo_struct *)
926                     pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
927                             &err_dma_handle);
928
929                 if (c->err_info == NULL) {
930                         pci_free_consistent(h->pdev,
931                                 sizeof(CommandList_struct), c, cmd_dma_handle);
932                         return NULL;
933                 }
934                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
935         } else {                /* get it out of the controllers pool */
936
937                 do {
938                         i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
939                         if (i == h->nr_cmds)
940                                 return NULL;
941                 } while (test_and_set_bit
942                          (i & (BITS_PER_LONG - 1),
943                           h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
944 #ifdef CCISS_DEBUG
945                 printk(KERN_DEBUG "cciss: using command buffer %d\n", i);
946 #endif
947                 c = h->cmd_pool + i;
948                 memset(c, 0, sizeof(CommandList_struct));
949                 cmd_dma_handle = h->cmd_pool_dhandle
950                     + i * sizeof(CommandList_struct);
951                 c->err_info = h->errinfo_pool + i;
952                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
953                 err_dma_handle = h->errinfo_pool_dhandle
954                     + i * sizeof(ErrorInfo_struct);
955                 h->nr_allocs++;
956
957                 c->cmdindex = i;
958         }
959
960         INIT_HLIST_NODE(&c->list);
961         c->busaddr = (__u32) cmd_dma_handle;
962         temp64.val = (__u64) err_dma_handle;
963         c->ErrDesc.Addr.lower = temp64.val32.lower;
964         c->ErrDesc.Addr.upper = temp64.val32.upper;
965         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
966
967         c->ctlr = h->ctlr;
968         return c;
969 }
970
971 /*
972  * Frees a command block that was previously allocated with cmd_alloc().
973  */
974 static void cmd_free(ctlr_info_t *h, CommandList_struct *c, int got_from_pool)
975 {
976         int i;
977         u64bit temp64;
978
979         if (!got_from_pool) {
980                 temp64.val32.lower = c->ErrDesc.Addr.lower;
981                 temp64.val32.upper = c->ErrDesc.Addr.upper;
982                 pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
983                                     c->err_info, (dma_addr_t) temp64.val);
984                 pci_free_consistent(h->pdev, sizeof(CommandList_struct),
985                                     c, (dma_addr_t) c->busaddr);
986         } else {
987                 i = c - h->cmd_pool;
988                 clear_bit(i & (BITS_PER_LONG - 1),
989                           h->cmd_pool_bits + (i / BITS_PER_LONG));
990                 h->nr_frees++;
991         }
992 }
993
994 static inline ctlr_info_t *get_host(struct gendisk *disk)
995 {
996         return disk->queue->queuedata;
997 }
998
999 static inline drive_info_struct *get_drv(struct gendisk *disk)
1000 {
1001         return disk->private_data;
1002 }
1003
1004 /*
1005  * Open.  Make sure the device is really there.
1006  */
1007 static int cciss_open(struct block_device *bdev, fmode_t mode)
1008 {
1009         ctlr_info_t *host = get_host(bdev->bd_disk);
1010         drive_info_struct *drv = get_drv(bdev->bd_disk);
1011
1012 #ifdef CCISS_DEBUG
1013         printk(KERN_DEBUG "cciss_open %s\n", bdev->bd_disk->disk_name);
1014 #endif                          /* CCISS_DEBUG */
1015
1016         if (drv->busy_configuring)
1017                 return -EBUSY;
1018         /*
1019          * Root is allowed to open raw volume zero even if it's not configured
1020          * so array config can still work. Root is also allowed to open any
1021          * volume that has a LUN ID, so it can issue IOCTL to reread the
1022          * disk information.  I don't think I really like this
1023          * but I'm already using way to many device nodes to claim another one
1024          * for "raw controller".
1025          */
1026         if (drv->heads == 0) {
1027                 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
1028                         /* if not node 0 make sure it is a partition = 0 */
1029                         if (MINOR(bdev->bd_dev) & 0x0f) {
1030                                 return -ENXIO;
1031                                 /* if it is, make sure we have a LUN ID */
1032                         } else if (memcmp(drv->LunID, CTLR_LUNID,
1033                                 sizeof(drv->LunID))) {
1034                                 return -ENXIO;
1035                         }
1036                 }
1037                 if (!capable(CAP_SYS_ADMIN))
1038                         return -EPERM;
1039         }
1040         drv->usage_count++;
1041         host->usage_count++;
1042         return 0;
1043 }
1044
1045 /*
1046  * Close.  Sync first.
1047  */
1048 static int cciss_release(struct gendisk *disk, fmode_t mode)
1049 {
1050         ctlr_info_t *host = get_host(disk);
1051         drive_info_struct *drv = get_drv(disk);
1052
1053 #ifdef CCISS_DEBUG
1054         printk(KERN_DEBUG "cciss_release %s\n", disk->disk_name);
1055 #endif                          /* CCISS_DEBUG */
1056
1057         drv->usage_count--;
1058         host->usage_count--;
1059         return 0;
1060 }
1061
1062 static int do_ioctl(struct block_device *bdev, fmode_t mode,
1063                     unsigned cmd, unsigned long arg)
1064 {
1065         int ret;
1066         lock_kernel();
1067         ret = cciss_ioctl(bdev, mode, cmd, arg);
1068         unlock_kernel();
1069         return ret;
1070 }
1071
1072 #ifdef CONFIG_COMPAT
1073
1074 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1075                                   unsigned cmd, unsigned long arg);
1076 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1077                                       unsigned cmd, unsigned long arg);
1078
1079 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
1080                               unsigned cmd, unsigned long arg)
1081 {
1082         switch (cmd) {
1083         case CCISS_GETPCIINFO:
1084         case CCISS_GETINTINFO:
1085         case CCISS_SETINTINFO:
1086         case CCISS_GETNODENAME:
1087         case CCISS_SETNODENAME:
1088         case CCISS_GETHEARTBEAT:
1089         case CCISS_GETBUSTYPES:
1090         case CCISS_GETFIRMVER:
1091         case CCISS_GETDRIVVER:
1092         case CCISS_REVALIDVOLS:
1093         case CCISS_DEREGDISK:
1094         case CCISS_REGNEWDISK:
1095         case CCISS_REGNEWD:
1096         case CCISS_RESCANDISK:
1097         case CCISS_GETLUNINFO:
1098                 return do_ioctl(bdev, mode, cmd, arg);
1099
1100         case CCISS_PASSTHRU32:
1101                 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
1102         case CCISS_BIG_PASSTHRU32:
1103                 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
1104
1105         default:
1106                 return -ENOIOCTLCMD;
1107         }
1108 }
1109
1110 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1111                                   unsigned cmd, unsigned long arg)
1112 {
1113         IOCTL32_Command_struct __user *arg32 =
1114             (IOCTL32_Command_struct __user *) arg;
1115         IOCTL_Command_struct arg64;
1116         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
1117         int err;
1118         u32 cp;
1119
1120         err = 0;
1121         err |=
1122             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1123                            sizeof(arg64.LUN_info));
1124         err |=
1125             copy_from_user(&arg64.Request, &arg32->Request,
1126                            sizeof(arg64.Request));
1127         err |=
1128             copy_from_user(&arg64.error_info, &arg32->error_info,
1129                            sizeof(arg64.error_info));
1130         err |= get_user(arg64.buf_size, &arg32->buf_size);
1131         err |= get_user(cp, &arg32->buf);
1132         arg64.buf = compat_ptr(cp);
1133         err |= copy_to_user(p, &arg64, sizeof(arg64));
1134
1135         if (err)
1136                 return -EFAULT;
1137
1138         err = do_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
1139         if (err)
1140                 return err;
1141         err |=
1142             copy_in_user(&arg32->error_info, &p->error_info,
1143                          sizeof(arg32->error_info));
1144         if (err)
1145                 return -EFAULT;
1146         return err;
1147 }
1148
1149 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1150                                       unsigned cmd, unsigned long arg)
1151 {
1152         BIG_IOCTL32_Command_struct __user *arg32 =
1153             (BIG_IOCTL32_Command_struct __user *) arg;
1154         BIG_IOCTL_Command_struct arg64;
1155         BIG_IOCTL_Command_struct __user *p =
1156             compat_alloc_user_space(sizeof(arg64));
1157         int err;
1158         u32 cp;
1159
1160         err = 0;
1161         err |=
1162             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1163                            sizeof(arg64.LUN_info));
1164         err |=
1165             copy_from_user(&arg64.Request, &arg32->Request,
1166                            sizeof(arg64.Request));
1167         err |=
1168             copy_from_user(&arg64.error_info, &arg32->error_info,
1169                            sizeof(arg64.error_info));
1170         err |= get_user(arg64.buf_size, &arg32->buf_size);
1171         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1172         err |= get_user(cp, &arg32->buf);
1173         arg64.buf = compat_ptr(cp);
1174         err |= copy_to_user(p, &arg64, sizeof(arg64));
1175
1176         if (err)
1177                 return -EFAULT;
1178
1179         err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1180         if (err)
1181                 return err;
1182         err |=
1183             copy_in_user(&arg32->error_info, &p->error_info,
1184                          sizeof(arg32->error_info));
1185         if (err)
1186                 return -EFAULT;
1187         return err;
1188 }
1189 #endif
1190
1191 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1192 {
1193         drive_info_struct *drv = get_drv(bdev->bd_disk);
1194
1195         if (!drv->cylinders)
1196                 return -ENXIO;
1197
1198         geo->heads = drv->heads;
1199         geo->sectors = drv->sectors;
1200         geo->cylinders = drv->cylinders;
1201         return 0;
1202 }
1203
1204 static void check_ioctl_unit_attention(ctlr_info_t *host, CommandList_struct *c)
1205 {
1206         if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1207                         c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1208                 (void)check_for_unit_attention(host, c);
1209 }
1210 /*
1211  * ioctl
1212  */
1213 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1214                        unsigned int cmd, unsigned long arg)
1215 {
1216         struct gendisk *disk = bdev->bd_disk;
1217         ctlr_info_t *host = get_host(disk);
1218         drive_info_struct *drv = get_drv(disk);
1219         int ctlr = host->ctlr;
1220         void __user *argp = (void __user *)arg;
1221
1222 #ifdef CCISS_DEBUG
1223         printk(KERN_DEBUG "cciss_ioctl: Called with cmd=%x %lx\n", cmd, arg);
1224 #endif                          /* CCISS_DEBUG */
1225
1226         switch (cmd) {
1227         case CCISS_GETPCIINFO:
1228                 {
1229                         cciss_pci_info_struct pciinfo;
1230
1231                         if (!arg)
1232                                 return -EINVAL;
1233                         pciinfo.domain = pci_domain_nr(host->pdev->bus);
1234                         pciinfo.bus = host->pdev->bus->number;
1235                         pciinfo.dev_fn = host->pdev->devfn;
1236                         pciinfo.board_id = host->board_id;
1237                         if (copy_to_user
1238                             (argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1239                                 return -EFAULT;
1240                         return 0;
1241                 }
1242         case CCISS_GETINTINFO:
1243                 {
1244                         cciss_coalint_struct intinfo;
1245                         if (!arg)
1246                                 return -EINVAL;
1247                         intinfo.delay =
1248                             readl(&host->cfgtable->HostWrite.CoalIntDelay);
1249                         intinfo.count =
1250                             readl(&host->cfgtable->HostWrite.CoalIntCount);
1251                         if (copy_to_user
1252                             (argp, &intinfo, sizeof(cciss_coalint_struct)))
1253                                 return -EFAULT;
1254                         return 0;
1255                 }
1256         case CCISS_SETINTINFO:
1257                 {
1258                         cciss_coalint_struct intinfo;
1259                         unsigned long flags;
1260                         int i;
1261
1262                         if (!arg)
1263                                 return -EINVAL;
1264                         if (!capable(CAP_SYS_ADMIN))
1265                                 return -EPERM;
1266                         if (copy_from_user
1267                             (&intinfo, argp, sizeof(cciss_coalint_struct)))
1268                                 return -EFAULT;
1269                         if ((intinfo.delay == 0) && (intinfo.count == 0))
1270                         {
1271 //                      printk("cciss_ioctl: delay and count cannot be 0\n");
1272                                 return -EINVAL;
1273                         }
1274                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1275                         /* Update the field, and then ring the doorbell */
1276                         writel(intinfo.delay,
1277                                &(host->cfgtable->HostWrite.CoalIntDelay));
1278                         writel(intinfo.count,
1279                                &(host->cfgtable->HostWrite.CoalIntCount));
1280                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1281
1282                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1283                                 if (!(readl(host->vaddr + SA5_DOORBELL)
1284                                       & CFGTBL_ChangeReq))
1285                                         break;
1286                                 /* delay and try again */
1287                                 udelay(1000);
1288                         }
1289                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1290                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1291                                 return -EAGAIN;
1292                         return 0;
1293                 }
1294         case CCISS_GETNODENAME:
1295                 {
1296                         NodeName_type NodeName;
1297                         int i;
1298
1299                         if (!arg)
1300                                 return -EINVAL;
1301                         for (i = 0; i < 16; i++)
1302                                 NodeName[i] =
1303                                     readb(&host->cfgtable->ServerName[i]);
1304                         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1305                                 return -EFAULT;
1306                         return 0;
1307                 }
1308         case CCISS_SETNODENAME:
1309                 {
1310                         NodeName_type NodeName;
1311                         unsigned long flags;
1312                         int i;
1313
1314                         if (!arg)
1315                                 return -EINVAL;
1316                         if (!capable(CAP_SYS_ADMIN))
1317                                 return -EPERM;
1318
1319                         if (copy_from_user
1320                             (NodeName, argp, sizeof(NodeName_type)))
1321                                 return -EFAULT;
1322
1323                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1324
1325                         /* Update the field, and then ring the doorbell */
1326                         for (i = 0; i < 16; i++)
1327                                 writeb(NodeName[i],
1328                                        &host->cfgtable->ServerName[i]);
1329
1330                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1331
1332                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1333                                 if (!(readl(host->vaddr + SA5_DOORBELL)
1334                                       & CFGTBL_ChangeReq))
1335                                         break;
1336                                 /* delay and try again */
1337                                 udelay(1000);
1338                         }
1339                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1340                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1341                                 return -EAGAIN;
1342                         return 0;
1343                 }
1344
1345         case CCISS_GETHEARTBEAT:
1346                 {
1347                         Heartbeat_type heartbeat;
1348
1349                         if (!arg)
1350                                 return -EINVAL;
1351                         heartbeat = readl(&host->cfgtable->HeartBeat);
1352                         if (copy_to_user
1353                             (argp, &heartbeat, sizeof(Heartbeat_type)))
1354                                 return -EFAULT;
1355                         return 0;
1356                 }
1357         case CCISS_GETBUSTYPES:
1358                 {
1359                         BusTypes_type BusTypes;
1360
1361                         if (!arg)
1362                                 return -EINVAL;
1363                         BusTypes = readl(&host->cfgtable->BusTypes);
1364                         if (copy_to_user
1365                             (argp, &BusTypes, sizeof(BusTypes_type)))
1366                                 return -EFAULT;
1367                         return 0;
1368                 }
1369         case CCISS_GETFIRMVER:
1370                 {
1371                         FirmwareVer_type firmware;
1372
1373                         if (!arg)
1374                                 return -EINVAL;
1375                         memcpy(firmware, host->firm_ver, 4);
1376
1377                         if (copy_to_user
1378                             (argp, firmware, sizeof(FirmwareVer_type)))
1379                                 return -EFAULT;
1380                         return 0;
1381                 }
1382         case CCISS_GETDRIVVER:
1383                 {
1384                         DriverVer_type DriverVer = DRIVER_VERSION;
1385
1386                         if (!arg)
1387                                 return -EINVAL;
1388
1389                         if (copy_to_user
1390                             (argp, &DriverVer, sizeof(DriverVer_type)))
1391                                 return -EFAULT;
1392                         return 0;
1393                 }
1394
1395         case CCISS_DEREGDISK:
1396         case CCISS_REGNEWD:
1397         case CCISS_REVALIDVOLS:
1398                 return rebuild_lun_table(host, 0, 1);
1399
1400         case CCISS_GETLUNINFO:{
1401                         LogvolInfo_struct luninfo;
1402
1403                         memcpy(&luninfo.LunID, drv->LunID,
1404                                 sizeof(luninfo.LunID));
1405                         luninfo.num_opens = drv->usage_count;
1406                         luninfo.num_parts = 0;
1407                         if (copy_to_user(argp, &luninfo,
1408                                          sizeof(LogvolInfo_struct)))
1409                                 return -EFAULT;
1410                         return 0;
1411                 }
1412         case CCISS_PASSTHRU:
1413                 {
1414                         IOCTL_Command_struct iocommand;
1415                         CommandList_struct *c;
1416                         char *buff = NULL;
1417                         u64bit temp64;
1418                         DECLARE_COMPLETION_ONSTACK(wait);
1419
1420                         if (!arg)
1421                                 return -EINVAL;
1422
1423                         if (!capable(CAP_SYS_RAWIO))
1424                                 return -EPERM;
1425
1426                         if (copy_from_user
1427                             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1428                                 return -EFAULT;
1429                         if ((iocommand.buf_size < 1) &&
1430                             (iocommand.Request.Type.Direction != XFER_NONE)) {
1431                                 return -EINVAL;
1432                         }
1433 #if 0                           /* 'buf_size' member is 16-bits, and always smaller than kmalloc limit */
1434                         /* Check kmalloc limits */
1435                         if (iocommand.buf_size > 128000)
1436                                 return -EINVAL;
1437 #endif
1438                         if (iocommand.buf_size > 0) {
1439                                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1440                                 if (buff == NULL)
1441                                         return -EFAULT;
1442                         }
1443                         if (iocommand.Request.Type.Direction == XFER_WRITE) {
1444                                 /* Copy the data into the buffer we created */
1445                                 if (copy_from_user
1446                                     (buff, iocommand.buf, iocommand.buf_size)) {
1447                                         kfree(buff);
1448                                         return -EFAULT;
1449                                 }
1450                         } else {
1451                                 memset(buff, 0, iocommand.buf_size);
1452                         }
1453                         if ((c = cmd_alloc(host, 0)) == NULL) {
1454                                 kfree(buff);
1455                                 return -ENOMEM;
1456                         }
1457                         /* Fill in the command type */
1458                         c->cmd_type = CMD_IOCTL_PEND;
1459                         /* Fill in Command Header */
1460                         c->Header.ReplyQueue = 0;   /* unused in simple mode */
1461                         if (iocommand.buf_size > 0) /* buffer to fill */
1462                         {
1463                                 c->Header.SGList = 1;
1464                                 c->Header.SGTotal = 1;
1465                         } else /* no buffers to fill */
1466                         {
1467                                 c->Header.SGList = 0;
1468                                 c->Header.SGTotal = 0;
1469                         }
1470                         c->Header.LUN = iocommand.LUN_info;
1471                         /* use the kernel address the cmd block for tag */
1472                         c->Header.Tag.lower = c->busaddr;
1473
1474                         /* Fill in Request block */
1475                         c->Request = iocommand.Request;
1476
1477                         /* Fill in the scatter gather information */
1478                         if (iocommand.buf_size > 0) {
1479                                 temp64.val = pci_map_single(host->pdev, buff,
1480                                         iocommand.buf_size,
1481                                         PCI_DMA_BIDIRECTIONAL);
1482                                 c->SG[0].Addr.lower = temp64.val32.lower;
1483                                 c->SG[0].Addr.upper = temp64.val32.upper;
1484                                 c->SG[0].Len = iocommand.buf_size;
1485                                 c->SG[0].Ext = 0;  /* we are not chaining */
1486                         }
1487                         c->waiting = &wait;
1488
1489                         enqueue_cmd_and_start_io(host, c);
1490                         wait_for_completion(&wait);
1491
1492                         /* unlock the buffers from DMA */
1493                         temp64.val32.lower = c->SG[0].Addr.lower;
1494                         temp64.val32.upper = c->SG[0].Addr.upper;
1495                         pci_unmap_single(host->pdev, (dma_addr_t) temp64.val,
1496                                          iocommand.buf_size,
1497                                          PCI_DMA_BIDIRECTIONAL);
1498
1499                         check_ioctl_unit_attention(host, c);
1500
1501                         /* Copy the error information out */
1502                         iocommand.error_info = *(c->err_info);
1503                         if (copy_to_user
1504                             (argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1505                                 kfree(buff);
1506                                 cmd_free(host, c, 0);
1507                                 return -EFAULT;
1508                         }
1509
1510                         if (iocommand.Request.Type.Direction == XFER_READ) {
1511                                 /* Copy the data out of the buffer we created */
1512                                 if (copy_to_user
1513                                     (iocommand.buf, buff, iocommand.buf_size)) {
1514                                         kfree(buff);
1515                                         cmd_free(host, c, 0);
1516                                         return -EFAULT;
1517                                 }
1518                         }
1519                         kfree(buff);
1520                         cmd_free(host, c, 0);
1521                         return 0;
1522                 }
1523         case CCISS_BIG_PASSTHRU:{
1524                         BIG_IOCTL_Command_struct *ioc;
1525                         CommandList_struct *c;
1526                         unsigned char **buff = NULL;
1527                         int *buff_size = NULL;
1528                         u64bit temp64;
1529                         BYTE sg_used = 0;
1530                         int status = 0;
1531                         int i;
1532                         DECLARE_COMPLETION_ONSTACK(wait);
1533                         __u32 left;
1534                         __u32 sz;
1535                         BYTE __user *data_ptr;
1536
1537                         if (!arg)
1538                                 return -EINVAL;
1539                         if (!capable(CAP_SYS_RAWIO))
1540                                 return -EPERM;
1541                         ioc = (BIG_IOCTL_Command_struct *)
1542                             kmalloc(sizeof(*ioc), GFP_KERNEL);
1543                         if (!ioc) {
1544                                 status = -ENOMEM;
1545                                 goto cleanup1;
1546                         }
1547                         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1548                                 status = -EFAULT;
1549                                 goto cleanup1;
1550                         }
1551                         if ((ioc->buf_size < 1) &&
1552                             (ioc->Request.Type.Direction != XFER_NONE)) {
1553                                 status = -EINVAL;
1554                                 goto cleanup1;
1555                         }
1556                         /* Check kmalloc limits  using all SGs */
1557                         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1558                                 status = -EINVAL;
1559                                 goto cleanup1;
1560                         }
1561                         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1562                                 status = -EINVAL;
1563                                 goto cleanup1;
1564                         }
1565                         buff =
1566                             kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1567                         if (!buff) {
1568                                 status = -ENOMEM;
1569                                 goto cleanup1;
1570                         }
1571                         buff_size = kmalloc(MAXSGENTRIES * sizeof(int),
1572                                                    GFP_KERNEL);
1573                         if (!buff_size) {
1574                                 status = -ENOMEM;
1575                                 goto cleanup1;
1576                         }
1577                         left = ioc->buf_size;
1578                         data_ptr = ioc->buf;
1579                         while (left) {
1580                                 sz = (left >
1581                                       ioc->malloc_size) ? ioc->
1582                                     malloc_size : left;
1583                                 buff_size[sg_used] = sz;
1584                                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1585                                 if (buff[sg_used] == NULL) {
1586                                         status = -ENOMEM;
1587                                         goto cleanup1;
1588                                 }
1589                                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1590                                         if (copy_from_user
1591                                             (buff[sg_used], data_ptr, sz)) {
1592                                                 status = -EFAULT;
1593                                                 goto cleanup1;
1594                                         }
1595                                 } else {
1596                                         memset(buff[sg_used], 0, sz);
1597                                 }
1598                                 left -= sz;
1599                                 data_ptr += sz;
1600                                 sg_used++;
1601                         }
1602                         if ((c = cmd_alloc(host, 0)) == NULL) {
1603                                 status = -ENOMEM;
1604                                 goto cleanup1;
1605                         }
1606                         c->cmd_type = CMD_IOCTL_PEND;
1607                         c->Header.ReplyQueue = 0;
1608
1609                         if (ioc->buf_size > 0) {
1610                                 c->Header.SGList = sg_used;
1611                                 c->Header.SGTotal = sg_used;
1612                         } else {
1613                                 c->Header.SGList = 0;
1614                                 c->Header.SGTotal = 0;
1615                         }
1616                         c->Header.LUN = ioc->LUN_info;
1617                         c->Header.Tag.lower = c->busaddr;
1618
1619                         c->Request = ioc->Request;
1620                         if (ioc->buf_size > 0) {
1621                                 for (i = 0; i < sg_used; i++) {
1622                                         temp64.val =
1623                                             pci_map_single(host->pdev, buff[i],
1624                                                     buff_size[i],
1625                                                     PCI_DMA_BIDIRECTIONAL);
1626                                         c->SG[i].Addr.lower =
1627                                             temp64.val32.lower;
1628                                         c->SG[i].Addr.upper =
1629                                             temp64.val32.upper;
1630                                         c->SG[i].Len = buff_size[i];
1631                                         c->SG[i].Ext = 0;       /* we are not chaining */
1632                                 }
1633                         }
1634                         c->waiting = &wait;
1635                         enqueue_cmd_and_start_io(host, c);
1636                         wait_for_completion(&wait);
1637                         /* unlock the buffers from DMA */
1638                         for (i = 0; i < sg_used; i++) {
1639                                 temp64.val32.lower = c->SG[i].Addr.lower;
1640                                 temp64.val32.upper = c->SG[i].Addr.upper;
1641                                 pci_unmap_single(host->pdev,
1642                                         (dma_addr_t) temp64.val, buff_size[i],
1643                                         PCI_DMA_BIDIRECTIONAL);
1644                         }
1645                         check_ioctl_unit_attention(host, c);
1646                         /* Copy the error information out */
1647                         ioc->error_info = *(c->err_info);
1648                         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1649                                 cmd_free(host, c, 0);
1650                                 status = -EFAULT;
1651                                 goto cleanup1;
1652                         }
1653                         if (ioc->Request.Type.Direction == XFER_READ) {
1654                                 /* Copy the data out of the buffer we created */
1655                                 BYTE __user *ptr = ioc->buf;
1656                                 for (i = 0; i < sg_used; i++) {
1657                                         if (copy_to_user
1658                                             (ptr, buff[i], buff_size[i])) {
1659                                                 cmd_free(host, c, 0);
1660                                                 status = -EFAULT;
1661                                                 goto cleanup1;
1662                                         }
1663                                         ptr += buff_size[i];
1664                                 }
1665                         }
1666                         cmd_free(host, c, 0);
1667                         status = 0;
1668                       cleanup1:
1669                         if (buff) {
1670                                 for (i = 0; i < sg_used; i++)
1671                                         kfree(buff[i]);
1672                                 kfree(buff);
1673                         }
1674                         kfree(buff_size);
1675                         kfree(ioc);
1676                         return status;
1677                 }
1678
1679         /* scsi_cmd_ioctl handles these, below, though some are not */
1680         /* very meaningful for cciss.  SG_IO is the main one people want. */
1681
1682         case SG_GET_VERSION_NUM:
1683         case SG_SET_TIMEOUT:
1684         case SG_GET_TIMEOUT:
1685         case SG_GET_RESERVED_SIZE:
1686         case SG_SET_RESERVED_SIZE:
1687         case SG_EMULATED_HOST:
1688         case SG_IO:
1689         case SCSI_IOCTL_SEND_COMMAND:
1690                 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
1691
1692         /* scsi_cmd_ioctl would normally handle these, below, but */
1693         /* they aren't a good fit for cciss, as CD-ROMs are */
1694         /* not supported, and we don't have any bus/target/lun */
1695         /* which we present to the kernel. */
1696
1697         case CDROM_SEND_PACKET:
1698         case CDROMCLOSETRAY:
1699         case CDROMEJECT:
1700         case SCSI_IOCTL_GET_IDLUN:
1701         case SCSI_IOCTL_GET_BUS_NUMBER:
1702         default:
1703                 return -ENOTTY;
1704         }
1705 }
1706
1707 static void cciss_check_queues(ctlr_info_t *h)
1708 {
1709         int start_queue = h->next_to_run;
1710         int i;
1711
1712         /* check to see if we have maxed out the number of commands that can
1713          * be placed on the queue.  If so then exit.  We do this check here
1714          * in case the interrupt we serviced was from an ioctl and did not
1715          * free any new commands.
1716          */
1717         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1718                 return;
1719
1720         /* We have room on the queue for more commands.  Now we need to queue
1721          * them up.  We will also keep track of the next queue to run so
1722          * that every queue gets a chance to be started first.
1723          */
1724         for (i = 0; i < h->highest_lun + 1; i++) {
1725                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1726                 /* make sure the disk has been added and the drive is real
1727                  * because this can be called from the middle of init_one.
1728                  */
1729                 if (!h->drv[curr_queue])
1730                         continue;
1731                 if (!(h->drv[curr_queue]->queue) ||
1732                         !(h->drv[curr_queue]->heads))
1733                         continue;
1734                 blk_start_queue(h->gendisk[curr_queue]->queue);
1735
1736                 /* check to see if we have maxed out the number of commands
1737                  * that can be placed on the queue.
1738                  */
1739                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1740                         if (curr_queue == start_queue) {
1741                                 h->next_to_run =
1742                                     (start_queue + 1) % (h->highest_lun + 1);
1743                                 break;
1744                         } else {
1745                                 h->next_to_run = curr_queue;
1746                                 break;
1747                         }
1748                 }
1749         }
1750 }
1751
1752 static void cciss_softirq_done(struct request *rq)
1753 {
1754         CommandList_struct *cmd = rq->completion_data;
1755         ctlr_info_t *h = hba[cmd->ctlr];
1756         SGDescriptor_struct *curr_sg = cmd->SG;
1757         u64bit temp64;
1758         unsigned long flags;
1759         int i, ddir;
1760         int sg_index = 0;
1761
1762         if (cmd->Request.Type.Direction == XFER_READ)
1763                 ddir = PCI_DMA_FROMDEVICE;
1764         else
1765                 ddir = PCI_DMA_TODEVICE;
1766
1767         /* command did not need to be retried */
1768         /* unmap the DMA mapping for all the scatter gather elements */
1769         for (i = 0; i < cmd->Header.SGList; i++) {
1770                 if (curr_sg[sg_index].Ext == CCISS_SG_CHAIN) {
1771                         cciss_unmap_sg_chain_block(h, cmd);
1772                         /* Point to the next block */
1773                         curr_sg = h->cmd_sg_list[cmd->cmdindex];
1774                         sg_index = 0;
1775                 }
1776                 temp64.val32.lower = curr_sg[sg_index].Addr.lower;
1777                 temp64.val32.upper = curr_sg[sg_index].Addr.upper;
1778                 pci_unmap_page(h->pdev, temp64.val, curr_sg[sg_index].Len,
1779                                 ddir);
1780                 ++sg_index;
1781         }
1782
1783 #ifdef CCISS_DEBUG
1784         printk("Done with %p\n", rq);
1785 #endif                          /* CCISS_DEBUG */
1786
1787         /* set the residual count for pc requests */
1788         if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
1789                 rq->resid_len = cmd->err_info->ResidualCnt;
1790
1791         blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1792
1793         spin_lock_irqsave(&h->lock, flags);
1794         cmd_free(h, cmd, 1);
1795         cciss_check_queues(h);
1796         spin_unlock_irqrestore(&h->lock, flags);
1797 }
1798
1799 static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1800         unsigned char scsi3addr[], uint32_t log_unit)
1801 {
1802         memcpy(scsi3addr, h->drv[log_unit]->LunID,
1803                 sizeof(h->drv[log_unit]->LunID));
1804 }
1805
1806 /* This function gets the SCSI vendor, model, and revision of a logical drive
1807  * via the inquiry page 0.  Model, vendor, and rev are set to empty strings if
1808  * they cannot be read.
1809  */
1810 static void cciss_get_device_descr(int ctlr, int logvol,
1811                                    char *vendor, char *model, char *rev)
1812 {
1813         int rc;
1814         InquiryData_struct *inq_buf;
1815         unsigned char scsi3addr[8];
1816
1817         *vendor = '\0';
1818         *model = '\0';
1819         *rev = '\0';
1820
1821         inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1822         if (!inq_buf)
1823                 return;
1824
1825         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1826         rc = sendcmd_withirq(CISS_INQUIRY, ctlr, inq_buf, sizeof(*inq_buf), 0,
1827                         scsi3addr, TYPE_CMD);
1828         if (rc == IO_OK) {
1829                 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1830                 vendor[VENDOR_LEN] = '\0';
1831                 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1832                 model[MODEL_LEN] = '\0';
1833                 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1834                 rev[REV_LEN] = '\0';
1835         }
1836
1837         kfree(inq_buf);
1838         return;
1839 }
1840
1841 /* This function gets the serial number of a logical drive via
1842  * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1843  * number cannot be had, for whatever reason, 16 bytes of 0xff
1844  * are returned instead.
1845  */
1846 static void cciss_get_serial_no(int ctlr, int logvol,
1847                                 unsigned char *serial_no, int buflen)
1848 {
1849 #define PAGE_83_INQ_BYTES 64
1850         int rc;
1851         unsigned char *buf;
1852         unsigned char scsi3addr[8];
1853
1854         if (buflen > 16)
1855                 buflen = 16;
1856         memset(serial_no, 0xff, buflen);
1857         buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1858         if (!buf)
1859                 return;
1860         memset(serial_no, 0, buflen);
1861         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1862         rc = sendcmd_withirq(CISS_INQUIRY, ctlr, buf,
1863                 PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1864         if (rc == IO_OK)
1865                 memcpy(serial_no, &buf[8], buflen);
1866         kfree(buf);
1867         return;
1868 }
1869
1870 /*
1871  * cciss_add_disk sets up the block device queue for a logical drive
1872  */
1873 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1874                                 int drv_index)
1875 {
1876         disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1877         if (!disk->queue)
1878                 goto init_queue_failure;
1879         sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1880         disk->major = h->major;
1881         disk->first_minor = drv_index << NWD_SHIFT;
1882         disk->fops = &cciss_fops;
1883         if (cciss_create_ld_sysfs_entry(h, drv_index))
1884                 goto cleanup_queue;
1885         disk->private_data = h->drv[drv_index];
1886         disk->driverfs_dev = &h->drv[drv_index]->dev;
1887
1888         /* Set up queue information */
1889         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1890
1891         /* This is a hardware imposed limit. */
1892         blk_queue_max_segments(disk->queue, h->maxsgentries);
1893
1894         blk_queue_max_hw_sectors(disk->queue, h->cciss_max_sectors);
1895
1896         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1897
1898         disk->queue->queuedata = h;
1899
1900         blk_queue_logical_block_size(disk->queue,
1901                                      h->drv[drv_index]->block_size);
1902
1903         /* Make sure all queue data is written out before */
1904         /* setting h->drv[drv_index]->queue, as setting this */
1905         /* allows the interrupt handler to start the queue */
1906         wmb();
1907         h->drv[drv_index]->queue = disk->queue;
1908         add_disk(disk);
1909         return 0;
1910
1911 cleanup_queue:
1912         blk_cleanup_queue(disk->queue);
1913         disk->queue = NULL;
1914 init_queue_failure:
1915         return -1;
1916 }
1917
1918 /* This function will check the usage_count of the drive to be updated/added.
1919  * If the usage_count is zero and it is a heretofore unknown drive, or,
1920  * the drive's capacity, geometry, or serial number has changed,
1921  * then the drive information will be updated and the disk will be
1922  * re-registered with the kernel.  If these conditions don't hold,
1923  * then it will be left alone for the next reboot.  The exception to this
1924  * is disk 0 which will always be left registered with the kernel since it
1925  * is also the controller node.  Any changes to disk 0 will show up on
1926  * the next reboot.
1927  */
1928 static void cciss_update_drive_info(int ctlr, int drv_index, int first_time,
1929         int via_ioctl)
1930 {
1931         ctlr_info_t *h = hba[ctlr];
1932         struct gendisk *disk;
1933         InquiryData_struct *inq_buff = NULL;
1934         unsigned int block_size;
1935         sector_t total_size;
1936         unsigned long flags = 0;
1937         int ret = 0;
1938         drive_info_struct *drvinfo;
1939
1940         /* Get information about the disk and modify the driver structure */
1941         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1942         drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
1943         if (inq_buff == NULL || drvinfo == NULL)
1944                 goto mem_msg;
1945
1946         /* testing to see if 16-byte CDBs are already being used */
1947         if (h->cciss_read == CCISS_READ_16) {
1948                 cciss_read_capacity_16(h->ctlr, drv_index,
1949                         &total_size, &block_size);
1950
1951         } else {
1952                 cciss_read_capacity(ctlr, drv_index, &total_size, &block_size);
1953                 /* if read_capacity returns all F's this volume is >2TB */
1954                 /* in size so we switch to 16-byte CDB's for all */
1955                 /* read/write ops */
1956                 if (total_size == 0xFFFFFFFFULL) {
1957                         cciss_read_capacity_16(ctlr, drv_index,
1958                         &total_size, &block_size);
1959                         h->cciss_read = CCISS_READ_16;
1960                         h->cciss_write = CCISS_WRITE_16;
1961                 } else {
1962                         h->cciss_read = CCISS_READ_10;
1963                         h->cciss_write = CCISS_WRITE_10;
1964                 }
1965         }
1966
1967         cciss_geometry_inquiry(ctlr, drv_index, total_size, block_size,
1968                                inq_buff, drvinfo);
1969         drvinfo->block_size = block_size;
1970         drvinfo->nr_blocks = total_size + 1;
1971
1972         cciss_get_device_descr(ctlr, drv_index, drvinfo->vendor,
1973                                 drvinfo->model, drvinfo->rev);
1974         cciss_get_serial_no(ctlr, drv_index, drvinfo->serial_no,
1975                         sizeof(drvinfo->serial_no));
1976         /* Save the lunid in case we deregister the disk, below. */
1977         memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
1978                 sizeof(drvinfo->LunID));
1979
1980         /* Is it the same disk we already know, and nothing's changed? */
1981         if (h->drv[drv_index]->raid_level != -1 &&
1982                 ((memcmp(drvinfo->serial_no,
1983                                 h->drv[drv_index]->serial_no, 16) == 0) &&
1984                 drvinfo->block_size == h->drv[drv_index]->block_size &&
1985                 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
1986                 drvinfo->heads == h->drv[drv_index]->heads &&
1987                 drvinfo->sectors == h->drv[drv_index]->sectors &&
1988                 drvinfo->cylinders == h->drv[drv_index]->cylinders))
1989                         /* The disk is unchanged, nothing to update */
1990                         goto freeret;
1991
1992         /* If we get here it's not the same disk, or something's changed,
1993          * so we need to * deregister it, and re-register it, if it's not
1994          * in use.
1995          * If the disk already exists then deregister it before proceeding
1996          * (unless it's the first disk (for the controller node).
1997          */
1998         if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
1999                 printk(KERN_WARNING "disk %d has changed.\n", drv_index);
2000                 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2001                 h->drv[drv_index]->busy_configuring = 1;
2002                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2003
2004                 /* deregister_disk sets h->drv[drv_index]->queue = NULL
2005                  * which keeps the interrupt handler from starting
2006                  * the queue.
2007                  */
2008                 ret = deregister_disk(h, drv_index, 0, via_ioctl);
2009         }
2010
2011         /* If the disk is in use return */
2012         if (ret)
2013                 goto freeret;
2014
2015         /* Save the new information from cciss_geometry_inquiry
2016          * and serial number inquiry.  If the disk was deregistered
2017          * above, then h->drv[drv_index] will be NULL.
2018          */
2019         if (h->drv[drv_index] == NULL) {
2020                 drvinfo->device_initialized = 0;
2021                 h->drv[drv_index] = drvinfo;
2022                 drvinfo = NULL; /* so it won't be freed below. */
2023         } else {
2024                 /* special case for cxd0 */
2025                 h->drv[drv_index]->block_size = drvinfo->block_size;
2026                 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
2027                 h->drv[drv_index]->heads = drvinfo->heads;
2028                 h->drv[drv_index]->sectors = drvinfo->sectors;
2029                 h->drv[drv_index]->cylinders = drvinfo->cylinders;
2030                 h->drv[drv_index]->raid_level = drvinfo->raid_level;
2031                 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
2032                 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
2033                         VENDOR_LEN + 1);
2034                 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
2035                 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
2036         }
2037
2038         ++h->num_luns;
2039         disk = h->gendisk[drv_index];
2040         set_capacity(disk, h->drv[drv_index]->nr_blocks);
2041
2042         /* If it's not disk 0 (drv_index != 0)
2043          * or if it was disk 0, but there was previously
2044          * no actual corresponding configured logical drive
2045          * (raid_leve == -1) then we want to update the
2046          * logical drive's information.
2047          */
2048         if (drv_index || first_time) {
2049                 if (cciss_add_disk(h, disk, drv_index) != 0) {
2050                         cciss_free_gendisk(h, drv_index);
2051                         cciss_free_drive_info(h, drv_index);
2052                         printk(KERN_WARNING "cciss:%d could not update "
2053                                 "disk %d\n", h->ctlr, drv_index);
2054                         --h->num_luns;
2055                 }
2056         }
2057
2058 freeret:
2059         kfree(inq_buff);
2060         kfree(drvinfo);
2061         return;
2062 mem_msg:
2063         printk(KERN_ERR "cciss: out of memory\n");
2064         goto freeret;
2065 }
2066
2067 /* This function will find the first index of the controllers drive array
2068  * that has a null drv pointer and allocate the drive info struct and
2069  * will return that index   This is where new drives will be added.
2070  * If the index to be returned is greater than the highest_lun index for
2071  * the controller then highest_lun is set * to this new index.
2072  * If there are no available indexes or if tha allocation fails, then -1
2073  * is returned.  * "controller_node" is used to know if this is a real
2074  * logical drive, or just the controller node, which determines if this
2075  * counts towards highest_lun.
2076  */
2077 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
2078 {
2079         int i;
2080         drive_info_struct *drv;
2081
2082         /* Search for an empty slot for our drive info */
2083         for (i = 0; i < CISS_MAX_LUN; i++) {
2084
2085                 /* if not cxd0 case, and it's occupied, skip it. */
2086                 if (h->drv[i] && i != 0)
2087                         continue;
2088                 /*
2089                  * If it's cxd0 case, and drv is alloc'ed already, and a
2090                  * disk is configured there, skip it.
2091                  */
2092                 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2093                         continue;
2094
2095                 /*
2096                  * We've found an empty slot.  Update highest_lun
2097                  * provided this isn't just the fake cxd0 controller node.
2098                  */
2099                 if (i > h->highest_lun && !controller_node)
2100                         h->highest_lun = i;
2101
2102                 /* If adding a real disk at cxd0, and it's already alloc'ed */
2103                 if (i == 0 && h->drv[i] != NULL)
2104                         return i;
2105
2106                 /*
2107                  * Found an empty slot, not already alloc'ed.  Allocate it.
2108                  * Mark it with raid_level == -1, so we know it's new later on.
2109                  */
2110                 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2111                 if (!drv)
2112                         return -1;
2113                 drv->raid_level = -1; /* so we know it's new */
2114                 h->drv[i] = drv;
2115                 return i;
2116         }
2117         return -1;
2118 }
2119
2120 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2121 {
2122         kfree(h->drv[drv_index]);
2123         h->drv[drv_index] = NULL;
2124 }
2125
2126 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2127 {
2128         put_disk(h->gendisk[drv_index]);
2129         h->gendisk[drv_index] = NULL;
2130 }
2131
2132 /* cciss_add_gendisk finds a free hba[]->drv structure
2133  * and allocates a gendisk if needed, and sets the lunid
2134  * in the drvinfo structure.   It returns the index into
2135  * the ->drv[] array, or -1 if none are free.
2136  * is_controller_node indicates whether highest_lun should
2137  * count this disk, or if it's only being added to provide
2138  * a means to talk to the controller in case no logical
2139  * drives have yet been configured.
2140  */
2141 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2142         int controller_node)
2143 {
2144         int drv_index;
2145
2146         drv_index = cciss_alloc_drive_info(h, controller_node);
2147         if (drv_index == -1)
2148                 return -1;
2149
2150         /*Check if the gendisk needs to be allocated */
2151         if (!h->gendisk[drv_index]) {
2152                 h->gendisk[drv_index] =
2153                         alloc_disk(1 << NWD_SHIFT);
2154                 if (!h->gendisk[drv_index]) {
2155                         printk(KERN_ERR "cciss%d: could not "
2156                                 "allocate a new disk %d\n",
2157                                 h->ctlr, drv_index);
2158                         goto err_free_drive_info;
2159                 }
2160         }
2161         memcpy(h->drv[drv_index]->LunID, lunid,
2162                 sizeof(h->drv[drv_index]->LunID));
2163         if (cciss_create_ld_sysfs_entry(h, drv_index))
2164                 goto err_free_disk;
2165         /* Don't need to mark this busy because nobody */
2166         /* else knows about this disk yet to contend */
2167         /* for access to it. */
2168         h->drv[drv_index]->busy_configuring = 0;
2169         wmb();
2170         return drv_index;
2171
2172 err_free_disk:
2173         cciss_free_gendisk(h, drv_index);
2174 err_free_drive_info:
2175         cciss_free_drive_info(h, drv_index);
2176         return -1;
2177 }
2178
2179 /* This is for the special case of a controller which
2180  * has no logical drives.  In this case, we still need
2181  * to register a disk so the controller can be accessed
2182  * by the Array Config Utility.
2183  */
2184 static void cciss_add_controller_node(ctlr_info_t *h)
2185 {
2186         struct gendisk *disk;
2187         int drv_index;
2188
2189         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2190                 return;
2191
2192         drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2193         if (drv_index == -1)
2194                 goto error;
2195         h->drv[drv_index]->block_size = 512;
2196         h->drv[drv_index]->nr_blocks = 0;
2197         h->drv[drv_index]->heads = 0;
2198         h->drv[drv_index]->sectors = 0;
2199         h->drv[drv_index]->cylinders = 0;
2200         h->drv[drv_index]->raid_level = -1;
2201         memset(h->drv[drv_index]->serial_no, 0, 16);
2202         disk = h->gendisk[drv_index];
2203         if (cciss_add_disk(h, disk, drv_index) == 0)
2204                 return;
2205         cciss_free_gendisk(h, drv_index);
2206         cciss_free_drive_info(h, drv_index);
2207 error:
2208         printk(KERN_WARNING "cciss%d: could not "
2209                 "add disk 0.\n", h->ctlr);
2210         return;
2211 }
2212
2213 /* This function will add and remove logical drives from the Logical
2214  * drive array of the controller and maintain persistency of ordering
2215  * so that mount points are preserved until the next reboot.  This allows
2216  * for the removal of logical drives in the middle of the drive array
2217  * without a re-ordering of those drives.
2218  * INPUT
2219  * h            = The controller to perform the operations on
2220  */
2221 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2222         int via_ioctl)
2223 {
2224         int ctlr = h->ctlr;
2225         int num_luns;
2226         ReportLunData_struct *ld_buff = NULL;
2227         int return_code;
2228         int listlength = 0;
2229         int i;
2230         int drv_found;
2231         int drv_index = 0;
2232         unsigned char lunid[8] = CTLR_LUNID;
2233         unsigned long flags;
2234
2235         if (!capable(CAP_SYS_RAWIO))
2236                 return -EPERM;
2237
2238         /* Set busy_configuring flag for this operation */
2239         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2240         if (h->busy_configuring) {
2241                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2242                 return -EBUSY;
2243         }
2244         h->busy_configuring = 1;
2245         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2246
2247         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2248         if (ld_buff == NULL)
2249                 goto mem_msg;
2250
2251         return_code = sendcmd_withirq(CISS_REPORT_LOG, ctlr, ld_buff,
2252                                       sizeof(ReportLunData_struct),
2253                                       0, CTLR_LUNID, TYPE_CMD);
2254
2255         if (return_code == IO_OK)
2256                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2257         else {  /* reading number of logical volumes failed */
2258                 printk(KERN_WARNING "cciss: report logical volume"
2259                        " command failed\n");
2260                 listlength = 0;
2261                 goto freeret;
2262         }
2263
2264         num_luns = listlength / 8;      /* 8 bytes per entry */
2265         if (num_luns > CISS_MAX_LUN) {
2266                 num_luns = CISS_MAX_LUN;
2267                 printk(KERN_WARNING "cciss: more luns configured"
2268                        " on controller than can be handled by"
2269                        " this driver.\n");
2270         }
2271
2272         if (num_luns == 0)
2273                 cciss_add_controller_node(h);
2274
2275         /* Compare controller drive array to driver's drive array
2276          * to see if any drives are missing on the controller due
2277          * to action of Array Config Utility (user deletes drive)
2278          * and deregister logical drives which have disappeared.
2279          */
2280         for (i = 0; i <= h->highest_lun; i++) {
2281                 int j;
2282                 drv_found = 0;
2283
2284                 /* skip holes in the array from already deleted drives */
2285                 if (h->drv[i] == NULL)
2286                         continue;
2287
2288                 for (j = 0; j < num_luns; j++) {
2289                         memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2290                         if (memcmp(h->drv[i]->LunID, lunid,
2291                                 sizeof(lunid)) == 0) {
2292                                 drv_found = 1;
2293                                 break;
2294                         }
2295                 }
2296                 if (!drv_found) {
2297                         /* Deregister it from the OS, it's gone. */
2298                         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2299                         h->drv[i]->busy_configuring = 1;
2300                         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2301                         return_code = deregister_disk(h, i, 1, via_ioctl);
2302                         if (h->drv[i] != NULL)
2303                                 h->drv[i]->busy_configuring = 0;
2304                 }
2305         }
2306
2307         /* Compare controller drive array to driver's drive array.
2308          * Check for updates in the drive information and any new drives
2309          * on the controller due to ACU adding logical drives, or changing
2310          * a logical drive's size, etc.  Reregister any new/changed drives
2311          */
2312         for (i = 0; i < num_luns; i++) {
2313                 int j;
2314
2315                 drv_found = 0;
2316
2317                 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2318                 /* Find if the LUN is already in the drive array
2319                  * of the driver.  If so then update its info
2320                  * if not in use.  If it does not exist then find
2321                  * the first free index and add it.
2322                  */
2323                 for (j = 0; j <= h->highest_lun; j++) {
2324                         if (h->drv[j] != NULL &&
2325                                 memcmp(h->drv[j]->LunID, lunid,
2326                                         sizeof(h->drv[j]->LunID)) == 0) {
2327                                 drv_index = j;
2328                                 drv_found = 1;
2329                                 break;
2330                         }
2331                 }
2332
2333                 /* check if the drive was found already in the array */
2334                 if (!drv_found) {
2335                         drv_index = cciss_add_gendisk(h, lunid, 0);
2336                         if (drv_index == -1)
2337                                 goto freeret;
2338                 }
2339                 cciss_update_drive_info(ctlr, drv_index, first_time,
2340                         via_ioctl);
2341         }               /* end for */
2342
2343 freeret:
2344         kfree(ld_buff);
2345         h->busy_configuring = 0;
2346         /* We return -1 here to tell the ACU that we have registered/updated
2347          * all of the drives that we can and to keep it from calling us
2348          * additional times.
2349          */
2350         return -1;
2351 mem_msg:
2352         printk(KERN_ERR "cciss: out of memory\n");
2353         h->busy_configuring = 0;
2354         goto freeret;
2355 }
2356
2357 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2358 {
2359         /* zero out the disk size info */
2360         drive_info->nr_blocks = 0;
2361         drive_info->block_size = 0;
2362         drive_info->heads = 0;
2363         drive_info->sectors = 0;
2364         drive_info->cylinders = 0;
2365         drive_info->raid_level = -1;
2366         memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2367         memset(drive_info->model, 0, sizeof(drive_info->model));
2368         memset(drive_info->rev, 0, sizeof(drive_info->rev));
2369         memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2370         /*
2371          * don't clear the LUNID though, we need to remember which
2372          * one this one is.
2373          */
2374 }
2375
2376 /* This function will deregister the disk and it's queue from the
2377  * kernel.  It must be called with the controller lock held and the
2378  * drv structures busy_configuring flag set.  It's parameters are:
2379  *
2380  * disk = This is the disk to be deregistered
2381  * drv  = This is the drive_info_struct associated with the disk to be
2382  *        deregistered.  It contains information about the disk used
2383  *        by the driver.
2384  * clear_all = This flag determines whether or not the disk information
2385  *             is going to be completely cleared out and the highest_lun
2386  *             reset.  Sometimes we want to clear out information about
2387  *             the disk in preparation for re-adding it.  In this case
2388  *             the highest_lun should be left unchanged and the LunID
2389  *             should not be cleared.
2390  * via_ioctl
2391  *    This indicates whether we've reached this path via ioctl.
2392  *    This affects the maximum usage count allowed for c0d0 to be messed with.
2393  *    If this path is reached via ioctl(), then the max_usage_count will
2394  *    be 1, as the process calling ioctl() has got to have the device open.
2395  *    If we get here via sysfs, then the max usage count will be zero.
2396 */
2397 static int deregister_disk(ctlr_info_t *h, int drv_index,
2398                            int clear_all, int via_ioctl)
2399 {
2400         int i;
2401         struct gendisk *disk;
2402         drive_info_struct *drv;
2403         int recalculate_highest_lun;
2404
2405         if (!capable(CAP_SYS_RAWIO))
2406                 return -EPERM;
2407
2408         drv = h->drv[drv_index];
2409         disk = h->gendisk[drv_index];
2410
2411         /* make sure logical volume is NOT is use */
2412         if (clear_all || (h->gendisk[0] == disk)) {
2413                 if (drv->usage_count > via_ioctl)
2414                         return -EBUSY;
2415         } else if (drv->usage_count > 0)
2416                 return -EBUSY;
2417
2418         recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2419
2420         /* invalidate the devices and deregister the disk.  If it is disk
2421          * zero do not deregister it but just zero out it's values.  This
2422          * allows us to delete disk zero but keep the controller registered.
2423          */
2424         if (h->gendisk[0] != disk) {
2425                 struct request_queue *q = disk->queue;
2426                 if (disk->flags & GENHD_FL_UP) {
2427                         cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2428                         del_gendisk(disk);
2429                 }
2430                 if (q)
2431                         blk_cleanup_queue(q);
2432                 /* If clear_all is set then we are deleting the logical
2433                  * drive, not just refreshing its info.  For drives
2434                  * other than disk 0 we will call put_disk.  We do not
2435                  * do this for disk 0 as we need it to be able to
2436                  * configure the controller.
2437                  */
2438                 if (clear_all){
2439                         /* This isn't pretty, but we need to find the
2440                          * disk in our array and NULL our the pointer.
2441                          * This is so that we will call alloc_disk if
2442                          * this index is used again later.
2443                          */
2444                         for (i=0; i < CISS_MAX_LUN; i++){
2445                                 if (h->gendisk[i] == disk) {
2446                                         h->gendisk[i] = NULL;
2447                                         break;
2448                                 }
2449                         }
2450                         put_disk(disk);
2451                 }
2452         } else {
2453                 set_capacity(disk, 0);
2454                 cciss_clear_drive_info(drv);
2455         }
2456
2457         --h->num_luns;
2458
2459         /* if it was the last disk, find the new hightest lun */
2460         if (clear_all && recalculate_highest_lun) {
2461                 int newhighest = -1;
2462                 for (i = 0; i <= h->highest_lun; i++) {
2463                         /* if the disk has size > 0, it is available */
2464                         if (h->drv[i] && h->drv[i]->heads)
2465                                 newhighest = i;
2466                 }
2467                 h->highest_lun = newhighest;
2468         }
2469         return 0;
2470 }
2471
2472 static int fill_cmd(CommandList_struct *c, __u8 cmd, int ctlr, void *buff,
2473                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2474                 int cmd_type)
2475 {
2476         ctlr_info_t *h = hba[ctlr];
2477         u64bit buff_dma_handle;
2478         int status = IO_OK;
2479
2480         c->cmd_type = CMD_IOCTL_PEND;
2481         c->Header.ReplyQueue = 0;
2482         if (buff != NULL) {
2483                 c->Header.SGList = 1;
2484                 c->Header.SGTotal = 1;
2485         } else {
2486                 c->Header.SGList = 0;
2487                 c->Header.SGTotal = 0;
2488         }
2489         c->Header.Tag.lower = c->busaddr;
2490         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2491
2492         c->Request.Type.Type = cmd_type;
2493         if (cmd_type == TYPE_CMD) {
2494                 switch (cmd) {
2495                 case CISS_INQUIRY:
2496                         /* are we trying to read a vital product page */
2497                         if (page_code != 0) {
2498                                 c->Request.CDB[1] = 0x01;
2499                                 c->Request.CDB[2] = page_code;
2500                         }
2501                         c->Request.CDBLen = 6;
2502                         c->Request.Type.Attribute = ATTR_SIMPLE;
2503                         c->Request.Type.Direction = XFER_READ;
2504                         c->Request.Timeout = 0;
2505                         c->Request.CDB[0] = CISS_INQUIRY;
2506                         c->Request.CDB[4] = size & 0xFF;
2507                         break;
2508                 case CISS_REPORT_LOG:
2509                 case CISS_REPORT_PHYS:
2510                         /* Talking to controller so It's a physical command
2511                            mode = 00 target = 0.  Nothing to write.
2512                          */
2513                         c->Request.CDBLen = 12;
2514                         c->Request.Type.Attribute = ATTR_SIMPLE;
2515                         c->Request.Type.Direction = XFER_READ;
2516                         c->Request.Timeout = 0;
2517                         c->Request.CDB[0] = cmd;
2518                         c->Request.CDB[6] = (size >> 24) & 0xFF; /* MSB */
2519                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2520                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2521                         c->Request.CDB[9] = size & 0xFF;
2522                         break;
2523
2524                 case CCISS_READ_CAPACITY:
2525                         c->Request.CDBLen = 10;
2526                         c->Request.Type.Attribute = ATTR_SIMPLE;
2527                         c->Request.Type.Direction = XFER_READ;
2528                         c->Request.Timeout = 0;
2529                         c->Request.CDB[0] = cmd;
2530                         break;
2531                 case CCISS_READ_CAPACITY_16:
2532                         c->Request.CDBLen = 16;
2533                         c->Request.Type.Attribute = ATTR_SIMPLE;
2534                         c->Request.Type.Direction = XFER_READ;
2535                         c->Request.Timeout = 0;
2536                         c->Request.CDB[0] = cmd;
2537                         c->Request.CDB[1] = 0x10;
2538                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2539                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2540                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2541                         c->Request.CDB[13] = size & 0xFF;
2542                         c->Request.Timeout = 0;
2543                         c->Request.CDB[0] = cmd;
2544                         break;
2545                 case CCISS_CACHE_FLUSH:
2546                         c->Request.CDBLen = 12;
2547                         c->Request.Type.Attribute = ATTR_SIMPLE;
2548                         c->Request.Type.Direction = XFER_WRITE;
2549                         c->Request.Timeout = 0;
2550                         c->Request.CDB[0] = BMIC_WRITE;
2551                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2552                         break;
2553                 case TEST_UNIT_READY:
2554                         c->Request.CDBLen = 6;
2555                         c->Request.Type.Attribute = ATTR_SIMPLE;
2556                         c->Request.Type.Direction = XFER_NONE;
2557                         c->Request.Timeout = 0;
2558                         break;
2559                 default:
2560                         printk(KERN_WARNING
2561                                "cciss%d:  Unknown Command 0x%c\n", ctlr, cmd);
2562                         return IO_ERROR;
2563                 }
2564         } else if (cmd_type == TYPE_MSG) {
2565                 switch (cmd) {
2566                 case 0: /* ABORT message */
2567                         c->Request.CDBLen = 12;
2568                         c->Request.Type.Attribute = ATTR_SIMPLE;
2569                         c->Request.Type.Direction = XFER_WRITE;
2570                         c->Request.Timeout = 0;
2571                         c->Request.CDB[0] = cmd;        /* abort */
2572                         c->Request.CDB[1] = 0;  /* abort a command */
2573                         /* buff contains the tag of the command to abort */
2574                         memcpy(&c->Request.CDB[4], buff, 8);
2575                         break;
2576                 case 1: /* RESET message */
2577                         c->Request.CDBLen = 16;
2578                         c->Request.Type.Attribute = ATTR_SIMPLE;
2579                         c->Request.Type.Direction = XFER_NONE;
2580                         c->Request.Timeout = 0;
2581                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2582                         c->Request.CDB[0] = cmd;        /* reset */
2583                         c->Request.CDB[1] = 0x03;       /* reset a target */
2584                         break;
2585                 case 3: /* No-Op message */
2586                         c->Request.CDBLen = 1;
2587                         c->Request.Type.Attribute = ATTR_SIMPLE;
2588                         c->Request.Type.Direction = XFER_WRITE;
2589                         c->Request.Timeout = 0;
2590                         c->Request.CDB[0] = cmd;
2591                         break;
2592                 default:
2593                         printk(KERN_WARNING
2594                                "cciss%d: unknown message type %d\n", ctlr, cmd);
2595                         return IO_ERROR;
2596                 }
2597         } else {
2598                 printk(KERN_WARNING
2599                        "cciss%d: unknown command type %d\n", ctlr, cmd_type);
2600                 return IO_ERROR;
2601         }
2602         /* Fill in the scatter gather information */
2603         if (size > 0) {
2604                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2605                                                              buff, size,
2606                                                              PCI_DMA_BIDIRECTIONAL);
2607                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2608                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2609                 c->SG[0].Len = size;
2610                 c->SG[0].Ext = 0;       /* we are not chaining */
2611         }
2612         return status;
2613 }
2614
2615 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2616 {
2617         switch (c->err_info->ScsiStatus) {
2618         case SAM_STAT_GOOD:
2619                 return IO_OK;
2620         case SAM_STAT_CHECK_CONDITION:
2621                 switch (0xf & c->err_info->SenseInfo[2]) {
2622                 case 0: return IO_OK; /* no sense */
2623                 case 1: return IO_OK; /* recovered error */
2624                 default:
2625                         if (check_for_unit_attention(h, c))
2626                                 return IO_NEEDS_RETRY;
2627                         printk(KERN_WARNING "cciss%d: cmd 0x%02x "
2628                                 "check condition, sense key = 0x%02x\n",
2629                                 h->ctlr, c->Request.CDB[0],
2630                                 c->err_info->SenseInfo[2]);
2631                 }
2632                 break;
2633         default:
2634                 printk(KERN_WARNING "cciss%d: cmd 0x%02x"
2635                         "scsi status = 0x%02x\n", h->ctlr,
2636                         c->Request.CDB[0], c->err_info->ScsiStatus);
2637                 break;
2638         }
2639         return IO_ERROR;
2640 }
2641
2642 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2643 {
2644         int return_status = IO_OK;
2645
2646         if (c->err_info->CommandStatus == CMD_SUCCESS)
2647                 return IO_OK;
2648
2649         switch (c->err_info->CommandStatus) {
2650         case CMD_TARGET_STATUS:
2651                 return_status = check_target_status(h, c);
2652                 break;
2653         case CMD_DATA_UNDERRUN:
2654         case CMD_DATA_OVERRUN:
2655                 /* expected for inquiry and report lun commands */
2656                 break;
2657         case CMD_INVALID:
2658                 printk(KERN_WARNING "cciss: cmd 0x%02x is "
2659                        "reported invalid\n", c->Request.CDB[0]);
2660                 return_status = IO_ERROR;
2661                 break;
2662         case CMD_PROTOCOL_ERR:
2663                 printk(KERN_WARNING "cciss: cmd 0x%02x has "
2664                        "protocol error \n", c->Request.CDB[0]);
2665                 return_status = IO_ERROR;
2666                 break;
2667         case CMD_HARDWARE_ERR:
2668                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2669                        " hardware error\n", c->Request.CDB[0]);
2670                 return_status = IO_ERROR;
2671                 break;
2672         case CMD_CONNECTION_LOST:
2673                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2674                        "connection lost\n", c->Request.CDB[0]);
2675                 return_status = IO_ERROR;
2676                 break;
2677         case CMD_ABORTED:
2678                 printk(KERN_WARNING "cciss: cmd 0x%02x was "
2679                        "aborted\n", c->Request.CDB[0]);
2680                 return_status = IO_ERROR;
2681                 break;
2682         case CMD_ABORT_FAILED:
2683                 printk(KERN_WARNING "cciss: cmd 0x%02x reports "
2684                        "abort failed\n", c->Request.CDB[0]);
2685                 return_status = IO_ERROR;
2686                 break;
2687         case CMD_UNSOLICITED_ABORT:
2688                 printk(KERN_WARNING
2689                        "cciss%d: unsolicited abort 0x%02x\n", h->ctlr,
2690                         c->Request.CDB[0]);
2691                 return_status = IO_NEEDS_RETRY;
2692                 break;
2693         default:
2694                 printk(KERN_WARNING "cciss: cmd 0x%02x returned "
2695                        "unknown status %x\n", c->Request.CDB[0],
2696                        c->err_info->CommandStatus);
2697                 return_status = IO_ERROR;
2698         }
2699         return return_status;
2700 }
2701
2702 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2703         int attempt_retry)
2704 {
2705         DECLARE_COMPLETION_ONSTACK(wait);
2706         u64bit buff_dma_handle;
2707         int return_status = IO_OK;
2708
2709 resend_cmd2:
2710         c->waiting = &wait;
2711         enqueue_cmd_and_start_io(h, c);
2712
2713         wait_for_completion(&wait);
2714
2715         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2716                 goto command_done;
2717
2718         return_status = process_sendcmd_error(h, c);
2719
2720         if (return_status == IO_NEEDS_RETRY &&
2721                 c->retry_count < MAX_CMD_RETRIES) {
2722                 printk(KERN_WARNING "cciss%d: retrying 0x%02x\n", h->ctlr,
2723                         c->Request.CDB[0]);
2724                 c->retry_count++;
2725                 /* erase the old error information */
2726                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2727                 return_status = IO_OK;
2728                 INIT_COMPLETION(wait);
2729                 goto resend_cmd2;
2730         }
2731
2732 command_done:
2733         /* unlock the buffers from DMA */
2734         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2735         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2736         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2737                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2738         return return_status;
2739 }
2740
2741 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
2742                            __u8 page_code, unsigned char scsi3addr[],
2743                         int cmd_type)
2744 {
2745         ctlr_info_t *h = hba[ctlr];
2746         CommandList_struct *c;
2747         int return_status;
2748
2749         c = cmd_alloc(h, 0);
2750         if (!c)
2751                 return -ENOMEM;
2752         return_status = fill_cmd(c, cmd, ctlr, buff, size, page_code,
2753                 scsi3addr, cmd_type);
2754         if (return_status == IO_OK)
2755                 return_status = sendcmd_withirq_core(h, c, 1);
2756
2757         cmd_free(h, c, 0);
2758         return return_status;
2759 }
2760
2761 static void cciss_geometry_inquiry(int ctlr, int logvol,
2762                                    sector_t total_size,
2763                                    unsigned int block_size,
2764                                    InquiryData_struct *inq_buff,
2765                                    drive_info_struct *drv)
2766 {
2767         int return_code;
2768         unsigned long t;
2769         unsigned char scsi3addr[8];
2770
2771         memset(inq_buff, 0, sizeof(InquiryData_struct));
2772         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2773         return_code = sendcmd_withirq(CISS_INQUIRY, ctlr, inq_buff,
2774                         sizeof(*inq_buff), 0xC1, scsi3addr, TYPE_CMD);
2775         if (return_code == IO_OK) {
2776                 if (inq_buff->data_byte[8] == 0xFF) {
2777                         printk(KERN_WARNING
2778                                "cciss: reading geometry failed, volume "
2779                                "does not support reading geometry\n");
2780                         drv->heads = 255;
2781                         drv->sectors = 32;      /* Sectors per track */
2782                         drv->cylinders = total_size + 1;
2783                         drv->raid_level = RAID_UNKNOWN;
2784                 } else {
2785                         drv->heads = inq_buff->data_byte[6];
2786                         drv->sectors = inq_buff->data_byte[7];
2787                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2788                         drv->cylinders += inq_buff->data_byte[5];
2789                         drv->raid_level = inq_buff->data_byte[8];
2790                 }
2791                 drv->block_size = block_size;
2792                 drv->nr_blocks = total_size + 1;
2793                 t = drv->heads * drv->sectors;
2794                 if (t > 1) {
2795                         sector_t real_size = total_size + 1;
2796                         unsigned long rem = sector_div(real_size, t);
2797                         if (rem)
2798                                 real_size++;
2799                         drv->cylinders = real_size;
2800                 }
2801         } else {                /* Get geometry failed */
2802                 printk(KERN_WARNING "cciss: reading geometry failed\n");
2803         }
2804 }
2805
2806 static void
2807 cciss_read_capacity(int ctlr, int logvol, sector_t *total_size,
2808                     unsigned int *block_size)
2809 {
2810         ReadCapdata_struct *buf;
2811         int return_code;
2812         unsigned char scsi3addr[8];
2813
2814         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2815         if (!buf) {
2816                 printk(KERN_WARNING "cciss: out of memory\n");
2817                 return;
2818         }
2819
2820         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2821         return_code = sendcmd_withirq(CCISS_READ_CAPACITY, ctlr, buf,
2822                 sizeof(ReadCapdata_struct), 0, scsi3addr, TYPE_CMD);
2823         if (return_code == IO_OK) {
2824                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2825                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2826         } else {                /* read capacity command failed */
2827                 printk(KERN_WARNING "cciss: read capacity failed\n");
2828                 *total_size = 0;
2829                 *block_size = BLOCK_SIZE;
2830         }
2831         kfree(buf);
2832 }
2833
2834 static void cciss_read_capacity_16(int ctlr, int logvol,
2835         sector_t *total_size, unsigned int *block_size)
2836 {
2837         ReadCapdata_struct_16 *buf;
2838         int return_code;
2839         unsigned char scsi3addr[8];
2840
2841         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2842         if (!buf) {
2843                 printk(KERN_WARNING "cciss: out of memory\n");
2844                 return;
2845         }
2846
2847         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2848         return_code = sendcmd_withirq(CCISS_READ_CAPACITY_16,
2849                 ctlr, buf, sizeof(ReadCapdata_struct_16),
2850                         0, scsi3addr, TYPE_CMD);
2851         if (return_code == IO_OK) {
2852                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2853                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2854         } else {                /* read capacity command failed */
2855                 printk(KERN_WARNING "cciss: read capacity failed\n");
2856                 *total_size = 0;
2857                 *block_size = BLOCK_SIZE;
2858         }
2859         printk(KERN_INFO "      blocks= %llu block_size= %d\n",
2860                (unsigned long long)*total_size+1, *block_size);
2861         kfree(buf);
2862 }
2863
2864 static int cciss_revalidate(struct gendisk *disk)
2865 {
2866         ctlr_info_t *h = get_host(disk);
2867         drive_info_struct *drv = get_drv(disk);
2868         int logvol;
2869         int FOUND = 0;
2870         unsigned int block_size;
2871         sector_t total_size;
2872         InquiryData_struct *inq_buff = NULL;
2873
2874         for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
2875                 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2876                         sizeof(drv->LunID)) == 0) {
2877                         FOUND = 1;
2878                         break;
2879                 }
2880         }
2881
2882         if (!FOUND)
2883                 return 1;
2884
2885         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2886         if (inq_buff == NULL) {
2887                 printk(KERN_WARNING "cciss: out of memory\n");
2888                 return 1;
2889         }
2890         if (h->cciss_read == CCISS_READ_10) {
2891                 cciss_read_capacity(h->ctlr, logvol,
2892                                         &total_size, &block_size);
2893         } else {
2894                 cciss_read_capacity_16(h->ctlr, logvol,
2895                                         &total_size, &block_size);
2896         }
2897         cciss_geometry_inquiry(h->ctlr, logvol, total_size, block_size,
2898                                inq_buff, drv);
2899
2900         blk_queue_logical_block_size(drv->queue, drv->block_size);
2901         set_capacity(disk, drv->nr_blocks);
2902
2903         kfree(inq_buff);
2904         return 0;
2905 }
2906
2907 /*
2908  * Map (physical) PCI mem into (virtual) kernel space
2909  */
2910 static void __iomem *remap_pci_mem(ulong base, ulong size)
2911 {
2912         ulong page_base = ((ulong) base) & PAGE_MASK;
2913         ulong page_offs = ((ulong) base) - page_base;
2914         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2915
2916         return page_remapped ? (page_remapped + page_offs) : NULL;
2917 }
2918
2919 /*
2920  * Takes jobs of the Q and sends them to the hardware, then puts it on
2921  * the Q to wait for completion.
2922  */
2923 static void start_io(ctlr_info_t *h)
2924 {
2925         CommandList_struct *c;
2926
2927         while (!hlist_empty(&h->reqQ)) {
2928                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
2929                 /* can't do anything if fifo is full */
2930                 if ((h->access.fifo_full(h))) {
2931                         printk(KERN_WARNING "cciss: fifo full\n");
2932                         break;
2933                 }
2934
2935                 /* Get the first entry from the Request Q */
2936                 removeQ(c);
2937                 h->Qdepth--;
2938
2939                 /* Tell the controller execute command */
2940                 h->access.submit_command(h, c);
2941
2942                 /* Put job onto the completed Q */
2943                 addQ(&h->cmpQ, c);
2944         }
2945 }
2946
2947 /* Assumes that CCISS_LOCK(h->ctlr) is held. */
2948 /* Zeros out the error record and then resends the command back */
2949 /* to the controller */
2950 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2951 {
2952         /* erase the old error information */
2953         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2954
2955         /* add it to software queue and then send it to the controller */
2956         addQ(&h->reqQ, c);
2957         h->Qdepth++;
2958         if (h->Qdepth > h->maxQsinceinit)
2959                 h->maxQsinceinit = h->Qdepth;
2960
2961         start_io(h);
2962 }
2963
2964 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
2965         unsigned int msg_byte, unsigned int host_byte,
2966         unsigned int driver_byte)
2967 {
2968         /* inverse of macros in scsi.h */
2969         return (scsi_status_byte & 0xff) |
2970                 ((msg_byte & 0xff) << 8) |
2971                 ((host_byte & 0xff) << 16) |
2972                 ((driver_byte & 0xff) << 24);
2973 }
2974
2975 static inline int evaluate_target_status(ctlr_info_t *h,
2976                         CommandList_struct *cmd, int *retry_cmd)
2977 {
2978         unsigned char sense_key;
2979         unsigned char status_byte, msg_byte, host_byte, driver_byte;
2980         int error_value;
2981
2982         *retry_cmd = 0;
2983         /* If we get in here, it means we got "target status", that is, scsi status */
2984         status_byte = cmd->err_info->ScsiStatus;
2985         driver_byte = DRIVER_OK;
2986         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
2987
2988         if (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC)
2989                 host_byte = DID_PASSTHROUGH;
2990         else
2991                 host_byte = DID_OK;
2992
2993         error_value = make_status_bytes(status_byte, msg_byte,
2994                 host_byte, driver_byte);
2995
2996         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
2997                 if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC)
2998                         printk(KERN_WARNING "cciss: cmd %p "
2999                                "has SCSI Status 0x%x\n",
3000                                cmd, cmd->err_info->ScsiStatus);
3001                 return error_value;
3002         }
3003
3004         /* check the sense key */
3005         sense_key = 0xf & cmd->err_info->SenseInfo[2];
3006         /* no status or recovered error */
3007         if (((sense_key == 0x0) || (sense_key == 0x1)) &&
3008             (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC))
3009                 error_value = 0;
3010
3011         if (check_for_unit_attention(h, cmd)) {
3012                 *retry_cmd = !(cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC);
3013                 return 0;
3014         }
3015
3016         /* Not SG_IO or similar? */
3017         if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC) {
3018                 if (error_value != 0)
3019                         printk(KERN_WARNING "cciss: cmd %p has CHECK CONDITION"
3020                                " sense key = 0x%x\n", cmd, sense_key);
3021                 return error_value;
3022         }
3023
3024         /* SG_IO or similar, copy sense data back */
3025         if (cmd->rq->sense) {
3026                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
3027                         cmd->rq->sense_len = cmd->err_info->SenseLen;
3028                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
3029                         cmd->rq->sense_len);
3030         } else
3031                 cmd->rq->sense_len = 0;
3032
3033         return error_value;
3034 }
3035
3036 /* checks the status of the job and calls complete buffers to mark all
3037  * buffers for the completed job. Note that this function does not need
3038  * to hold the hba/queue lock.
3039  */
3040 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
3041                                     int timeout)
3042 {
3043         int retry_cmd = 0;
3044         struct request *rq = cmd->rq;
3045
3046         rq->errors = 0;
3047
3048         if (timeout)
3049                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
3050
3051         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
3052                 goto after_error_processing;
3053
3054         switch (cmd->err_info->CommandStatus) {
3055         case CMD_TARGET_STATUS:
3056                 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
3057                 break;
3058         case CMD_DATA_UNDERRUN:
3059                 if (cmd->rq->cmd_type == REQ_TYPE_FS) {
3060                         printk(KERN_WARNING "cciss: cmd %p has"
3061                                " completed with data underrun "
3062                                "reported\n", cmd);
3063                         cmd->rq->resid_len = cmd->err_info->ResidualCnt;
3064                 }
3065                 break;
3066         case CMD_DATA_OVERRUN:
3067                 if (cmd->rq->cmd_type == REQ_TYPE_FS)
3068                         printk(KERN_WARNING "cciss: cmd %p has"
3069                                " completed with data overrun "
3070                                "reported\n", cmd);
3071                 break;
3072         case CMD_INVALID:
3073                 printk(KERN_WARNING "cciss: cmd %p is "
3074                        "reported invalid\n", cmd);
3075                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3076                         cmd->err_info->CommandStatus, DRIVER_OK,
3077                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3078                                 DID_PASSTHROUGH : DID_ERROR);
3079                 break;
3080         case CMD_PROTOCOL_ERR:
3081                 printk(KERN_WARNING "cciss: cmd %p has "
3082                        "protocol error \n", cmd);
3083                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3084                         cmd->err_info->CommandStatus, DRIVER_OK,
3085                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3086                                 DID_PASSTHROUGH : DID_ERROR);
3087                 break;
3088         case CMD_HARDWARE_ERR:
3089                 printk(KERN_WARNING "cciss: cmd %p had "
3090                        " hardware error\n", cmd);
3091                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3092                         cmd->err_info->CommandStatus, DRIVER_OK,
3093                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3094                                 DID_PASSTHROUGH : DID_ERROR);
3095                 break;
3096         case CMD_CONNECTION_LOST:
3097                 printk(KERN_WARNING "cciss: cmd %p had "
3098                        "connection lost\n", cmd);
3099                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3100                         cmd->err_info->CommandStatus, DRIVER_OK,
3101                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3102                                 DID_PASSTHROUGH : DID_ERROR);
3103                 break;
3104         case CMD_ABORTED:
3105                 printk(KERN_WARNING "cciss: cmd %p was "
3106                        "aborted\n", cmd);
3107                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3108                         cmd->err_info->CommandStatus, DRIVER_OK,
3109                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3110                                 DID_PASSTHROUGH : DID_ABORT);
3111                 break;
3112         case CMD_ABORT_FAILED:
3113                 printk(KERN_WARNING "cciss: cmd %p reports "
3114                        "abort failed\n", cmd);
3115                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3116                         cmd->err_info->CommandStatus, DRIVER_OK,
3117                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3118                                 DID_PASSTHROUGH : DID_ERROR);
3119                 break;
3120         case CMD_UNSOLICITED_ABORT:
3121                 printk(KERN_WARNING "cciss%d: unsolicited "
3122                        "abort %p\n", h->ctlr, cmd);
3123                 if (cmd->retry_count < MAX_CMD_RETRIES) {
3124                         retry_cmd = 1;
3125                         printk(KERN_WARNING
3126                                "cciss%d: retrying %p\n", h->ctlr, cmd);
3127                         cmd->retry_count++;
3128                 } else
3129                         printk(KERN_WARNING
3130                                "cciss%d: %p retried too "
3131                                "many times\n", h->ctlr, cmd);
3132                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3133                         cmd->err_info->CommandStatus, DRIVER_OK,
3134                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3135                                 DID_PASSTHROUGH : DID_ABORT);
3136                 break;
3137         case CMD_TIMEOUT:
3138                 printk(KERN_WARNING "cciss: cmd %p timedout\n", cmd);
3139                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3140                         cmd->err_info->CommandStatus, DRIVER_OK,
3141                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3142                                 DID_PASSTHROUGH : DID_ERROR);
3143                 break;
3144         default:
3145                 printk(KERN_WARNING "cciss: cmd %p returned "
3146                        "unknown status %x\n", cmd,
3147                        cmd->err_info->CommandStatus);
3148                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3149                         cmd->err_info->CommandStatus, DRIVER_OK,
3150                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3151                                 DID_PASSTHROUGH : DID_ERROR);
3152         }
3153
3154 after_error_processing:
3155
3156         /* We need to return this command */
3157         if (retry_cmd) {
3158                 resend_cciss_cmd(h, cmd);
3159                 return;
3160         }
3161         cmd->rq->completion_data = cmd;
3162         blk_complete_request(cmd->rq);
3163 }
3164
3165 static inline u32 cciss_tag_contains_index(u32 tag)
3166 {
3167 #define DIRECT_LOOKUP_BIT 0x10
3168         return tag & DIRECT_LOOKUP_BIT;
3169 }
3170
3171 static inline u32 cciss_tag_to_index(u32 tag)
3172 {
3173 #define DIRECT_LOOKUP_SHIFT 5
3174         return tag >> DIRECT_LOOKUP_SHIFT;
3175 }
3176
3177 static inline u32 cciss_tag_discard_error_bits(u32 tag)
3178 {
3179 #define CCISS_ERROR_BITS 0x03
3180         return tag & ~CCISS_ERROR_BITS;
3181 }
3182
3183 static inline void cciss_mark_tag_indexed(u32 *tag)
3184 {
3185         *tag |= DIRECT_LOOKUP_BIT;
3186 }
3187
3188 static inline void cciss_set_tag_index(u32 *tag, u32 index)
3189 {
3190         *tag |= (index << DIRECT_LOOKUP_SHIFT);
3191 }
3192
3193 /*
3194  * Get a request and submit it to the controller.
3195  */
3196 static void do_cciss_request(struct request_queue *q)
3197 {
3198         ctlr_info_t *h = q->queuedata;
3199         CommandList_struct *c;
3200         sector_t start_blk;
3201         int seg;
3202         struct request *creq;
3203         u64bit temp64;
3204         struct scatterlist *tmp_sg;
3205         SGDescriptor_struct *curr_sg;
3206         drive_info_struct *drv;
3207         int i, dir;
3208         int sg_index = 0;
3209         int chained = 0;
3210
3211         /* We call start_io here in case there is a command waiting on the
3212          * queue that has not been sent.
3213          */
3214         if (blk_queue_plugged(q))
3215                 goto startio;
3216
3217       queue:
3218         creq = blk_peek_request(q);
3219         if (!creq)
3220                 goto startio;
3221
3222         BUG_ON(creq->nr_phys_segments > h->maxsgentries);
3223
3224         if ((c = cmd_alloc(h, 1)) == NULL)
3225                 goto full;
3226
3227         blk_start_request(creq);
3228
3229         tmp_sg = h->scatter_list[c->cmdindex];
3230         spin_unlock_irq(q->queue_lock);
3231
3232         c->cmd_type = CMD_RWREQ;
3233         c->rq = creq;
3234
3235         /* fill in the request */
3236         drv = creq->rq_disk->private_data;
3237         c->Header.ReplyQueue = 0;       /* unused in simple mode */
3238         /* got command from pool, so use the command block index instead */
3239         /* for direct lookups. */
3240         /* The first 2 bits are reserved for controller error reporting. */
3241         cciss_set_tag_index(&c->Header.Tag.lower, c->cmdindex);
3242         cciss_mark_tag_indexed(&c->Header.Tag.lower);
3243         memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3244         c->Request.CDBLen = 10; /* 12 byte commands not in FW yet; */
3245         c->Request.Type.Type = TYPE_CMD;        /* It is a command. */
3246         c->Request.Type.Attribute = ATTR_SIMPLE;
3247         c->Request.Type.Direction =
3248             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3249         c->Request.Timeout = 0; /* Don't time out */
3250         c->Request.CDB[0] =
3251             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3252         start_blk = blk_rq_pos(creq);
3253 #ifdef CCISS_DEBUG
3254         printk(KERN_DEBUG "ciss: sector =%d nr_sectors=%d\n",
3255                (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3256 #endif                          /* CCISS_DEBUG */
3257
3258         sg_init_table(tmp_sg, h->maxsgentries);
3259         seg = blk_rq_map_sg(q, creq, tmp_sg);
3260
3261         /* get the DMA records for the setup */
3262         if (c->Request.Type.Direction == XFER_READ)
3263                 dir = PCI_DMA_FROMDEVICE;
3264         else
3265                 dir = PCI_DMA_TODEVICE;
3266
3267         curr_sg = c->SG;
3268         sg_index = 0;
3269         chained = 0;
3270
3271         for (i = 0; i < seg; i++) {
3272                 if (((sg_index+1) == (h->max_cmd_sgentries)) &&
3273                         !chained && ((seg - i) > 1)) {
3274                         /* Point to next chain block. */
3275                         curr_sg = h->cmd_sg_list[c->cmdindex];
3276                         sg_index = 0;
3277                         chained = 1;
3278                 }
3279                 curr_sg[sg_index].Len = tmp_sg[i].length;
3280                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3281                                                 tmp_sg[i].offset,
3282                                                 tmp_sg[i].length, dir);
3283                 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3284                 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3285                 curr_sg[sg_index].Ext = 0;  /* we are not chaining */
3286                 ++sg_index;
3287         }
3288         if (chained)
3289                 cciss_map_sg_chain_block(h, c, h->cmd_sg_list[c->cmdindex],
3290                         (seg - (h->max_cmd_sgentries - 1)) *
3291                                 sizeof(SGDescriptor_struct));
3292
3293         /* track how many SG entries we are using */
3294         if (seg > h->maxSG)
3295                 h->maxSG = seg;
3296
3297 #ifdef CCISS_DEBUG
3298         printk(KERN_DEBUG "cciss: Submitting %ld sectors in %d segments "
3299                         "chained[%d]\n",
3300                         blk_rq_sectors(creq), seg, chained);
3301 #endif                          /* CCISS_DEBUG */
3302
3303         c->Header.SGTotal = seg + chained;
3304         if (seg <= h->max_cmd_sgentries)
3305                 c->Header.SGList = c->Header.SGTotal;
3306         else
3307                 c->Header.SGList = h->max_cmd_sgentries;
3308         set_performant_mode(h, c);
3309
3310         if (likely(creq->cmd_type == REQ_TYPE_FS)) {
3311                 if(h->cciss_read == CCISS_READ_10) {
3312                         c->Request.CDB[1] = 0;
3313                         c->Request.CDB[2] = (start_blk >> 24) & 0xff; /* MSB */
3314                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3315                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3316                         c->Request.CDB[5] = start_blk & 0xff;
3317                         c->Request.CDB[6] = 0; /* (sect >> 24) & 0xff; MSB */
3318                         c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3319                         c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3320                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3321                 } else {
3322                         u32 upper32 = upper_32_bits(start_blk);
3323
3324                         c->Request.CDBLen = 16;
3325                         c->Request.CDB[1]= 0;
3326                         c->Request.CDB[2]= (upper32 >> 24) & 0xff; /* MSB */
3327                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3328                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3329                         c->Request.CDB[5]= upper32 & 0xff;
3330                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3331                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3332                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3333                         c->Request.CDB[9]= start_blk & 0xff;
3334                         c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3335                         c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3336                         c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3337                         c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3338                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
3339                 }
3340         } else if (creq->cmd_type == REQ_TYPE_BLOCK_PC) {
3341                 c->Request.CDBLen = creq->cmd_len;
3342                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3343         } else {
3344                 printk(KERN_WARNING "cciss%d: bad request type %d\n", h->ctlr, creq->cmd_type);
3345                 BUG();
3346         }
3347
3348         spin_lock_irq(q->queue_lock);
3349
3350         addQ(&h->reqQ, c);
3351         h->Qdepth++;
3352         if (h->Qdepth > h->maxQsinceinit)
3353                 h->maxQsinceinit = h->Qdepth;
3354
3355         goto queue;
3356 full:
3357         blk_stop_queue(q);
3358 startio:
3359         /* We will already have the driver lock here so not need
3360          * to lock it.
3361          */
3362         start_io(h);
3363 }
3364
3365 static inline unsigned long get_next_completion(ctlr_info_t *h)
3366 {
3367         return h->access.command_completed(h);
3368 }
3369
3370 static inline int interrupt_pending(ctlr_info_t *h)
3371 {
3372         return h->access.intr_pending(h);
3373 }
3374
3375 static inline long interrupt_not_for_us(ctlr_info_t *h)
3376 {
3377         return !(h->msi_vector || h->msix_vector) &&
3378                 ((h->access.intr_pending(h) == 0) ||
3379                 (h->interrupts_enabled == 0));
3380 }
3381
3382 static inline int bad_tag(ctlr_info_t *h, u32 tag_index,
3383                         u32 raw_tag)
3384 {
3385         if (unlikely(tag_index >= h->nr_cmds)) {
3386                 dev_warn(&h->pdev->dev, "bad tag 0x%08x ignored.\n", raw_tag);
3387                 return 1;
3388         }
3389         return 0;
3390 }
3391
3392 static inline void finish_cmd(ctlr_info_t *h, CommandList_struct *c,
3393                                 u32 raw_tag)
3394 {
3395         removeQ(c);
3396         if (likely(c->cmd_type == CMD_RWREQ))
3397                 complete_command(h, c, 0);
3398         else if (c->cmd_type == CMD_IOCTL_PEND)
3399                 complete(c->waiting);
3400 #ifdef CONFIG_CISS_SCSI_TAPE
3401         else if (c->cmd_type == CMD_SCSI)
3402                 complete_scsi_command(c, 0, raw_tag);
3403 #endif
3404 }
3405
3406 static inline u32 next_command(ctlr_info_t *h)
3407 {
3408         u32 a;
3409
3410         if (unlikely(h->transMethod != CFGTBL_Trans_Performant))
3411                 return h->access.command_completed(h);
3412
3413         if ((*(h->reply_pool_head) & 1) == (h->reply_pool_wraparound)) {
3414                 a = *(h->reply_pool_head); /* Next cmd in ring buffer */
3415                 (h->reply_pool_head)++;
3416                 h->commands_outstanding--;
3417         } else {
3418                 a = FIFO_EMPTY;
3419         }
3420         /* Check for wraparound */
3421         if (h->reply_pool_head == (h->reply_pool + h->max_commands)) {
3422                 h->reply_pool_head = h->reply_pool;
3423                 h->reply_pool_wraparound ^= 1;
3424         }
3425         return a;
3426 }
3427
3428 /* process completion of an indexed ("direct lookup") command */
3429 static inline u32 process_indexed_cmd(ctlr_info_t *h, u32 raw_tag)
3430 {
3431         u32 tag_index;
3432         CommandList_struct *c;
3433
3434         tag_index = cciss_tag_to_index(raw_tag);
3435         if (bad_tag(h, tag_index, raw_tag))
3436                 return next_command(h);
3437         c = h->cmd_pool + tag_index;
3438         finish_cmd(h, c, raw_tag);
3439         return next_command(h);
3440 }
3441
3442 /* process completion of a non-indexed command */
3443 static inline u32 process_nonindexed_cmd(ctlr_info_t *h, u32 raw_tag)
3444 {
3445         u32 tag;
3446         CommandList_struct *c = NULL;
3447         struct hlist_node *tmp;
3448         __u32 busaddr_masked, tag_masked;
3449
3450         tag = cciss_tag_discard_error_bits(raw_tag);
3451         hlist_for_each_entry(c, tmp, &h->cmpQ, list) {
3452                 busaddr_masked = cciss_tag_discard_error_bits(c->busaddr);
3453                 tag_masked = cciss_tag_discard_error_bits(tag);
3454                 if (busaddr_masked == tag_masked) {
3455                         finish_cmd(h, c, raw_tag);
3456                         return next_command(h);
3457                 }
3458         }
3459         bad_tag(h, h->nr_cmds + 1, raw_tag);
3460         return next_command(h);
3461 }
3462
3463 static irqreturn_t do_cciss_intx(int irq, void *dev_id)
3464 {
3465         ctlr_info_t *h = dev_id;
3466         unsigned long flags;
3467         u32 raw_tag;
3468
3469         if (interrupt_not_for_us(h))
3470                 return IRQ_NONE;
3471         /*
3472          * If there are completed commands in the completion queue,
3473          * we had better do something about it.
3474          */
3475         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
3476         while (interrupt_pending(h)) {
3477                 raw_tag = get_next_completion(h);
3478                 while (raw_tag != FIFO_EMPTY) {
3479                         if (cciss_tag_contains_index(raw_tag))
3480                                 raw_tag = process_indexed_cmd(h, raw_tag);
3481                         else
3482                                 raw_tag = process_nonindexed_cmd(h, raw_tag);
3483                 }
3484         }
3485
3486         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
3487         return IRQ_HANDLED;
3488 }
3489
3490 /* Add a second interrupt handler for MSI/MSI-X mode. In this mode we never
3491  * check the interrupt pending register because it is not set.
3492  */
3493 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id)
3494 {
3495         ctlr_info_t *h = dev_id;
3496         unsigned long flags;
3497         u32 raw_tag;
3498
3499         if (interrupt_not_for_us(h))
3500                 return IRQ_NONE;
3501         /*
3502          * If there are completed commands in the completion queue,
3503          * we had better do something about it.
3504          */
3505         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
3506         raw_tag = get_next_completion(h);
3507         while (raw_tag != FIFO_EMPTY) {
3508                 if (cciss_tag_contains_index(raw_tag))
3509                         raw_tag = process_indexed_cmd(h, raw_tag);
3510                 else
3511                         raw_tag = process_nonindexed_cmd(h, raw_tag);
3512         }
3513
3514         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
3515         return IRQ_HANDLED;
3516 }
3517
3518 /**
3519  * add_to_scan_list() - add controller to rescan queue
3520  * @h:                Pointer to the controller.
3521  *
3522  * Adds the controller to the rescan queue if not already on the queue.
3523  *
3524  * returns 1 if added to the queue, 0 if skipped (could be on the
3525  * queue already, or the controller could be initializing or shutting
3526  * down).
3527  **/
3528 static int add_to_scan_list(struct ctlr_info *h)
3529 {
3530         struct ctlr_info *test_h;
3531         int found = 0;
3532         int ret = 0;
3533
3534         if (h->busy_initializing)
3535                 return 0;
3536
3537         if (!mutex_trylock(&h->busy_shutting_down))
3538                 return 0;
3539
3540         mutex_lock(&scan_mutex);
3541         list_for_each_entry(test_h, &scan_q, scan_list) {
3542                 if (test_h == h) {
3543                         found = 1;
3544                         break;
3545                 }
3546         }
3547         if (!found && !h->busy_scanning) {
3548                 INIT_COMPLETION(h->scan_wait);
3549                 list_add_tail(&h->scan_list, &scan_q);
3550                 ret = 1;
3551         }
3552         mutex_unlock(&scan_mutex);
3553         mutex_unlock(&h->busy_shutting_down);
3554
3555         return ret;
3556 }
3557
3558 /**
3559  * remove_from_scan_list() - remove controller from rescan queue
3560  * @h:                     Pointer to the controller.
3561  *
3562  * Removes the controller from the rescan queue if present. Blocks if
3563  * the controller is currently conducting a rescan.  The controller
3564  * can be in one of three states:
3565  * 1. Doesn't need a scan
3566  * 2. On the scan list, but not scanning yet (we remove it)
3567  * 3. Busy scanning (and not on the list). In this case we want to wait for
3568  *    the scan to complete to make sure the scanning thread for this
3569  *    controller is completely idle.
3570  **/
3571 static void remove_from_scan_list(struct ctlr_info *h)
3572 {
3573         struct ctlr_info *test_h, *tmp_h;
3574
3575         mutex_lock(&scan_mutex);
3576         list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3577                 if (test_h == h) { /* state 2. */
3578                         list_del(&h->scan_list);
3579                         complete_all(&h->scan_wait);
3580                         mutex_unlock(&scan_mutex);
3581                         return;
3582                 }
3583         }
3584         if (h->busy_scanning) { /* state 3. */
3585                 mutex_unlock(&scan_mutex);
3586                 wait_for_completion(&h->scan_wait);
3587         } else { /* state 1, nothing to do. */
3588                 mutex_unlock(&scan_mutex);
3589         }
3590 }
3591
3592 /**
3593  * scan_thread() - kernel thread used to rescan controllers
3594  * @data:        Ignored.
3595  *
3596  * A kernel thread used scan for drive topology changes on
3597  * controllers. The thread processes only one controller at a time
3598  * using a queue.  Controllers are added to the queue using
3599  * add_to_scan_list() and removed from the queue either after done
3600  * processing or using remove_from_scan_list().
3601  *
3602  * returns 0.
3603  **/
3604 static int scan_thread(void *data)
3605 {
3606         struct ctlr_info *h;
3607
3608         while (1) {
3609                 set_current_state(TASK_INTERRUPTIBLE);
3610                 schedule();
3611                 if (kthread_should_stop())
3612                         break;
3613
3614                 while (1) {
3615                         mutex_lock(&scan_mutex);
3616                         if (list_empty(&scan_q)) {
3617                                 mutex_unlock(&scan_mutex);
3618                                 break;
3619                         }
3620
3621                         h = list_entry(scan_q.next,
3622                                        struct ctlr_info,
3623                                        scan_list);
3624                         list_del(&h->scan_list);
3625                         h->busy_scanning = 1;
3626                         mutex_unlock(&scan_mutex);
3627
3628                         rebuild_lun_table(h, 0, 0);
3629                         complete_all(&h->scan_wait);
3630                         mutex_lock(&scan_mutex);
3631                         h->busy_scanning = 0;
3632                         mutex_unlock(&scan_mutex);
3633                 }
3634         }
3635
3636         return 0;
3637 }
3638
3639 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3640 {
3641         if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3642                 return 0;
3643
3644         switch (c->err_info->SenseInfo[12]) {
3645         case STATE_CHANGED:
3646                 printk(KERN_WARNING "cciss%d: a state change "
3647                         "detected, command retried\n", h->ctlr);
3648                 return 1;
3649         break;
3650         case LUN_FAILED:
3651                 printk(KERN_WARNING "cciss%d: LUN failure "
3652                         "detected, action required\n", h->ctlr);
3653                 return 1;
3654         break;
3655         case REPORT_LUNS_CHANGED:
3656                 printk(KERN_WARNING "cciss%d: report LUN data "
3657                         "changed\n", h->ctlr);
3658         /*
3659          * Here, we could call add_to_scan_list and wake up the scan thread,
3660          * except that it's quite likely that we will get more than one
3661          * REPORT_LUNS_CHANGED condition in quick succession, which means
3662          * that those which occur after the first one will likely happen
3663          * *during* the scan_thread's rescan.  And the rescan code is not
3664          * robust enough to restart in the middle, undoing what it has already
3665          * done, and it's not clear that it's even possible to do this, since
3666          * part of what it does is notify the block layer, which starts
3667          * doing it's own i/o to read partition tables and so on, and the
3668          * driver doesn't have visibility to know what might need undoing.
3669          * In any event, if possible, it is horribly complicated to get right
3670          * so we just don't do it for now.
3671          *
3672          * Note: this REPORT_LUNS_CHANGED condition only occurs on the MSA2012.
3673          */
3674                 return 1;
3675         break;
3676         case POWER_OR_RESET:
3677                 printk(KERN_WARNING "cciss%d: a power on "
3678                         "or device reset detected\n", h->ctlr);
3679                 return 1;
3680         break;
3681         case UNIT_ATTENTION_CLEARED:
3682                 printk(KERN_WARNING "cciss%d: unit attention "
3683                     "cleared by another initiator\n", h->ctlr);
3684                 return 1;
3685         break;
3686         default:
3687                 printk(KERN_WARNING "cciss%d: unknown "
3688                         "unit attention detected\n", h->ctlr);
3689                                 return 1;
3690         }
3691 }
3692
3693 /*
3694  *  We cannot read the structure directly, for portability we must use
3695  *   the io functions.
3696  *   This is for debug only.
3697  */
3698 #ifdef CCISS_DEBUG
3699 static void print_cfg_table(CfgTable_struct *tb)
3700 {
3701         int i;
3702         char temp_name[17];
3703
3704         printk("Controller Configuration information\n");
3705         printk("------------------------------------\n");
3706         for (i = 0; i < 4; i++)
3707                 temp_name[i] = readb(&(tb->Signature[i]));
3708         temp_name[4] = '\0';
3709         printk("   Signature = %s\n", temp_name);
3710         printk("   Spec Number = %d\n", readl(&(tb->SpecValence)));
3711         printk("   Transport methods supported = 0x%x\n",
3712                readl(&(tb->TransportSupport)));
3713         printk("   Transport methods active = 0x%x\n",
3714                readl(&(tb->TransportActive)));
3715         printk("   Requested transport Method = 0x%x\n",
3716                readl(&(tb->HostWrite.TransportRequest)));
3717         printk("   Coalesce Interrupt Delay = 0x%x\n",
3718                readl(&(tb->HostWrite.CoalIntDelay)));
3719         printk("   Coalesce Interrupt Count = 0x%x\n",
3720                readl(&(tb->HostWrite.CoalIntCount)));
3721         printk("   Max outstanding commands = 0x%d\n",
3722                readl(&(tb->CmdsOutMax)));
3723         printk("   Bus Types = 0x%x\n", readl(&(tb->BusTypes)));
3724         for (i = 0; i < 16; i++)
3725                 temp_name[i] = readb(&(tb->ServerName[i]));
3726         temp_name[16] = '\0';
3727         printk("   Server Name = %s\n", temp_name);
3728         printk("   Heartbeat Counter = 0x%x\n\n\n", readl(&(tb->HeartBeat)));
3729 }
3730 #endif                          /* CCISS_DEBUG */
3731
3732 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3733 {
3734         int i, offset, mem_type, bar_type;
3735         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3736                 return 0;
3737         offset = 0;
3738         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3739                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3740                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3741                         offset += 4;
3742                 else {
3743                         mem_type = pci_resource_flags(pdev, i) &
3744                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3745                         switch (mem_type) {
3746                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
3747                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3748                                 offset += 4;    /* 32 bit */
3749                                 break;
3750                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
3751                                 offset += 8;
3752                                 break;
3753                         default:        /* reserved in PCI 2.2 */
3754                                 printk(KERN_WARNING
3755                                        "Base address is invalid\n");
3756                                 return -1;
3757                                 break;
3758                         }
3759                 }
3760                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3761                         return i + 1;
3762         }
3763         return -1;
3764 }
3765
3766 /* Fill in bucket_map[], given nsgs (the max number of
3767  * scatter gather elements supported) and bucket[],
3768  * which is an array of 8 integers.  The bucket[] array
3769  * contains 8 different DMA transfer sizes (in 16
3770  * byte increments) which the controller uses to fetch
3771  * commands.  This function fills in bucket_map[], which
3772  * maps a given number of scatter gather elements to one of
3773  * the 8 DMA transfer sizes.  The point of it is to allow the
3774  * controller to only do as much DMA as needed to fetch the
3775  * command, with the DMA transfer size encoded in the lower
3776  * bits of the command address.
3777  */
3778 static void  calc_bucket_map(int bucket[], int num_buckets,
3779         int nsgs, int *bucket_map)
3780 {
3781         int i, j, b, size;
3782
3783         /* even a command with 0 SGs requires 4 blocks */
3784 #define MINIMUM_TRANSFER_BLOCKS 4
3785 #define NUM_BUCKETS 8
3786         /* Note, bucket_map must have nsgs+1 entries. */
3787         for (i = 0; i <= nsgs; i++) {
3788                 /* Compute size of a command with i SG entries */
3789                 size = i + MINIMUM_TRANSFER_BLOCKS;
3790                 b = num_buckets; /* Assume the biggest bucket */
3791                 /* Find the bucket that is just big enough */
3792                 for (j = 0; j < 8; j++) {
3793                         if (bucket[j] >= size) {
3794                                 b = j;
3795                                 break;
3796                         }
3797                 }
3798                 /* for a command with i SG entries, use bucket b. */
3799                 bucket_map[i] = b;
3800         }
3801 }
3802
3803 static void
3804 cciss_put_controller_into_performant_mode(ctlr_info_t *h)
3805 {
3806         int l = 0;
3807         __u32 trans_support;
3808         __u32 trans_offset;
3809                         /*
3810                          *  5 = 1 s/g entry or 4k
3811                          *  6 = 2 s/g entry or 8k
3812                          *  8 = 4 s/g entry or 16k
3813                          * 10 = 6 s/g entry or 24k
3814                          */
3815         int bft[8] = { 5, 6, 8, 10, 12, 20, 28, MAXSGENTRIES + 4};
3816         unsigned long register_value;
3817
3818         BUILD_BUG_ON(28 > MAXSGENTRIES + 4);
3819
3820         /* Attempt to put controller into performant mode if supported */
3821         /* Does board support performant mode? */
3822         trans_support = readl(&(h->cfgtable->TransportSupport));
3823         if (!(trans_support & PERFORMANT_MODE))
3824                 return;
3825
3826         printk(KERN_WARNING "cciss%d: Placing controller into "
3827                                 "performant mode\n", h->ctlr);
3828         /* Performant mode demands commands on a 32 byte boundary
3829          * pci_alloc_consistent aligns on page boundarys already.
3830          * Just need to check if divisible by 32
3831          */
3832         if ((sizeof(CommandList_struct) % 32) != 0) {
3833                 printk(KERN_WARNING "%s %d %s\n",
3834                         "cciss info: command size[",
3835                         (int)sizeof(CommandList_struct),
3836                         "] not divisible by 32, no performant mode..\n");
3837                 return;
3838         }
3839
3840         /* Performant mode ring buffer and supporting data structures */
3841         h->reply_pool = (__u64 *)pci_alloc_consistent(
3842                 h->pdev, h->max_commands * sizeof(__u64),
3843                 &(h->reply_pool_dhandle));
3844
3845         /* Need a block fetch table for performant mode */
3846         h->blockFetchTable = kmalloc(((h->maxsgentries+1) *
3847                 sizeof(__u32)), GFP_KERNEL);
3848
3849         if ((h->reply_pool == NULL) || (h->blockFetchTable == NULL))
3850                 goto clean_up;
3851
3852         h->reply_pool_wraparound = 1; /* spec: init to 1 */
3853
3854         /* Controller spec: zero out this buffer. */
3855         memset(h->reply_pool, 0, h->max_commands * sizeof(__u64));
3856         h->reply_pool_head = h->reply_pool;
3857
3858         trans_offset = readl(&(h->cfgtable->TransMethodOffset));
3859         calc_bucket_map(bft, ARRAY_SIZE(bft), h->maxsgentries,
3860                                 h->blockFetchTable);
3861         writel(bft[0], &h->transtable->BlockFetch0);
3862         writel(bft[1], &h->transtable->BlockFetch1);
3863         writel(bft[2], &h->transtable->BlockFetch2);
3864         writel(bft[3], &h->transtable->BlockFetch3);
3865         writel(bft[4], &h->transtable->BlockFetch4);
3866         writel(bft[5], &h->transtable->BlockFetch5);
3867         writel(bft[6], &h->transtable->BlockFetch6);
3868         writel(bft[7], &h->transtable->BlockFetch7);
3869
3870         /* size of controller ring buffer */
3871         writel(h->max_commands, &h->transtable->RepQSize);
3872         writel(1, &h->transtable->RepQCount);
3873         writel(0, &h->transtable->RepQCtrAddrLow32);
3874         writel(0, &h->transtable->RepQCtrAddrHigh32);
3875         writel(h->reply_pool_dhandle, &h->transtable->RepQAddr0Low32);
3876         writel(0, &h->transtable->RepQAddr0High32);
3877         writel(CFGTBL_Trans_Performant,
3878                         &(h->cfgtable->HostWrite.TransportRequest));
3879
3880         h->transMethod = CFGTBL_Trans_Performant;
3881         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
3882         /* under certain very rare conditions, this can take awhile.
3883          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3884          * as we enter this code.) */
3885         for (l = 0; l < MAX_CONFIG_WAIT; l++) {
3886                 register_value = readl(h->vaddr + SA5_DOORBELL);
3887                 if (!(register_value & CFGTBL_ChangeReq))
3888                         break;
3889                 /* delay and try again */
3890                 set_current_state(TASK_INTERRUPTIBLE);
3891                 schedule_timeout(10);
3892         }
3893         register_value = readl(&(h->cfgtable->TransportActive));
3894         if (!(register_value & CFGTBL_Trans_Performant)) {
3895                 printk(KERN_WARNING "cciss: unable to get board into"
3896                                         " performant mode\n");
3897                 return;
3898         }
3899
3900         /* Change the access methods to the performant access methods */
3901         h->access = SA5_performant_access;
3902
3903         return;
3904 clean_up:
3905         kfree(h->blockFetchTable);
3906         if (h->reply_pool)
3907                 pci_free_consistent(h->pdev,
3908                                 h->max_commands * sizeof(__u64),
3909                                 h->reply_pool,
3910                                 h->reply_pool_dhandle);
3911         return;
3912
3913 } /* cciss_put_controller_into_performant_mode */
3914
3915 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
3916  * controllers that are capable. If not, we use IO-APIC mode.
3917  */
3918
3919 static void __devinit cciss_interrupt_mode(ctlr_info_t *c,
3920                                            struct pci_dev *pdev, __u32 board_id)
3921 {
3922 #ifdef CONFIG_PCI_MSI
3923         int err;
3924         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
3925         {0, 2}, {0, 3}
3926         };
3927
3928         /* Some boards advertise MSI but don't really support it */
3929         if ((board_id == 0x40700E11) ||
3930             (board_id == 0x40800E11) ||
3931             (board_id == 0x40820E11) || (board_id == 0x40830E11))
3932                 goto default_int_mode;
3933
3934         if (pci_find_capability(pdev, PCI_CAP_ID_MSIX)) {
3935                 err = pci_enable_msix(pdev, cciss_msix_entries, 4);
3936                 if (!err) {
3937                         c->intr[0] = cciss_msix_entries[0].vector;
3938                         c->intr[1] = cciss_msix_entries[1].vector;
3939                         c->intr[2] = cciss_msix_entries[2].vector;
3940                         c->intr[3] = cciss_msix_entries[3].vector;
3941                         c->msix_vector = 1;
3942                         return;
3943                 }
3944                 if (err > 0) {
3945                         printk(KERN_WARNING "cciss: only %d MSI-X vectors "
3946                                "available\n", err);
3947                         goto default_int_mode;
3948                 } else {
3949                         printk(KERN_WARNING "cciss: MSI-X init failed %d\n",
3950                                err);
3951                         goto default_int_mode;
3952                 }
3953         }
3954         if (pci_find_capability(pdev, PCI_CAP_ID_MSI)) {
3955                 if (!pci_enable_msi(pdev)) {
3956                         c->msi_vector = 1;
3957                 } else {
3958                         printk(KERN_WARNING "cciss: MSI init failed\n");
3959                 }
3960         }
3961 default_int_mode:
3962 #endif                          /* CONFIG_PCI_MSI */
3963         /* if we get here we're going to use the default interrupt mode */
3964         c->intr[PERF_MODE_INT] = pdev->irq;
3965         return;
3966 }
3967
3968 static int __devinit cciss_pci_init(ctlr_info_t *c, struct pci_dev *pdev)
3969 {
3970         ushort subsystem_vendor_id, subsystem_device_id, command;
3971         __u32 board_id, scratchpad = 0;
3972         __u64 cfg_offset;
3973         __u32 cfg_base_addr;
3974         __u64 cfg_base_addr_index;
3975         int i, prod_index, err;
3976         __u32 trans_offset;
3977
3978         subsystem_vendor_id = pdev->subsystem_vendor;
3979         subsystem_device_id = pdev->subsystem_device;
3980         board_id = (((__u32) (subsystem_device_id << 16) & 0xffff0000) |
3981                     subsystem_vendor_id);
3982
3983         for (i = 0; i < ARRAY_SIZE(products); i++) {
3984                 /* Stand aside for hpsa driver on request */
3985                 if (cciss_allow_hpsa && products[i].board_id == HPSA_BOUNDARY)
3986                         return -ENODEV;
3987                 if (board_id == products[i].board_id)
3988                         break;
3989         }
3990         prod_index = i;
3991         if (prod_index == ARRAY_SIZE(products)) {
3992                 dev_warn(&pdev->dev,
3993                         "unrecognized board ID: 0x%08lx, ignoring.\n",
3994                         (unsigned long) board_id);
3995                 return -ENODEV;
3996         }
3997
3998         /* check to see if controller has been disabled */
3999         /* BEFORE trying to enable it */
4000         (void)pci_read_config_word(pdev, PCI_COMMAND, &command);
4001         if (!(command & 0x02)) {
4002                 printk(KERN_WARNING
4003                        "cciss: controller appears to be disabled\n");
4004                 return -ENODEV;
4005         }
4006
4007         err = pci_enable_device(pdev);
4008         if (err) {
4009                 printk(KERN_ERR "cciss: Unable to Enable PCI device\n");
4010                 return err;
4011         }
4012
4013         err = pci_request_regions(pdev, "cciss");
4014         if (err) {
4015                 printk(KERN_ERR "cciss: Cannot obtain PCI resources, "
4016                        "aborting\n");
4017                 return err;
4018         }
4019
4020 #ifdef CCISS_DEBUG
4021         printk("command = %x\n", command);
4022         printk("irq = %x\n", pdev->irq);
4023         printk("board_id = %x\n", board_id);
4024 #endif                          /* CCISS_DEBUG */
4025
4026 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
4027  * else we use the IO-APIC interrupt assigned to us by system ROM.
4028  */
4029         cciss_interrupt_mode(c, pdev, board_id);
4030
4031         /* find the memory BAR */
4032         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
4033                 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM)
4034                         break;
4035         }
4036         if (i == DEVICE_COUNT_RESOURCE) {
4037                 printk(KERN_WARNING "cciss: No memory BAR found\n");
4038                 err = -ENODEV;
4039                 goto err_out_free_res;
4040         }
4041
4042         c->paddr = pci_resource_start(pdev, i); /* addressing mode bits
4043                                                  * already removed
4044                                                  */
4045
4046 #ifdef CCISS_DEBUG
4047         printk("address 0 = %lx\n", c->paddr);
4048 #endif                          /* CCISS_DEBUG */
4049         c->vaddr = remap_pci_mem(c->paddr, 0x250);
4050
4051         /* Wait for the board to become ready.  (PCI hotplug needs this.)
4052          * We poll for up to 120 secs, once per 100ms. */
4053         for (i = 0; i < 1200; i++) {
4054                 scratchpad = readl(c->vaddr + SA5_SCRATCHPAD_OFFSET);
4055                 if (scratchpad == CCISS_FIRMWARE_READY)
4056                         break;
4057                 set_current_state(TASK_INTERRUPTIBLE);
4058                 schedule_timeout(msecs_to_jiffies(100));        /* wait 100ms */
4059         }
4060         if (scratchpad != CCISS_FIRMWARE_READY) {
4061                 printk(KERN_WARNING "cciss: Board not ready.  Timed out.\n");
4062                 err = -ENODEV;
4063                 goto err_out_free_res;
4064         }
4065
4066         /* get the address index number */
4067         cfg_base_addr = readl(c->vaddr + SA5_CTCFG_OFFSET);
4068         cfg_base_addr &= (__u32) 0x0000ffff;
4069 #ifdef CCISS_DEBUG
4070         printk("cfg base address = %x\n", cfg_base_addr);
4071 #endif                          /* CCISS_DEBUG */
4072         cfg_base_addr_index = find_PCI_BAR_index(pdev, cfg_base_addr);
4073 #ifdef CCISS_DEBUG
4074         printk("cfg base address index = %llx\n",
4075                 (unsigned long long)cfg_base_addr_index);
4076 #endif                          /* CCISS_DEBUG */
4077         if (cfg_base_addr_index == -1) {
4078                 printk(KERN_WARNING "cciss: Cannot find cfg_base_addr_index\n");
4079                 err = -ENODEV;
4080                 goto err_out_free_res;
4081         }
4082
4083         cfg_offset = readl(c->vaddr + SA5_CTMEM_OFFSET);
4084 #ifdef CCISS_DEBUG
4085         printk("cfg offset = %llx\n", (unsigned long long)cfg_offset);
4086 #endif                          /* CCISS_DEBUG */
4087         c->cfgtable = remap_pci_mem(pci_resource_start(pdev,
4088                                                        cfg_base_addr_index) +
4089                                     cfg_offset, sizeof(CfgTable_struct));
4090         /* Find performant mode table. */
4091         trans_offset = readl(&(c->cfgtable->TransMethodOffset));
4092         c->transtable = remap_pci_mem(pci_resource_start(pdev,
4093                 cfg_base_addr_index) + cfg_offset+trans_offset,
4094                 sizeof(*c->transtable));
4095         c->board_id = board_id;
4096
4097         #ifdef CCISS_DEBUG
4098                 print_cfg_table(c->cfgtable);
4099         #endif                          /* CCISS_DEBUG */
4100
4101         /* Some controllers support Zero Memory Raid (ZMR).
4102          * When configured in ZMR mode the number of supported
4103          * commands drops to 64. So instead of just setting an
4104          * arbitrary value we make the driver a little smarter.
4105          * We read the config table to tell us how many commands
4106          * are supported on the controller then subtract 4 to
4107          * leave a little room for ioctl calls.
4108          */
4109         c->max_commands = readl(&(c->cfgtable->MaxPerformantModeCommands));
4110         c->maxsgentries = readl(&(c->cfgtable->MaxSGElements));
4111
4112         /*
4113          * Limit native command to 32 s/g elements to save dma'able memory.
4114          * Howvever spec says if 0, use 31
4115          */
4116
4117         c->max_cmd_sgentries = 31;
4118         if (c->maxsgentries > 512) {
4119                 c->max_cmd_sgentries = 32;
4120                 c->chainsize = c->maxsgentries - c->max_cmd_sgentries + 1;
4121                 c->maxsgentries -= 1;   /* account for chain pointer */
4122         } else {
4123                 c->maxsgentries = 31;   /* Default to traditional value */
4124                 c->chainsize = 0;       /* traditional */
4125         }
4126
4127         c->product_name = products[prod_index].product_name;
4128         c->access = *(products[prod_index].access);
4129         c->nr_cmds = c->max_commands - 4;
4130         if ((readb(&c->cfgtable->Signature[0]) != 'C') ||
4131             (readb(&c->cfgtable->Signature[1]) != 'I') ||
4132             (readb(&c->cfgtable->Signature[2]) != 'S') ||
4133             (readb(&c->cfgtable->Signature[3]) != 'S')) {
4134                 printk("Does not appear to be a valid CISS config table\n");
4135                 err = -ENODEV;
4136                 goto err_out_free_res;
4137         }
4138 #ifdef CONFIG_X86
4139         {
4140                 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
4141                 __u32 prefetch;
4142                 prefetch = readl(&(c->cfgtable->SCSI_Prefetch));
4143                 prefetch |= 0x100;
4144                 writel(prefetch, &(c->cfgtable->SCSI_Prefetch));
4145         }
4146 #endif
4147
4148         /* Disabling DMA prefetch and refetch for the P600.
4149          * An ASIC bug may result in accesses to invalid memory addresses.
4150          * We've disabled prefetch for some time now. Testing with XEN
4151          * kernels revealed a bug in the refetch if dom0 resides on a P600.
4152          */
4153         if(board_id == 0x3225103C) {
4154                         __u32 dma_prefetch;
4155                 __u32 dma_refetch;
4156                 dma_prefetch = readl(c->vaddr + I2O_DMA1_CFG);
4157                 dma_prefetch |= 0x8000;
4158                 writel(dma_prefetch, c->vaddr + I2O_DMA1_CFG);
4159                 pci_read_config_dword(pdev, PCI_COMMAND_PARITY, &dma_refetch);
4160                 dma_refetch |= 0x1;
4161                 pci_write_config_dword(pdev, PCI_COMMAND_PARITY, dma_refetch);
4162         }
4163
4164 #ifdef CCISS_DEBUG
4165         printk(KERN_WARNING "Trying to put board into Performant mode\n");
4166 #endif                          /* CCISS_DEBUG */
4167         cciss_put_controller_into_performant_mode(c);
4168         return 0;
4169
4170 err_out_free_res:
4171         /*
4172          * Deliberately omit pci_disable_device(): it does something nasty to
4173          * Smart Array controllers that pci_enable_device does not undo
4174          */
4175         pci_release_regions(pdev);
4176         return err;
4177 }
4178
4179 /* Function to find the first free pointer into our hba[] array
4180  * Returns -1 if no free entries are left.
4181  */
4182 static int alloc_cciss_hba(void)
4183 {
4184         int i;
4185
4186         for (i = 0; i < MAX_CTLR; i++) {
4187                 if (!hba[i]) {
4188                         ctlr_info_t *p;
4189
4190                         p = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
4191                         if (!p)
4192                                 goto Enomem;
4193                         hba[i] = p;
4194                         return i;
4195                 }
4196         }
4197         printk(KERN_WARNING "cciss: This driver supports a maximum"
4198                " of %d controllers.\n", MAX_CTLR);
4199         return -1;
4200 Enomem:
4201         printk(KERN_ERR "cciss: out of memory.\n");
4202         return -1;
4203 }
4204
4205 static void free_hba(int n)
4206 {
4207         ctlr_info_t *h = hba[n];
4208         int i;
4209
4210         hba[n] = NULL;
4211         for (i = 0; i < h->highest_lun + 1; i++)
4212                 if (h->gendisk[i] != NULL)
4213                         put_disk(h->gendisk[i]);
4214         kfree(h);
4215 }
4216
4217 /* Send a message CDB to the firmware. */
4218 static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
4219 {
4220         typedef struct {
4221                 CommandListHeader_struct CommandHeader;
4222                 RequestBlock_struct Request;
4223                 ErrDescriptor_struct ErrorDescriptor;
4224         } Command;
4225         static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
4226         Command *cmd;
4227         dma_addr_t paddr64;
4228         uint32_t paddr32, tag;
4229         void __iomem *vaddr;
4230         int i, err;
4231
4232         vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
4233         if (vaddr == NULL)
4234                 return -ENOMEM;
4235
4236         /* The Inbound Post Queue only accepts 32-bit physical addresses for the
4237            CCISS commands, so they must be allocated from the lower 4GiB of
4238            memory. */
4239         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
4240         if (err) {
4241                 iounmap(vaddr);
4242                 return -ENOMEM;
4243         }
4244
4245         cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
4246         if (cmd == NULL) {
4247                 iounmap(vaddr);
4248                 return -ENOMEM;
4249         }
4250
4251         /* This must fit, because of the 32-bit consistent DMA mask.  Also,
4252            although there's no guarantee, we assume that the address is at
4253            least 4-byte aligned (most likely, it's page-aligned). */
4254         paddr32 = paddr64;
4255
4256         cmd->CommandHeader.ReplyQueue = 0;
4257         cmd->CommandHeader.SGList = 0;
4258         cmd->CommandHeader.SGTotal = 0;
4259         cmd->CommandHeader.Tag.lower = paddr32;
4260         cmd->CommandHeader.Tag.upper = 0;
4261         memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
4262
4263         cmd->Request.CDBLen = 16;
4264         cmd->Request.Type.Type = TYPE_MSG;
4265         cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
4266         cmd->Request.Type.Direction = XFER_NONE;
4267         cmd->Request.Timeout = 0; /* Don't time out */
4268         cmd->Request.CDB[0] = opcode;
4269         cmd->Request.CDB[1] = type;
4270         memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
4271
4272         cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
4273         cmd->ErrorDescriptor.Addr.upper = 0;
4274         cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
4275
4276         writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
4277
4278         for (i = 0; i < 10; i++) {
4279                 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
4280                 if ((tag & ~3) == paddr32)
4281                         break;
4282                 schedule_timeout_uninterruptible(HZ);
4283         }
4284
4285         iounmap(vaddr);
4286
4287         /* we leak the DMA buffer here ... no choice since the controller could
4288            still complete the command. */
4289         if (i == 10) {
4290                 printk(KERN_ERR "cciss: controller message %02x:%02x timed out\n",
4291                         opcode, type);
4292                 return -ETIMEDOUT;
4293         }
4294
4295         pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
4296
4297         if (tag & 2) {
4298                 printk(KERN_ERR "cciss: controller message %02x:%02x failed\n",
4299                         opcode, type);
4300                 return -EIO;
4301         }
4302
4303         printk(KERN_INFO "cciss: controller message %02x:%02x succeeded\n",
4304                 opcode, type);
4305         return 0;
4306 }
4307
4308 #define cciss_soft_reset_controller(p) cciss_message(p, 1, 0)
4309 #define cciss_noop(p) cciss_message(p, 3, 0)
4310
4311 static __devinit int cciss_reset_msi(struct pci_dev *pdev)
4312 {
4313 /* the #defines are stolen from drivers/pci/msi.h. */
4314 #define msi_control_reg(base)           (base + PCI_MSI_FLAGS)
4315 #define PCI_MSIX_FLAGS_ENABLE           (1 << 15)
4316
4317         int pos;
4318         u16 control = 0;
4319
4320         pos = pci_find_capability(pdev, PCI_CAP_ID_MSI);
4321         if (pos) {
4322                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4323                 if (control & PCI_MSI_FLAGS_ENABLE) {
4324                         printk(KERN_INFO "cciss: resetting MSI\n");
4325                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSI_FLAGS_ENABLE);
4326                 }
4327         }
4328
4329         pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
4330         if (pos) {
4331                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4332                 if (control & PCI_MSIX_FLAGS_ENABLE) {
4333                         printk(KERN_INFO "cciss: resetting MSI-X\n");
4334                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSIX_FLAGS_ENABLE);
4335                 }
4336         }
4337
4338         return 0;
4339 }
4340
4341 /* This does a hard reset of the controller using PCI power management
4342  * states. */
4343 static __devinit int cciss_hard_reset_controller(struct pci_dev *pdev)
4344 {
4345         u16 pmcsr, saved_config_space[32];
4346         int i, pos;
4347
4348         printk(KERN_INFO "cciss: using PCI PM to reset controller\n");
4349
4350         /* This is very nearly the same thing as
4351
4352            pci_save_state(pci_dev);
4353            pci_set_power_state(pci_dev, PCI_D3hot);
4354            pci_set_power_state(pci_dev, PCI_D0);
4355            pci_restore_state(pci_dev);
4356
4357            but we can't use these nice canned kernel routines on
4358            kexec, because they also check the MSI/MSI-X state in PCI
4359            configuration space and do the wrong thing when it is
4360            set/cleared.  Also, the pci_save/restore_state functions
4361            violate the ordering requirements for restoring the
4362            configuration space from the CCISS document (see the
4363            comment below).  So we roll our own .... */
4364
4365         for (i = 0; i < 32; i++)
4366                 pci_read_config_word(pdev, 2*i, &saved_config_space[i]);
4367
4368         pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4369         if (pos == 0) {
4370                 printk(KERN_ERR "cciss_reset_controller: PCI PM not supported\n");
4371                 return -ENODEV;
4372         }
4373
4374         /* Quoting from the Open CISS Specification: "The Power
4375          * Management Control/Status Register (CSR) controls the power
4376          * state of the device.  The normal operating state is D0,
4377          * CSR=00h.  The software off state is D3, CSR=03h.  To reset
4378          * the controller, place the interface device in D3 then to
4379          * D0, this causes a secondary PCI reset which will reset the
4380          * controller." */
4381
4382         /* enter the D3hot power management state */
4383         pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4384         pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4385         pmcsr |= PCI_D3hot;
4386         pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4387
4388         schedule_timeout_uninterruptible(HZ >> 1);
4389
4390         /* enter the D0 power management state */
4391         pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4392         pmcsr |= PCI_D0;
4393         pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4394
4395         schedule_timeout_uninterruptible(HZ >> 1);
4396
4397         /* Restore the PCI configuration space.  The Open CISS
4398          * Specification says, "Restore the PCI Configuration
4399          * Registers, offsets 00h through 60h. It is important to
4400          * restore the command register, 16-bits at offset 04h,
4401          * last. Do not restore the configuration status register,
4402          * 16-bits at offset 06h."  Note that the offset is 2*i. */
4403         for (i = 0; i < 32; i++) {
4404                 if (i == 2 || i == 3)
4405                         continue;
4406                 pci_write_config_word(pdev, 2*i, saved_config_space[i]);
4407         }
4408         wmb();
4409         pci_write_config_word(pdev, 4, saved_config_space[2]);
4410
4411         return 0;
4412 }
4413
4414 /*
4415  *  This is it.  Find all the controllers and register them.  I really hate
4416  *  stealing all these major device numbers.
4417  *  returns the number of block devices registered.
4418  */
4419 static int __devinit cciss_init_one(struct pci_dev *pdev,
4420                                     const struct pci_device_id *ent)
4421 {
4422         int i;
4423         int j = 0;
4424         int k = 0;
4425         int rc;
4426         int dac, return_code;
4427         InquiryData_struct *inq_buff;
4428
4429         if (reset_devices) {
4430                 /* Reset the controller with a PCI power-cycle */
4431                 if (cciss_hard_reset_controller(pdev) || cciss_reset_msi(pdev))
4432                         return -ENODEV;
4433
4434                 /* Now try to get the controller to respond to a no-op. Some
4435                    devices (notably the HP Smart Array 5i Controller) need
4436                    up to 30 seconds to respond. */
4437                 for (i=0; i<30; i++) {
4438                         if (cciss_noop(pdev) == 0)
4439                                 break;
4440
4441                         schedule_timeout_uninterruptible(HZ);
4442                 }
4443                 if (i == 30) {
4444                         printk(KERN_ERR "cciss: controller seems dead\n");
4445                         return -EBUSY;
4446                 }
4447         }
4448
4449         i = alloc_cciss_hba();
4450         if (i < 0)
4451                 return -1;
4452         hba[i]->busy_initializing = 1;
4453         INIT_HLIST_HEAD(&hba[i]->cmpQ);
4454         INIT_HLIST_HEAD(&hba[i]->reqQ);
4455         mutex_init(&hba[i]->busy_shutting_down);
4456
4457         if (cciss_pci_init(hba[i], pdev) != 0)
4458                 goto clean_no_release_regions;
4459
4460         sprintf(hba[i]->devname, "cciss%d", i);
4461         hba[i]->ctlr = i;
4462         hba[i]->pdev = pdev;
4463
4464         init_completion(&hba[i]->scan_wait);
4465
4466         if (cciss_create_hba_sysfs_entry(hba[i]))
4467                 goto clean0;
4468
4469         /* configure PCI DMA stuff */
4470         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
4471                 dac = 1;
4472         else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
4473                 dac = 0;
4474         else {
4475                 printk(KERN_ERR "cciss: no suitable DMA available\n");
4476                 goto clean1;
4477         }
4478
4479         /*
4480          * register with the major number, or get a dynamic major number
4481          * by passing 0 as argument.  This is done for greater than
4482          * 8 controller support.
4483          */
4484         if (i < MAX_CTLR_ORIG)
4485                 hba[i]->major = COMPAQ_CISS_MAJOR + i;
4486         rc = register_blkdev(hba[i]->major, hba[i]->devname);
4487         if (rc == -EBUSY || rc == -EINVAL) {
4488                 printk(KERN_ERR
4489                        "cciss:  Unable to get major number %d for %s "
4490                        "on hba %d\n", hba[i]->major, hba[i]->devname, i);
4491                 goto clean1;
4492         } else {
4493                 if (i >= MAX_CTLR_ORIG)
4494                         hba[i]->major = rc;
4495         }
4496
4497         /* make sure the board interrupts are off */
4498         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_OFF);
4499         if (hba[i]->msi_vector || hba[i]->msix_vector) {
4500                 if (request_irq(hba[i]->intr[PERF_MODE_INT],
4501                                 do_cciss_msix_intr,
4502                                 IRQF_DISABLED, hba[i]->devname, hba[i])) {
4503                         printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
4504                                hba[i]->intr[PERF_MODE_INT], hba[i]->devname);
4505                         goto clean2;
4506                 }
4507         } else {
4508                 if (request_irq(hba[i]->intr[PERF_MODE_INT], do_cciss_intx,
4509                                 IRQF_DISABLED, hba[i]->devname, hba[i])) {
4510                         printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
4511                                hba[i]->intr[PERF_MODE_INT], hba[i]->devname);
4512                         goto clean2;
4513                 }
4514         }
4515
4516         printk(KERN_INFO "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
4517                hba[i]->devname, pdev->device, pci_name(pdev),
4518                hba[i]->intr[PERF_MODE_INT], dac ? "" : " not");
4519
4520         hba[i]->cmd_pool_bits =
4521             kmalloc(DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4522                         * sizeof(unsigned long), GFP_KERNEL);
4523         hba[i]->cmd_pool = (CommandList_struct *)
4524             pci_alloc_consistent(hba[i]->pdev,
4525                     hba[i]->nr_cmds * sizeof(CommandList_struct),
4526                     &(hba[i]->cmd_pool_dhandle));
4527         hba[i]->errinfo_pool = (ErrorInfo_struct *)
4528             pci_alloc_consistent(hba[i]->pdev,
4529                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4530                     &(hba[i]->errinfo_pool_dhandle));
4531         if ((hba[i]->cmd_pool_bits == NULL)
4532             || (hba[i]->cmd_pool == NULL)
4533             || (hba[i]->errinfo_pool == NULL)) {
4534                 printk(KERN_ERR "cciss: out of memory");
4535                 goto clean4;
4536         }
4537
4538         /* Need space for temp scatter list */
4539         hba[i]->scatter_list = kmalloc(hba[i]->max_commands *
4540                                                 sizeof(struct scatterlist *),
4541                                                 GFP_KERNEL);
4542         for (k = 0; k < hba[i]->nr_cmds; k++) {
4543                 hba[i]->scatter_list[k] = kmalloc(sizeof(struct scatterlist) *
4544                                                         hba[i]->maxsgentries,
4545                                                         GFP_KERNEL);
4546                 if (hba[i]->scatter_list[k] == NULL) {
4547                         printk(KERN_ERR "cciss%d: could not allocate "
4548                                 "s/g lists\n", i);
4549                         goto clean4;
4550                 }
4551         }
4552         hba[i]->cmd_sg_list = cciss_allocate_sg_chain_blocks(hba[i],
4553                 hba[i]->chainsize, hba[i]->nr_cmds);
4554         if (!hba[i]->cmd_sg_list && hba[i]->chainsize > 0)
4555                 goto clean4;
4556
4557         spin_lock_init(&hba[i]->lock);
4558
4559         /* Initialize the pdev driver private data.
4560            have it point to hba[i].  */
4561         pci_set_drvdata(pdev, hba[i]);
4562         /* command and error info recs zeroed out before
4563            they are used */
4564         memset(hba[i]->cmd_pool_bits, 0,
4565                DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4566                         * sizeof(unsigned long));
4567
4568         hba[i]->num_luns = 0;
4569         hba[i]->highest_lun = -1;
4570         for (j = 0; j < CISS_MAX_LUN; j++) {
4571                 hba[i]->drv[j] = NULL;
4572                 hba[i]->gendisk[j] = NULL;
4573         }
4574
4575         cciss_scsi_setup(i);
4576
4577         /* Turn the interrupts on so we can service requests */
4578         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_ON);
4579
4580         /* Get the firmware version */
4581         inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
4582         if (inq_buff == NULL) {
4583                 printk(KERN_ERR "cciss: out of memory\n");
4584                 goto clean4;
4585         }
4586
4587         return_code = sendcmd_withirq(CISS_INQUIRY, i, inq_buff,
4588                 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
4589         if (return_code == IO_OK) {
4590                 hba[i]->firm_ver[0] = inq_buff->data_byte[32];
4591                 hba[i]->firm_ver[1] = inq_buff->data_byte[33];
4592                 hba[i]->firm_ver[2] = inq_buff->data_byte[34];
4593                 hba[i]->firm_ver[3] = inq_buff->data_byte[35];
4594         } else {         /* send command failed */
4595                 printk(KERN_WARNING "cciss: unable to determine firmware"
4596                         " version of controller\n");
4597         }
4598         kfree(inq_buff);
4599
4600         cciss_procinit(i);
4601
4602         hba[i]->cciss_max_sectors = 8192;
4603
4604         rebuild_lun_table(hba[i], 1, 0);
4605         hba[i]->busy_initializing = 0;
4606         return 1;
4607
4608 clean4:
4609         kfree(hba[i]->cmd_pool_bits);
4610         /* Free up sg elements */
4611         for (k = 0; k < hba[i]->nr_cmds; k++)
4612                 kfree(hba[i]->scatter_list[k]);
4613         kfree(hba[i]->scatter_list);
4614         cciss_free_sg_chain_blocks(hba[i]->cmd_sg_list, hba[i]->nr_cmds);
4615         if (hba[i]->cmd_pool)
4616                 pci_free_consistent(hba[i]->pdev,
4617                                     hba[i]->nr_cmds * sizeof(CommandList_struct),
4618                                     hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4619         if (hba[i]->errinfo_pool)
4620                 pci_free_consistent(hba[i]->pdev,
4621                                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4622                                     hba[i]->errinfo_pool,
4623                                     hba[i]->errinfo_pool_dhandle);
4624         free_irq(hba[i]->intr[PERF_MODE_INT], hba[i]);
4625 clean2:
4626         unregister_blkdev(hba[i]->major, hba[i]->devname);
4627 clean1:
4628         cciss_destroy_hba_sysfs_entry(hba[i]);
4629 clean0:
4630         pci_release_regions(pdev);
4631 clean_no_release_regions:
4632         hba[i]->busy_initializing = 0;
4633
4634         /*
4635          * Deliberately omit pci_disable_device(): it does something nasty to
4636          * Smart Array controllers that pci_enable_device does not undo
4637          */
4638         pci_set_drvdata(pdev, NULL);
4639         free_hba(i);
4640         return -1;
4641 }
4642
4643 static void cciss_shutdown(struct pci_dev *pdev)
4644 {
4645         ctlr_info_t *h;
4646         char *flush_buf;
4647         int return_code;
4648
4649         h = pci_get_drvdata(pdev);
4650         flush_buf = kzalloc(4, GFP_KERNEL);
4651         if (!flush_buf) {
4652                 printk(KERN_WARNING
4653                         "cciss:%d cache not flushed, out of memory.\n",
4654                         h->ctlr);
4655                 return;
4656         }
4657         /* write all data in the battery backed cache to disk */
4658         memset(flush_buf, 0, 4);
4659         return_code = sendcmd_withirq(CCISS_CACHE_FLUSH, h->ctlr, flush_buf,
4660                 4, 0, CTLR_LUNID, TYPE_CMD);
4661         kfree(flush_buf);
4662         if (return_code != IO_OK)
4663                 printk(KERN_WARNING "cciss%d: Error flushing cache\n",
4664                         h->ctlr);
4665         h->access.set_intr_mask(h, CCISS_INTR_OFF);
4666         free_irq(h->intr[PERF_MODE_INT], h);
4667 }
4668
4669 static void __devexit cciss_remove_one(struct pci_dev *pdev)
4670 {
4671         ctlr_info_t *tmp_ptr;
4672         int i, j;
4673
4674         if (pci_get_drvdata(pdev) == NULL) {
4675                 printk(KERN_ERR "cciss: Unable to remove device \n");
4676                 return;
4677         }
4678
4679         tmp_ptr = pci_get_drvdata(pdev);
4680         i = tmp_ptr->ctlr;
4681         if (hba[i] == NULL) {
4682                 printk(KERN_ERR "cciss: device appears to "
4683                        "already be removed \n");
4684                 return;
4685         }
4686
4687         mutex_lock(&hba[i]->busy_shutting_down);
4688
4689         remove_from_scan_list(hba[i]);
4690         remove_proc_entry(hba[i]->devname, proc_cciss);
4691         unregister_blkdev(hba[i]->major, hba[i]->devname);
4692
4693         /* remove it from the disk list */
4694         for (j = 0; j < CISS_MAX_LUN; j++) {
4695                 struct gendisk *disk = hba[i]->gendisk[j];
4696                 if (disk) {
4697                         struct request_queue *q = disk->queue;
4698
4699                         if (disk->flags & GENHD_FL_UP) {
4700                                 cciss_destroy_ld_sysfs_entry(hba[i], j, 1);
4701                                 del_gendisk(disk);
4702                         }
4703                         if (q)
4704                                 blk_cleanup_queue(q);
4705                 }
4706         }
4707
4708 #ifdef CONFIG_CISS_SCSI_TAPE
4709         cciss_unregister_scsi(i);       /* unhook from SCSI subsystem */
4710 #endif
4711
4712         cciss_shutdown(pdev);
4713
4714 #ifdef CONFIG_PCI_MSI
4715         if (hba[i]->msix_vector)
4716                 pci_disable_msix(hba[i]->pdev);
4717         else if (hba[i]->msi_vector)
4718                 pci_disable_msi(hba[i]->pdev);
4719 #endif                          /* CONFIG_PCI_MSI */
4720
4721         iounmap(hba[i]->vaddr);
4722
4723         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(CommandList_struct),
4724                             hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4725         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4726                             hba[i]->errinfo_pool, hba[i]->errinfo_pool_dhandle);
4727         kfree(hba[i]->cmd_pool_bits);
4728         /* Free up sg elements */
4729         for (j = 0; j < hba[i]->nr_cmds; j++)
4730                 kfree(hba[i]->scatter_list[j]);
4731         kfree(hba[i]->scatter_list);
4732         cciss_free_sg_chain_blocks(hba[i]->cmd_sg_list, hba[i]->nr_cmds);
4733         /*
4734          * Deliberately omit pci_disable_device(): it does something nasty to
4735          * Smart Array controllers that pci_enable_device does not undo
4736          */
4737         pci_release_regions(pdev);
4738         pci_set_drvdata(pdev, NULL);
4739         cciss_destroy_hba_sysfs_entry(hba[i]);
4740         mutex_unlock(&hba[i]->busy_shutting_down);
4741         free_hba(i);
4742 }
4743
4744 static struct pci_driver cciss_pci_driver = {
4745         .name = "cciss",
4746         .probe = cciss_init_one,
4747         .remove = __devexit_p(cciss_remove_one),
4748         .id_table = cciss_pci_device_id,        /* id_table */
4749         .shutdown = cciss_shutdown,
4750 };
4751
4752 /*
4753  *  This is it.  Register the PCI driver information for the cards we control
4754  *  the OS will call our registered routines when it finds one of our cards.
4755  */
4756 static int __init cciss_init(void)
4757 {
4758         int err;
4759
4760         /*
4761          * The hardware requires that commands are aligned on a 64-bit
4762          * boundary. Given that we use pci_alloc_consistent() to allocate an
4763          * array of them, the size must be a multiple of 8 bytes.
4764          */
4765         BUILD_BUG_ON(sizeof(CommandList_struct) % COMMANDLIST_ALIGNMENT);
4766         printk(KERN_INFO DRIVER_NAME "\n");
4767
4768         err = bus_register(&cciss_bus_type);
4769         if (err)
4770                 return err;
4771
4772         /* Start the scan thread */
4773         cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
4774         if (IS_ERR(cciss_scan_thread)) {
4775                 err = PTR_ERR(cciss_scan_thread);
4776                 goto err_bus_unregister;
4777         }
4778
4779         /* Register for our PCI devices */
4780         err = pci_register_driver(&cciss_pci_driver);
4781         if (err)
4782                 goto err_thread_stop;
4783
4784         return err;
4785
4786 err_thread_stop:
4787         kthread_stop(cciss_scan_thread);
4788 err_bus_unregister:
4789         bus_unregister(&cciss_bus_type);
4790
4791         return err;
4792 }
4793
4794 static void __exit cciss_cleanup(void)
4795 {
4796         int i;
4797
4798         pci_unregister_driver(&cciss_pci_driver);
4799         /* double check that all controller entrys have been removed */
4800         for (i = 0; i < MAX_CTLR; i++) {
4801                 if (hba[i] != NULL) {
4802                         printk(KERN_WARNING "cciss: had to remove"
4803                                " controller %d\n", i);
4804                         cciss_remove_one(hba[i]->pdev);
4805                 }
4806         }
4807         kthread_stop(cciss_scan_thread);
4808         remove_proc_entry("driver/cciss", NULL);
4809         bus_unregister(&cciss_bus_type);
4810 }
4811
4812 module_init(cciss_init);
4813 module_exit(cciss_cleanup);