2cdbc247d0ac92ca2765ea906174dcbc7bacb890
[~shefty/rdma-dev.git] / drivers / block / cciss.c
1 /*
2  *    Disk Array driver for HP Smart Array controllers.
3  *    (C) Copyright 2000, 2007 Hewlett-Packard Development Company, L.P.
4  *
5  *    This program is free software; you can redistribute it and/or modify
6  *    it under the terms of the GNU General Public License as published by
7  *    the Free Software Foundation; version 2 of the License.
8  *
9  *    This program is distributed in the hope that it will be useful,
10  *    but WITHOUT ANY WARRANTY; without even the implied warranty of
11  *    MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12  *    General Public License for more details.
13  *
14  *    You should have received a copy of the GNU General Public License
15  *    along with this program; if not, write to the Free Software
16  *    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
17  *    02111-1307, USA.
18  *
19  *    Questions/Comments/Bugfixes to iss_storagedev@hp.com
20  *
21  */
22
23 #include <linux/module.h>
24 #include <linux/interrupt.h>
25 #include <linux/types.h>
26 #include <linux/pci.h>
27 #include <linux/kernel.h>
28 #include <linux/slab.h>
29 #include <linux/delay.h>
30 #include <linux/major.h>
31 #include <linux/fs.h>
32 #include <linux/bio.h>
33 #include <linux/blkpg.h>
34 #include <linux/timer.h>
35 #include <linux/proc_fs.h>
36 #include <linux/seq_file.h>
37 #include <linux/init.h>
38 #include <linux/jiffies.h>
39 #include <linux/hdreg.h>
40 #include <linux/spinlock.h>
41 #include <linux/compat.h>
42 #include <linux/mutex.h>
43 #include <asm/uaccess.h>
44 #include <asm/io.h>
45
46 #include <linux/dma-mapping.h>
47 #include <linux/blkdev.h>
48 #include <linux/genhd.h>
49 #include <linux/completion.h>
50 #include <scsi/scsi.h>
51 #include <scsi/sg.h>
52 #include <scsi/scsi_ioctl.h>
53 #include <linux/cdrom.h>
54 #include <linux/scatterlist.h>
55 #include <linux/kthread.h>
56
57 #define CCISS_DRIVER_VERSION(maj,min,submin) ((maj<<16)|(min<<8)|(submin))
58 #define DRIVER_NAME "HP CISS Driver (v 3.6.26)"
59 #define DRIVER_VERSION CCISS_DRIVER_VERSION(3, 6, 26)
60
61 /* Embedded module documentation macros - see modules.h */
62 MODULE_AUTHOR("Hewlett-Packard Company");
63 MODULE_DESCRIPTION("Driver for HP Smart Array Controllers");
64 MODULE_SUPPORTED_DEVICE("HP Smart Array Controllers");
65 MODULE_VERSION("3.6.26");
66 MODULE_LICENSE("GPL");
67
68 static DEFINE_MUTEX(cciss_mutex);
69
70 #include "cciss_cmd.h"
71 #include "cciss.h"
72 #include <linux/cciss_ioctl.h>
73
74 /* define the PCI info for the cards we can control */
75 static const struct pci_device_id cciss_pci_device_id[] = {
76         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISS,  0x0E11, 0x4070},
77         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4080},
78         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4082},
79         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4083},
80         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x4091},
81         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409A},
82         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409B},
83         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409C},
84         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409D},
85         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSA,     0x103C, 0x3225},
86         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3223},
87         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3234},
88         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3235},
89         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3211},
90         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3212},
91         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3213},
92         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3214},
93         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3215},
94         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3237},
95         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x323D},
96         {0,}
97 };
98
99 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
100
101 /*  board_id = Subsystem Device ID & Vendor ID
102  *  product = Marketing Name for the board
103  *  access = Address of the struct of function pointers
104  */
105 static struct board_type products[] = {
106         {0x40700E11, "Smart Array 5300", &SA5_access},
107         {0x40800E11, "Smart Array 5i", &SA5B_access},
108         {0x40820E11, "Smart Array 532", &SA5B_access},
109         {0x40830E11, "Smart Array 5312", &SA5B_access},
110         {0x409A0E11, "Smart Array 641", &SA5_access},
111         {0x409B0E11, "Smart Array 642", &SA5_access},
112         {0x409C0E11, "Smart Array 6400", &SA5_access},
113         {0x409D0E11, "Smart Array 6400 EM", &SA5_access},
114         {0x40910E11, "Smart Array 6i", &SA5_access},
115         {0x3225103C, "Smart Array P600", &SA5_access},
116         {0x3235103C, "Smart Array P400i", &SA5_access},
117         {0x3211103C, "Smart Array E200i", &SA5_access},
118         {0x3212103C, "Smart Array E200", &SA5_access},
119         {0x3213103C, "Smart Array E200i", &SA5_access},
120         {0x3214103C, "Smart Array E200i", &SA5_access},
121         {0x3215103C, "Smart Array E200i", &SA5_access},
122         {0x3237103C, "Smart Array E500", &SA5_access},
123         {0x3223103C, "Smart Array P800", &SA5_access},
124         {0x3234103C, "Smart Array P400", &SA5_access},
125         {0x323D103C, "Smart Array P700m", &SA5_access},
126 };
127
128 /* How long to wait (in milliseconds) for board to go into simple mode */
129 #define MAX_CONFIG_WAIT 30000
130 #define MAX_IOCTL_CONFIG_WAIT 1000
131
132 /*define how many times we will try a command because of bus resets */
133 #define MAX_CMD_RETRIES 3
134
135 #define MAX_CTLR        32
136
137 /* Originally cciss driver only supports 8 major numbers */
138 #define MAX_CTLR_ORIG   8
139
140 static ctlr_info_t *hba[MAX_CTLR];
141
142 static struct task_struct *cciss_scan_thread;
143 static DEFINE_MUTEX(scan_mutex);
144 static LIST_HEAD(scan_q);
145
146 static void do_cciss_request(struct request_queue *q);
147 static irqreturn_t do_cciss_intx(int irq, void *dev_id);
148 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id);
149 static int cciss_open(struct block_device *bdev, fmode_t mode);
150 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode);
151 static int cciss_release(struct gendisk *disk, fmode_t mode);
152 static int do_ioctl(struct block_device *bdev, fmode_t mode,
153                     unsigned int cmd, unsigned long arg);
154 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
155                        unsigned int cmd, unsigned long arg);
156 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
157
158 static int cciss_revalidate(struct gendisk *disk);
159 static int rebuild_lun_table(ctlr_info_t *h, int first_time, int via_ioctl);
160 static int deregister_disk(ctlr_info_t *h, int drv_index,
161                            int clear_all, int via_ioctl);
162
163 static void cciss_read_capacity(ctlr_info_t *h, int logvol,
164                         sector_t *total_size, unsigned int *block_size);
165 static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
166                         sector_t *total_size, unsigned int *block_size);
167 static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
168                         sector_t total_size,
169                         unsigned int block_size, InquiryData_struct *inq_buff,
170                                    drive_info_struct *drv);
171 static void __devinit cciss_interrupt_mode(ctlr_info_t *);
172 static void start_io(ctlr_info_t *h);
173 static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
174                         __u8 page_code, unsigned char scsi3addr[],
175                         int cmd_type);
176 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
177         int attempt_retry);
178 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c);
179
180 static int add_to_scan_list(struct ctlr_info *h);
181 static int scan_thread(void *data);
182 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c);
183 static void cciss_hba_release(struct device *dev);
184 static void cciss_device_release(struct device *dev);
185 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index);
186 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index);
187 static inline u32 next_command(ctlr_info_t *h);
188 static int __devinit cciss_find_cfg_addrs(struct pci_dev *pdev,
189         void __iomem *vaddr, u32 *cfg_base_addr, u64 *cfg_base_addr_index,
190         u64 *cfg_offset);
191 static int __devinit cciss_pci_find_memory_BAR(struct pci_dev *pdev,
192         unsigned long *memory_bar);
193
194
195 /* performant mode helper functions */
196 static void  calc_bucket_map(int *bucket, int num_buckets, int nsgs,
197                                 int *bucket_map);
198 static void cciss_put_controller_into_performant_mode(ctlr_info_t *h);
199
200 #ifdef CONFIG_PROC_FS
201 static void cciss_procinit(ctlr_info_t *h);
202 #else
203 static void cciss_procinit(ctlr_info_t *h)
204 {
205 }
206 #endif                          /* CONFIG_PROC_FS */
207
208 #ifdef CONFIG_COMPAT
209 static int cciss_compat_ioctl(struct block_device *, fmode_t,
210                               unsigned, unsigned long);
211 #endif
212
213 static const struct block_device_operations cciss_fops = {
214         .owner = THIS_MODULE,
215         .open = cciss_unlocked_open,
216         .release = cciss_release,
217         .ioctl = do_ioctl,
218         .getgeo = cciss_getgeo,
219 #ifdef CONFIG_COMPAT
220         .compat_ioctl = cciss_compat_ioctl,
221 #endif
222         .revalidate_disk = cciss_revalidate,
223 };
224
225 /* set_performant_mode: Modify the tag for cciss performant
226  * set bit 0 for pull model, bits 3-1 for block fetch
227  * register number
228  */
229 static void set_performant_mode(ctlr_info_t *h, CommandList_struct *c)
230 {
231         if (likely(h->transMethod == CFGTBL_Trans_Performant))
232                 c->busaddr |= 1 | (h->blockFetchTable[c->Header.SGList] << 1);
233 }
234
235 /*
236  * Enqueuing and dequeuing functions for cmdlists.
237  */
238 static inline void addQ(struct hlist_head *list, CommandList_struct *c)
239 {
240         hlist_add_head(&c->list, list);
241 }
242
243 static inline void removeQ(CommandList_struct *c)
244 {
245         /*
246          * After kexec/dump some commands might still
247          * be in flight, which the firmware will try
248          * to complete. Resetting the firmware doesn't work
249          * with old fw revisions, so we have to mark
250          * them off as 'stale' to prevent the driver from
251          * falling over.
252          */
253         if (WARN_ON(hlist_unhashed(&c->list))) {
254                 c->cmd_type = CMD_MSG_STALE;
255                 return;
256         }
257
258         hlist_del_init(&c->list);
259 }
260
261 static void enqueue_cmd_and_start_io(ctlr_info_t *h,
262         CommandList_struct *c)
263 {
264         unsigned long flags;
265         set_performant_mode(h, c);
266         spin_lock_irqsave(&h->lock, flags);
267         addQ(&h->reqQ, c);
268         h->Qdepth++;
269         if (h->Qdepth > h->maxQsinceinit)
270                 h->maxQsinceinit = h->Qdepth;
271         start_io(h);
272         spin_unlock_irqrestore(&h->lock, flags);
273 }
274
275 static void cciss_free_sg_chain_blocks(SGDescriptor_struct **cmd_sg_list,
276         int nr_cmds)
277 {
278         int i;
279
280         if (!cmd_sg_list)
281                 return;
282         for (i = 0; i < nr_cmds; i++) {
283                 kfree(cmd_sg_list[i]);
284                 cmd_sg_list[i] = NULL;
285         }
286         kfree(cmd_sg_list);
287 }
288
289 static SGDescriptor_struct **cciss_allocate_sg_chain_blocks(
290         ctlr_info_t *h, int chainsize, int nr_cmds)
291 {
292         int j;
293         SGDescriptor_struct **cmd_sg_list;
294
295         if (chainsize <= 0)
296                 return NULL;
297
298         cmd_sg_list = kmalloc(sizeof(*cmd_sg_list) * nr_cmds, GFP_KERNEL);
299         if (!cmd_sg_list)
300                 return NULL;
301
302         /* Build up chain blocks for each command */
303         for (j = 0; j < nr_cmds; j++) {
304                 /* Need a block of chainsized s/g elements. */
305                 cmd_sg_list[j] = kmalloc((chainsize *
306                         sizeof(*cmd_sg_list[j])), GFP_KERNEL);
307                 if (!cmd_sg_list[j]) {
308                         dev_err(&h->pdev->dev, "Cannot get memory "
309                                 "for s/g chains.\n");
310                         goto clean;
311                 }
312         }
313         return cmd_sg_list;
314 clean:
315         cciss_free_sg_chain_blocks(cmd_sg_list, nr_cmds);
316         return NULL;
317 }
318
319 static void cciss_unmap_sg_chain_block(ctlr_info_t *h, CommandList_struct *c)
320 {
321         SGDescriptor_struct *chain_sg;
322         u64bit temp64;
323
324         if (c->Header.SGTotal <= h->max_cmd_sgentries)
325                 return;
326
327         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
328         temp64.val32.lower = chain_sg->Addr.lower;
329         temp64.val32.upper = chain_sg->Addr.upper;
330         pci_unmap_single(h->pdev, temp64.val, chain_sg->Len, PCI_DMA_TODEVICE);
331 }
332
333 static void cciss_map_sg_chain_block(ctlr_info_t *h, CommandList_struct *c,
334         SGDescriptor_struct *chain_block, int len)
335 {
336         SGDescriptor_struct *chain_sg;
337         u64bit temp64;
338
339         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
340         chain_sg->Ext = CCISS_SG_CHAIN;
341         chain_sg->Len = len;
342         temp64.val = pci_map_single(h->pdev, chain_block, len,
343                                 PCI_DMA_TODEVICE);
344         chain_sg->Addr.lower = temp64.val32.lower;
345         chain_sg->Addr.upper = temp64.val32.upper;
346 }
347
348 #include "cciss_scsi.c"         /* For SCSI tape support */
349
350 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
351         "UNKNOWN"
352 };
353 #define RAID_UNKNOWN (ARRAY_SIZE(raid_label)-1)
354
355 #ifdef CONFIG_PROC_FS
356
357 /*
358  * Report information about this controller.
359  */
360 #define ENG_GIG 1000000000
361 #define ENG_GIG_FACTOR (ENG_GIG/512)
362 #define ENGAGE_SCSI     "engage scsi"
363
364 static struct proc_dir_entry *proc_cciss;
365
366 static void cciss_seq_show_header(struct seq_file *seq)
367 {
368         ctlr_info_t *h = seq->private;
369
370         seq_printf(seq, "%s: HP %s Controller\n"
371                 "Board ID: 0x%08lx\n"
372                 "Firmware Version: %c%c%c%c\n"
373                 "IRQ: %d\n"
374                 "Logical drives: %d\n"
375                 "Current Q depth: %d\n"
376                 "Current # commands on controller: %d\n"
377                 "Max Q depth since init: %d\n"
378                 "Max # commands on controller since init: %d\n"
379                 "Max SG entries since init: %d\n",
380                 h->devname,
381                 h->product_name,
382                 (unsigned long)h->board_id,
383                 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
384                 h->firm_ver[3], (unsigned int)h->intr[PERF_MODE_INT],
385                 h->num_luns,
386                 h->Qdepth, h->commands_outstanding,
387                 h->maxQsinceinit, h->max_outstanding, h->maxSG);
388
389 #ifdef CONFIG_CISS_SCSI_TAPE
390         cciss_seq_tape_report(seq, h);
391 #endif /* CONFIG_CISS_SCSI_TAPE */
392 }
393
394 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
395 {
396         ctlr_info_t *h = seq->private;
397         unsigned long flags;
398
399         /* prevent displaying bogus info during configuration
400          * or deconfiguration of a logical volume
401          */
402         spin_lock_irqsave(&h->lock, flags);
403         if (h->busy_configuring) {
404                 spin_unlock_irqrestore(&h->lock, flags);
405                 return ERR_PTR(-EBUSY);
406         }
407         h->busy_configuring = 1;
408         spin_unlock_irqrestore(&h->lock, flags);
409
410         if (*pos == 0)
411                 cciss_seq_show_header(seq);
412
413         return pos;
414 }
415
416 static int cciss_seq_show(struct seq_file *seq, void *v)
417 {
418         sector_t vol_sz, vol_sz_frac;
419         ctlr_info_t *h = seq->private;
420         unsigned ctlr = h->ctlr;
421         loff_t *pos = v;
422         drive_info_struct *drv = h->drv[*pos];
423
424         if (*pos > h->highest_lun)
425                 return 0;
426
427         if (drv == NULL) /* it's possible for h->drv[] to have holes. */
428                 return 0;
429
430         if (drv->heads == 0)
431                 return 0;
432
433         vol_sz = drv->nr_blocks;
434         vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
435         vol_sz_frac *= 100;
436         sector_div(vol_sz_frac, ENG_GIG_FACTOR);
437
438         if (drv->raid_level < 0 || drv->raid_level > RAID_UNKNOWN)
439                 drv->raid_level = RAID_UNKNOWN;
440         seq_printf(seq, "cciss/c%dd%d:"
441                         "\t%4u.%02uGB\tRAID %s\n",
442                         ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
443                         raid_label[drv->raid_level]);
444         return 0;
445 }
446
447 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
448 {
449         ctlr_info_t *h = seq->private;
450
451         if (*pos > h->highest_lun)
452                 return NULL;
453         *pos += 1;
454
455         return pos;
456 }
457
458 static void cciss_seq_stop(struct seq_file *seq, void *v)
459 {
460         ctlr_info_t *h = seq->private;
461
462         /* Only reset h->busy_configuring if we succeeded in setting
463          * it during cciss_seq_start. */
464         if (v == ERR_PTR(-EBUSY))
465                 return;
466
467         h->busy_configuring = 0;
468 }
469
470 static const struct seq_operations cciss_seq_ops = {
471         .start = cciss_seq_start,
472         .show  = cciss_seq_show,
473         .next  = cciss_seq_next,
474         .stop  = cciss_seq_stop,
475 };
476
477 static int cciss_seq_open(struct inode *inode, struct file *file)
478 {
479         int ret = seq_open(file, &cciss_seq_ops);
480         struct seq_file *seq = file->private_data;
481
482         if (!ret)
483                 seq->private = PDE(inode)->data;
484
485         return ret;
486 }
487
488 static ssize_t
489 cciss_proc_write(struct file *file, const char __user *buf,
490                  size_t length, loff_t *ppos)
491 {
492         int err;
493         char *buffer;
494
495 #ifndef CONFIG_CISS_SCSI_TAPE
496         return -EINVAL;
497 #endif
498
499         if (!buf || length > PAGE_SIZE - 1)
500                 return -EINVAL;
501
502         buffer = (char *)__get_free_page(GFP_KERNEL);
503         if (!buffer)
504                 return -ENOMEM;
505
506         err = -EFAULT;
507         if (copy_from_user(buffer, buf, length))
508                 goto out;
509         buffer[length] = '\0';
510
511 #ifdef CONFIG_CISS_SCSI_TAPE
512         if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
513                 struct seq_file *seq = file->private_data;
514                 ctlr_info_t *h = seq->private;
515
516                 err = cciss_engage_scsi(h);
517                 if (err == 0)
518                         err = length;
519         } else
520 #endif /* CONFIG_CISS_SCSI_TAPE */
521                 err = -EINVAL;
522         /* might be nice to have "disengage" too, but it's not
523            safely possible. (only 1 module use count, lock issues.) */
524
525 out:
526         free_page((unsigned long)buffer);
527         return err;
528 }
529
530 static const struct file_operations cciss_proc_fops = {
531         .owner   = THIS_MODULE,
532         .open    = cciss_seq_open,
533         .read    = seq_read,
534         .llseek  = seq_lseek,
535         .release = seq_release,
536         .write   = cciss_proc_write,
537 };
538
539 static void __devinit cciss_procinit(ctlr_info_t *h)
540 {
541         struct proc_dir_entry *pde;
542
543         if (proc_cciss == NULL)
544                 proc_cciss = proc_mkdir("driver/cciss", NULL);
545         if (!proc_cciss)
546                 return;
547         pde = proc_create_data(h->devname, S_IWUSR | S_IRUSR | S_IRGRP |
548                                         S_IROTH, proc_cciss,
549                                         &cciss_proc_fops, h);
550 }
551 #endif                          /* CONFIG_PROC_FS */
552
553 #define MAX_PRODUCT_NAME_LEN 19
554
555 #define to_hba(n) container_of(n, struct ctlr_info, dev)
556 #define to_drv(n) container_of(n, drive_info_struct, dev)
557
558 static ssize_t host_store_rescan(struct device *dev,
559                                  struct device_attribute *attr,
560                                  const char *buf, size_t count)
561 {
562         struct ctlr_info *h = to_hba(dev);
563
564         add_to_scan_list(h);
565         wake_up_process(cciss_scan_thread);
566         wait_for_completion_interruptible(&h->scan_wait);
567
568         return count;
569 }
570 static DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
571
572 static ssize_t dev_show_unique_id(struct device *dev,
573                                  struct device_attribute *attr,
574                                  char *buf)
575 {
576         drive_info_struct *drv = to_drv(dev);
577         struct ctlr_info *h = to_hba(drv->dev.parent);
578         __u8 sn[16];
579         unsigned long flags;
580         int ret = 0;
581
582         spin_lock_irqsave(&h->lock, flags);
583         if (h->busy_configuring)
584                 ret = -EBUSY;
585         else
586                 memcpy(sn, drv->serial_no, sizeof(sn));
587         spin_unlock_irqrestore(&h->lock, flags);
588
589         if (ret)
590                 return ret;
591         else
592                 return snprintf(buf, 16 * 2 + 2,
593                                 "%02X%02X%02X%02X%02X%02X%02X%02X"
594                                 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
595                                 sn[0], sn[1], sn[2], sn[3],
596                                 sn[4], sn[5], sn[6], sn[7],
597                                 sn[8], sn[9], sn[10], sn[11],
598                                 sn[12], sn[13], sn[14], sn[15]);
599 }
600 static DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
601
602 static ssize_t dev_show_vendor(struct device *dev,
603                                struct device_attribute *attr,
604                                char *buf)
605 {
606         drive_info_struct *drv = to_drv(dev);
607         struct ctlr_info *h = to_hba(drv->dev.parent);
608         char vendor[VENDOR_LEN + 1];
609         unsigned long flags;
610         int ret = 0;
611
612         spin_lock_irqsave(&h->lock, flags);
613         if (h->busy_configuring)
614                 ret = -EBUSY;
615         else
616                 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
617         spin_unlock_irqrestore(&h->lock, flags);
618
619         if (ret)
620                 return ret;
621         else
622                 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
623 }
624 static DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
625
626 static ssize_t dev_show_model(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 model[MODEL_LEN + 1];
633         unsigned long flags;
634         int ret = 0;
635
636         spin_lock_irqsave(&h->lock, flags);
637         if (h->busy_configuring)
638                 ret = -EBUSY;
639         else
640                 memcpy(model, drv->model, MODEL_LEN + 1);
641         spin_unlock_irqrestore(&h->lock, flags);
642
643         if (ret)
644                 return ret;
645         else
646                 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
647 }
648 static DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
649
650 static ssize_t dev_show_rev(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 rev[REV_LEN + 1];
657         unsigned long flags;
658         int ret = 0;
659
660         spin_lock_irqsave(&h->lock, flags);
661         if (h->busy_configuring)
662                 ret = -EBUSY;
663         else
664                 memcpy(rev, drv->rev, REV_LEN + 1);
665         spin_unlock_irqrestore(&h->lock, flags);
666
667         if (ret)
668                 return ret;
669         else
670                 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
671 }
672 static DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
673
674 static ssize_t cciss_show_lunid(struct device *dev,
675                                 struct device_attribute *attr, char *buf)
676 {
677         drive_info_struct *drv = to_drv(dev);
678         struct ctlr_info *h = to_hba(drv->dev.parent);
679         unsigned long flags;
680         unsigned char lunid[8];
681
682         spin_lock_irqsave(&h->lock, flags);
683         if (h->busy_configuring) {
684                 spin_unlock_irqrestore(&h->lock, flags);
685                 return -EBUSY;
686         }
687         if (!drv->heads) {
688                 spin_unlock_irqrestore(&h->lock, flags);
689                 return -ENOTTY;
690         }
691         memcpy(lunid, drv->LunID, sizeof(lunid));
692         spin_unlock_irqrestore(&h->lock, flags);
693         return snprintf(buf, 20, "0x%02x%02x%02x%02x%02x%02x%02x%02x\n",
694                 lunid[0], lunid[1], lunid[2], lunid[3],
695                 lunid[4], lunid[5], lunid[6], lunid[7]);
696 }
697 static DEVICE_ATTR(lunid, S_IRUGO, cciss_show_lunid, NULL);
698
699 static ssize_t cciss_show_raid_level(struct device *dev,
700                                      struct device_attribute *attr, char *buf)
701 {
702         drive_info_struct *drv = to_drv(dev);
703         struct ctlr_info *h = to_hba(drv->dev.parent);
704         int raid;
705         unsigned long flags;
706
707         spin_lock_irqsave(&h->lock, flags);
708         if (h->busy_configuring) {
709                 spin_unlock_irqrestore(&h->lock, flags);
710                 return -EBUSY;
711         }
712         raid = drv->raid_level;
713         spin_unlock_irqrestore(&h->lock, flags);
714         if (raid < 0 || raid > RAID_UNKNOWN)
715                 raid = RAID_UNKNOWN;
716
717         return snprintf(buf, strlen(raid_label[raid]) + 7, "RAID %s\n",
718                         raid_label[raid]);
719 }
720 static DEVICE_ATTR(raid_level, S_IRUGO, cciss_show_raid_level, NULL);
721
722 static ssize_t cciss_show_usage_count(struct device *dev,
723                                       struct device_attribute *attr, char *buf)
724 {
725         drive_info_struct *drv = to_drv(dev);
726         struct ctlr_info *h = to_hba(drv->dev.parent);
727         unsigned long flags;
728         int count;
729
730         spin_lock_irqsave(&h->lock, flags);
731         if (h->busy_configuring) {
732                 spin_unlock_irqrestore(&h->lock, flags);
733                 return -EBUSY;
734         }
735         count = drv->usage_count;
736         spin_unlock_irqrestore(&h->lock, flags);
737         return snprintf(buf, 20, "%d\n", count);
738 }
739 static DEVICE_ATTR(usage_count, S_IRUGO, cciss_show_usage_count, NULL);
740
741 static struct attribute *cciss_host_attrs[] = {
742         &dev_attr_rescan.attr,
743         NULL
744 };
745
746 static struct attribute_group cciss_host_attr_group = {
747         .attrs = cciss_host_attrs,
748 };
749
750 static const struct attribute_group *cciss_host_attr_groups[] = {
751         &cciss_host_attr_group,
752         NULL
753 };
754
755 static struct device_type cciss_host_type = {
756         .name           = "cciss_host",
757         .groups         = cciss_host_attr_groups,
758         .release        = cciss_hba_release,
759 };
760
761 static struct attribute *cciss_dev_attrs[] = {
762         &dev_attr_unique_id.attr,
763         &dev_attr_model.attr,
764         &dev_attr_vendor.attr,
765         &dev_attr_rev.attr,
766         &dev_attr_lunid.attr,
767         &dev_attr_raid_level.attr,
768         &dev_attr_usage_count.attr,
769         NULL
770 };
771
772 static struct attribute_group cciss_dev_attr_group = {
773         .attrs = cciss_dev_attrs,
774 };
775
776 static const struct attribute_group *cciss_dev_attr_groups[] = {
777         &cciss_dev_attr_group,
778         NULL
779 };
780
781 static struct device_type cciss_dev_type = {
782         .name           = "cciss_device",
783         .groups         = cciss_dev_attr_groups,
784         .release        = cciss_device_release,
785 };
786
787 static struct bus_type cciss_bus_type = {
788         .name           = "cciss",
789 };
790
791 /*
792  * cciss_hba_release is called when the reference count
793  * of h->dev goes to zero.
794  */
795 static void cciss_hba_release(struct device *dev)
796 {
797         /*
798          * nothing to do, but need this to avoid a warning
799          * about not having a release handler from lib/kref.c.
800          */
801 }
802
803 /*
804  * Initialize sysfs entry for each controller.  This sets up and registers
805  * the 'cciss#' directory for each individual controller under
806  * /sys/bus/pci/devices/<dev>/.
807  */
808 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
809 {
810         device_initialize(&h->dev);
811         h->dev.type = &cciss_host_type;
812         h->dev.bus = &cciss_bus_type;
813         dev_set_name(&h->dev, "%s", h->devname);
814         h->dev.parent = &h->pdev->dev;
815
816         return device_add(&h->dev);
817 }
818
819 /*
820  * Remove sysfs entries for an hba.
821  */
822 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
823 {
824         device_del(&h->dev);
825         put_device(&h->dev); /* final put. */
826 }
827
828 /* cciss_device_release is called when the reference count
829  * of h->drv[x]dev goes to zero.
830  */
831 static void cciss_device_release(struct device *dev)
832 {
833         drive_info_struct *drv = to_drv(dev);
834         kfree(drv);
835 }
836
837 /*
838  * Initialize sysfs for each logical drive.  This sets up and registers
839  * the 'c#d#' directory for each individual logical drive under
840  * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
841  * /sys/block/cciss!c#d# to this entry.
842  */
843 static long cciss_create_ld_sysfs_entry(struct ctlr_info *h,
844                                        int drv_index)
845 {
846         struct device *dev;
847
848         if (h->drv[drv_index]->device_initialized)
849                 return 0;
850
851         dev = &h->drv[drv_index]->dev;
852         device_initialize(dev);
853         dev->type = &cciss_dev_type;
854         dev->bus = &cciss_bus_type;
855         dev_set_name(dev, "c%dd%d", h->ctlr, drv_index);
856         dev->parent = &h->dev;
857         h->drv[drv_index]->device_initialized = 1;
858         return device_add(dev);
859 }
860
861 /*
862  * Remove sysfs entries for a logical drive.
863  */
864 static void cciss_destroy_ld_sysfs_entry(struct ctlr_info *h, int drv_index,
865         int ctlr_exiting)
866 {
867         struct device *dev = &h->drv[drv_index]->dev;
868
869         /* special case for c*d0, we only destroy it on controller exit */
870         if (drv_index == 0 && !ctlr_exiting)
871                 return;
872
873         device_del(dev);
874         put_device(dev); /* the "final" put. */
875         h->drv[drv_index] = NULL;
876 }
877
878 /*
879  * For operations that cannot sleep, a command block is allocated at init,
880  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
881  * which ones are free or in use.
882  */
883 static CommandList_struct *cmd_alloc(ctlr_info_t *h)
884 {
885         CommandList_struct *c;
886         int i;
887         u64bit temp64;
888         dma_addr_t cmd_dma_handle, err_dma_handle;
889
890         do {
891                 i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
892                 if (i == h->nr_cmds)
893                         return NULL;
894         } while (test_and_set_bit(i & (BITS_PER_LONG - 1),
895                   h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
896         c = h->cmd_pool + i;
897         memset(c, 0, sizeof(CommandList_struct));
898         cmd_dma_handle = h->cmd_pool_dhandle + i * sizeof(CommandList_struct);
899         c->err_info = h->errinfo_pool + i;
900         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
901         err_dma_handle = h->errinfo_pool_dhandle
902             + i * sizeof(ErrorInfo_struct);
903         h->nr_allocs++;
904
905         c->cmdindex = i;
906
907         INIT_HLIST_NODE(&c->list);
908         c->busaddr = (__u32) cmd_dma_handle;
909         temp64.val = (__u64) err_dma_handle;
910         c->ErrDesc.Addr.lower = temp64.val32.lower;
911         c->ErrDesc.Addr.upper = temp64.val32.upper;
912         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
913
914         c->ctlr = h->ctlr;
915         return c;
916 }
917
918 /* allocate a command using pci_alloc_consistent, used for ioctls,
919  * etc., not for the main i/o path.
920  */
921 static CommandList_struct *cmd_special_alloc(ctlr_info_t *h)
922 {
923         CommandList_struct *c;
924         u64bit temp64;
925         dma_addr_t cmd_dma_handle, err_dma_handle;
926
927         c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
928                 sizeof(CommandList_struct), &cmd_dma_handle);
929         if (c == NULL)
930                 return NULL;
931         memset(c, 0, sizeof(CommandList_struct));
932
933         c->cmdindex = -1;
934
935         c->err_info = (ErrorInfo_struct *)
936             pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
937                     &err_dma_handle);
938
939         if (c->err_info == NULL) {
940                 pci_free_consistent(h->pdev,
941                         sizeof(CommandList_struct), c, cmd_dma_handle);
942                 return NULL;
943         }
944         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
945
946         INIT_HLIST_NODE(&c->list);
947         c->busaddr = (__u32) cmd_dma_handle;
948         temp64.val = (__u64) err_dma_handle;
949         c->ErrDesc.Addr.lower = temp64.val32.lower;
950         c->ErrDesc.Addr.upper = temp64.val32.upper;
951         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
952
953         c->ctlr = h->ctlr;
954         return c;
955 }
956
957 static void cmd_free(ctlr_info_t *h, CommandList_struct *c)
958 {
959         int i;
960
961         i = c - h->cmd_pool;
962         clear_bit(i & (BITS_PER_LONG - 1),
963                   h->cmd_pool_bits + (i / BITS_PER_LONG));
964         h->nr_frees++;
965 }
966
967 static void cmd_special_free(ctlr_info_t *h, CommandList_struct *c)
968 {
969         u64bit temp64;
970
971         temp64.val32.lower = c->ErrDesc.Addr.lower;
972         temp64.val32.upper = c->ErrDesc.Addr.upper;
973         pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
974                             c->err_info, (dma_addr_t) temp64.val);
975         pci_free_consistent(h->pdev, sizeof(CommandList_struct),
976                             c, (dma_addr_t) c->busaddr);
977 }
978
979 static inline ctlr_info_t *get_host(struct gendisk *disk)
980 {
981         return disk->queue->queuedata;
982 }
983
984 static inline drive_info_struct *get_drv(struct gendisk *disk)
985 {
986         return disk->private_data;
987 }
988
989 /*
990  * Open.  Make sure the device is really there.
991  */
992 static int cciss_open(struct block_device *bdev, fmode_t mode)
993 {
994         ctlr_info_t *h = get_host(bdev->bd_disk);
995         drive_info_struct *drv = get_drv(bdev->bd_disk);
996
997         dev_dbg(&h->pdev->dev, "cciss_open %s\n", bdev->bd_disk->disk_name);
998         if (drv->busy_configuring)
999                 return -EBUSY;
1000         /*
1001          * Root is allowed to open raw volume zero even if it's not configured
1002          * so array config can still work. Root is also allowed to open any
1003          * volume that has a LUN ID, so it can issue IOCTL to reread the
1004          * disk information.  I don't think I really like this
1005          * but I'm already using way to many device nodes to claim another one
1006          * for "raw controller".
1007          */
1008         if (drv->heads == 0) {
1009                 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
1010                         /* if not node 0 make sure it is a partition = 0 */
1011                         if (MINOR(bdev->bd_dev) & 0x0f) {
1012                                 return -ENXIO;
1013                                 /* if it is, make sure we have a LUN ID */
1014                         } else if (memcmp(drv->LunID, CTLR_LUNID,
1015                                 sizeof(drv->LunID))) {
1016                                 return -ENXIO;
1017                         }
1018                 }
1019                 if (!capable(CAP_SYS_ADMIN))
1020                         return -EPERM;
1021         }
1022         drv->usage_count++;
1023         h->usage_count++;
1024         return 0;
1025 }
1026
1027 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode)
1028 {
1029         int ret;
1030
1031         mutex_lock(&cciss_mutex);
1032         ret = cciss_open(bdev, mode);
1033         mutex_unlock(&cciss_mutex);
1034
1035         return ret;
1036 }
1037
1038 /*
1039  * Close.  Sync first.
1040  */
1041 static int cciss_release(struct gendisk *disk, fmode_t mode)
1042 {
1043         ctlr_info_t *h;
1044         drive_info_struct *drv;
1045
1046         mutex_lock(&cciss_mutex);
1047         h = get_host(disk);
1048         drv = get_drv(disk);
1049         dev_dbg(&h->pdev->dev, "cciss_release %s\n", disk->disk_name);
1050         drv->usage_count--;
1051         h->usage_count--;
1052         mutex_unlock(&cciss_mutex);
1053         return 0;
1054 }
1055
1056 static int do_ioctl(struct block_device *bdev, fmode_t mode,
1057                     unsigned cmd, unsigned long arg)
1058 {
1059         int ret;
1060         mutex_lock(&cciss_mutex);
1061         ret = cciss_ioctl(bdev, mode, cmd, arg);
1062         mutex_unlock(&cciss_mutex);
1063         return ret;
1064 }
1065
1066 #ifdef CONFIG_COMPAT
1067
1068 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1069                                   unsigned cmd, unsigned long arg);
1070 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1071                                       unsigned cmd, unsigned long arg);
1072
1073 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
1074                               unsigned cmd, unsigned long arg)
1075 {
1076         switch (cmd) {
1077         case CCISS_GETPCIINFO:
1078         case CCISS_GETINTINFO:
1079         case CCISS_SETINTINFO:
1080         case CCISS_GETNODENAME:
1081         case CCISS_SETNODENAME:
1082         case CCISS_GETHEARTBEAT:
1083         case CCISS_GETBUSTYPES:
1084         case CCISS_GETFIRMVER:
1085         case CCISS_GETDRIVVER:
1086         case CCISS_REVALIDVOLS:
1087         case CCISS_DEREGDISK:
1088         case CCISS_REGNEWDISK:
1089         case CCISS_REGNEWD:
1090         case CCISS_RESCANDISK:
1091         case CCISS_GETLUNINFO:
1092                 return do_ioctl(bdev, mode, cmd, arg);
1093
1094         case CCISS_PASSTHRU32:
1095                 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
1096         case CCISS_BIG_PASSTHRU32:
1097                 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
1098
1099         default:
1100                 return -ENOIOCTLCMD;
1101         }
1102 }
1103
1104 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1105                                   unsigned cmd, unsigned long arg)
1106 {
1107         IOCTL32_Command_struct __user *arg32 =
1108             (IOCTL32_Command_struct __user *) arg;
1109         IOCTL_Command_struct arg64;
1110         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
1111         int err;
1112         u32 cp;
1113
1114         err = 0;
1115         err |=
1116             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1117                            sizeof(arg64.LUN_info));
1118         err |=
1119             copy_from_user(&arg64.Request, &arg32->Request,
1120                            sizeof(arg64.Request));
1121         err |=
1122             copy_from_user(&arg64.error_info, &arg32->error_info,
1123                            sizeof(arg64.error_info));
1124         err |= get_user(arg64.buf_size, &arg32->buf_size);
1125         err |= get_user(cp, &arg32->buf);
1126         arg64.buf = compat_ptr(cp);
1127         err |= copy_to_user(p, &arg64, sizeof(arg64));
1128
1129         if (err)
1130                 return -EFAULT;
1131
1132         err = do_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
1133         if (err)
1134                 return err;
1135         err |=
1136             copy_in_user(&arg32->error_info, &p->error_info,
1137                          sizeof(arg32->error_info));
1138         if (err)
1139                 return -EFAULT;
1140         return err;
1141 }
1142
1143 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1144                                       unsigned cmd, unsigned long arg)
1145 {
1146         BIG_IOCTL32_Command_struct __user *arg32 =
1147             (BIG_IOCTL32_Command_struct __user *) arg;
1148         BIG_IOCTL_Command_struct arg64;
1149         BIG_IOCTL_Command_struct __user *p =
1150             compat_alloc_user_space(sizeof(arg64));
1151         int err;
1152         u32 cp;
1153
1154         memset(&arg64, 0, sizeof(arg64));
1155         err = 0;
1156         err |=
1157             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1158                            sizeof(arg64.LUN_info));
1159         err |=
1160             copy_from_user(&arg64.Request, &arg32->Request,
1161                            sizeof(arg64.Request));
1162         err |=
1163             copy_from_user(&arg64.error_info, &arg32->error_info,
1164                            sizeof(arg64.error_info));
1165         err |= get_user(arg64.buf_size, &arg32->buf_size);
1166         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1167         err |= get_user(cp, &arg32->buf);
1168         arg64.buf = compat_ptr(cp);
1169         err |= copy_to_user(p, &arg64, sizeof(arg64));
1170
1171         if (err)
1172                 return -EFAULT;
1173
1174         err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1175         if (err)
1176                 return err;
1177         err |=
1178             copy_in_user(&arg32->error_info, &p->error_info,
1179                          sizeof(arg32->error_info));
1180         if (err)
1181                 return -EFAULT;
1182         return err;
1183 }
1184 #endif
1185
1186 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1187 {
1188         drive_info_struct *drv = get_drv(bdev->bd_disk);
1189
1190         if (!drv->cylinders)
1191                 return -ENXIO;
1192
1193         geo->heads = drv->heads;
1194         geo->sectors = drv->sectors;
1195         geo->cylinders = drv->cylinders;
1196         return 0;
1197 }
1198
1199 static void check_ioctl_unit_attention(ctlr_info_t *h, CommandList_struct *c)
1200 {
1201         if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1202                         c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1203                 (void)check_for_unit_attention(h, c);
1204 }
1205
1206 static int cciss_getpciinfo(ctlr_info_t *h, void __user *argp)
1207 {
1208         cciss_pci_info_struct pciinfo;
1209
1210         if (!argp)
1211                 return -EINVAL;
1212         pciinfo.domain = pci_domain_nr(h->pdev->bus);
1213         pciinfo.bus = h->pdev->bus->number;
1214         pciinfo.dev_fn = h->pdev->devfn;
1215         pciinfo.board_id = h->board_id;
1216         if (copy_to_user(argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1217                 return -EFAULT;
1218         return 0;
1219 }
1220
1221 static int cciss_getintinfo(ctlr_info_t *h, void __user *argp)
1222 {
1223         cciss_coalint_struct intinfo;
1224
1225         if (!argp)
1226                 return -EINVAL;
1227         intinfo.delay = readl(&h->cfgtable->HostWrite.CoalIntDelay);
1228         intinfo.count = readl(&h->cfgtable->HostWrite.CoalIntCount);
1229         if (copy_to_user
1230             (argp, &intinfo, sizeof(cciss_coalint_struct)))
1231                 return -EFAULT;
1232         return 0;
1233 }
1234
1235 static int cciss_setintinfo(ctlr_info_t *h, void __user *argp)
1236 {
1237         cciss_coalint_struct intinfo;
1238         unsigned long flags;
1239         int i;
1240
1241         if (!argp)
1242                 return -EINVAL;
1243         if (!capable(CAP_SYS_ADMIN))
1244                 return -EPERM;
1245         if (copy_from_user(&intinfo, argp, sizeof(intinfo)))
1246                 return -EFAULT;
1247         if ((intinfo.delay == 0) && (intinfo.count == 0))
1248                 return -EINVAL;
1249         spin_lock_irqsave(&h->lock, flags);
1250         /* Update the field, and then ring the doorbell */
1251         writel(intinfo.delay, &(h->cfgtable->HostWrite.CoalIntDelay));
1252         writel(intinfo.count, &(h->cfgtable->HostWrite.CoalIntCount));
1253         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1254
1255         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1256                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
1257                         break;
1258                 udelay(1000); /* delay and try again */
1259         }
1260         spin_unlock_irqrestore(&h->lock, flags);
1261         if (i >= MAX_IOCTL_CONFIG_WAIT)
1262                 return -EAGAIN;
1263         return 0;
1264 }
1265
1266 static int cciss_getnodename(ctlr_info_t *h, void __user *argp)
1267 {
1268         NodeName_type NodeName;
1269         int i;
1270
1271         if (!argp)
1272                 return -EINVAL;
1273         for (i = 0; i < 16; i++)
1274                 NodeName[i] = readb(&h->cfgtable->ServerName[i]);
1275         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1276                 return -EFAULT;
1277         return 0;
1278 }
1279
1280 static int cciss_setnodename(ctlr_info_t *h, void __user *argp)
1281 {
1282         NodeName_type NodeName;
1283         unsigned long flags;
1284         int i;
1285
1286         if (!argp)
1287                 return -EINVAL;
1288         if (!capable(CAP_SYS_ADMIN))
1289                 return -EPERM;
1290         if (copy_from_user(NodeName, argp, sizeof(NodeName_type)))
1291                 return -EFAULT;
1292         spin_lock_irqsave(&h->lock, flags);
1293         /* Update the field, and then ring the doorbell */
1294         for (i = 0; i < 16; i++)
1295                 writeb(NodeName[i], &h->cfgtable->ServerName[i]);
1296         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1297         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1298                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
1299                         break;
1300                 udelay(1000); /* delay and try again */
1301         }
1302         spin_unlock_irqrestore(&h->lock, flags);
1303         if (i >= MAX_IOCTL_CONFIG_WAIT)
1304                 return -EAGAIN;
1305         return 0;
1306 }
1307
1308 static int cciss_getheartbeat(ctlr_info_t *h, void __user *argp)
1309 {
1310         Heartbeat_type heartbeat;
1311
1312         if (!argp)
1313                 return -EINVAL;
1314         heartbeat = readl(&h->cfgtable->HeartBeat);
1315         if (copy_to_user(argp, &heartbeat, sizeof(Heartbeat_type)))
1316                 return -EFAULT;
1317         return 0;
1318 }
1319
1320 static int cciss_getbustypes(ctlr_info_t *h, void __user *argp)
1321 {
1322         BusTypes_type BusTypes;
1323
1324         if (!argp)
1325                 return -EINVAL;
1326         BusTypes = readl(&h->cfgtable->BusTypes);
1327         if (copy_to_user(argp, &BusTypes, sizeof(BusTypes_type)))
1328                 return -EFAULT;
1329         return 0;
1330 }
1331
1332 static int cciss_getfirmver(ctlr_info_t *h, void __user *argp)
1333 {
1334         FirmwareVer_type firmware;
1335
1336         if (!argp)
1337                 return -EINVAL;
1338         memcpy(firmware, h->firm_ver, 4);
1339
1340         if (copy_to_user
1341             (argp, firmware, sizeof(FirmwareVer_type)))
1342                 return -EFAULT;
1343         return 0;
1344 }
1345
1346 static int cciss_getdrivver(ctlr_info_t *h, void __user *argp)
1347 {
1348         DriverVer_type DriverVer = DRIVER_VERSION;
1349
1350         if (!argp)
1351                 return -EINVAL;
1352         if (copy_to_user(argp, &DriverVer, sizeof(DriverVer_type)))
1353                 return -EFAULT;
1354         return 0;
1355 }
1356
1357 static int cciss_getluninfo(ctlr_info_t *h,
1358         struct gendisk *disk, void __user *argp)
1359 {
1360         LogvolInfo_struct luninfo;
1361         drive_info_struct *drv = get_drv(disk);
1362
1363         if (!argp)
1364                 return -EINVAL;
1365         memcpy(&luninfo.LunID, drv->LunID, sizeof(luninfo.LunID));
1366         luninfo.num_opens = drv->usage_count;
1367         luninfo.num_parts = 0;
1368         if (copy_to_user(argp, &luninfo, sizeof(LogvolInfo_struct)))
1369                 return -EFAULT;
1370         return 0;
1371 }
1372
1373 static int cciss_passthru(ctlr_info_t *h, void __user *argp)
1374 {
1375         IOCTL_Command_struct iocommand;
1376         CommandList_struct *c;
1377         char *buff = NULL;
1378         u64bit temp64;
1379         DECLARE_COMPLETION_ONSTACK(wait);
1380
1381         if (!argp)
1382                 return -EINVAL;
1383
1384         if (!capable(CAP_SYS_RAWIO))
1385                 return -EPERM;
1386
1387         if (copy_from_user
1388             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1389                 return -EFAULT;
1390         if ((iocommand.buf_size < 1) &&
1391             (iocommand.Request.Type.Direction != XFER_NONE)) {
1392                 return -EINVAL;
1393         }
1394         if (iocommand.buf_size > 0) {
1395                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1396                 if (buff == NULL)
1397                         return -EFAULT;
1398         }
1399         if (iocommand.Request.Type.Direction == XFER_WRITE) {
1400                 /* Copy the data into the buffer we created */
1401                 if (copy_from_user(buff, iocommand.buf, iocommand.buf_size)) {
1402                         kfree(buff);
1403                         return -EFAULT;
1404                 }
1405         } else {
1406                 memset(buff, 0, iocommand.buf_size);
1407         }
1408         c = cmd_special_alloc(h);
1409         if (!c) {
1410                 kfree(buff);
1411                 return -ENOMEM;
1412         }
1413         /* Fill in the command type */
1414         c->cmd_type = CMD_IOCTL_PEND;
1415         /* Fill in Command Header */
1416         c->Header.ReplyQueue = 0;   /* unused in simple mode */
1417         if (iocommand.buf_size > 0) { /* buffer to fill */
1418                 c->Header.SGList = 1;
1419                 c->Header.SGTotal = 1;
1420         } else { /* no buffers to fill */
1421                 c->Header.SGList = 0;
1422                 c->Header.SGTotal = 0;
1423         }
1424         c->Header.LUN = iocommand.LUN_info;
1425         /* use the kernel address the cmd block for tag */
1426         c->Header.Tag.lower = c->busaddr;
1427
1428         /* Fill in Request block */
1429         c->Request = iocommand.Request;
1430
1431         /* Fill in the scatter gather information */
1432         if (iocommand.buf_size > 0) {
1433                 temp64.val = pci_map_single(h->pdev, buff,
1434                         iocommand.buf_size, PCI_DMA_BIDIRECTIONAL);
1435                 c->SG[0].Addr.lower = temp64.val32.lower;
1436                 c->SG[0].Addr.upper = temp64.val32.upper;
1437                 c->SG[0].Len = iocommand.buf_size;
1438                 c->SG[0].Ext = 0;  /* we are not chaining */
1439         }
1440         c->waiting = &wait;
1441
1442         enqueue_cmd_and_start_io(h, c);
1443         wait_for_completion(&wait);
1444
1445         /* unlock the buffers from DMA */
1446         temp64.val32.lower = c->SG[0].Addr.lower;
1447         temp64.val32.upper = c->SG[0].Addr.upper;
1448         pci_unmap_single(h->pdev, (dma_addr_t) temp64.val, iocommand.buf_size,
1449                          PCI_DMA_BIDIRECTIONAL);
1450         check_ioctl_unit_attention(h, c);
1451
1452         /* Copy the error information out */
1453         iocommand.error_info = *(c->err_info);
1454         if (copy_to_user(argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1455                 kfree(buff);
1456                 cmd_special_free(h, c);
1457                 return -EFAULT;
1458         }
1459
1460         if (iocommand.Request.Type.Direction == XFER_READ) {
1461                 /* Copy the data out of the buffer we created */
1462                 if (copy_to_user(iocommand.buf, buff, iocommand.buf_size)) {
1463                         kfree(buff);
1464                         cmd_special_free(h, c);
1465                         return -EFAULT;
1466                 }
1467         }
1468         kfree(buff);
1469         cmd_special_free(h, c);
1470         return 0;
1471 }
1472
1473 static int cciss_bigpassthru(ctlr_info_t *h, void __user *argp)
1474 {
1475         BIG_IOCTL_Command_struct *ioc;
1476         CommandList_struct *c;
1477         unsigned char **buff = NULL;
1478         int *buff_size = NULL;
1479         u64bit temp64;
1480         BYTE sg_used = 0;
1481         int status = 0;
1482         int i;
1483         DECLARE_COMPLETION_ONSTACK(wait);
1484         __u32 left;
1485         __u32 sz;
1486         BYTE __user *data_ptr;
1487
1488         if (!argp)
1489                 return -EINVAL;
1490         if (!capable(CAP_SYS_RAWIO))
1491                 return -EPERM;
1492         ioc = (BIG_IOCTL_Command_struct *)
1493             kmalloc(sizeof(*ioc), GFP_KERNEL);
1494         if (!ioc) {
1495                 status = -ENOMEM;
1496                 goto cleanup1;
1497         }
1498         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1499                 status = -EFAULT;
1500                 goto cleanup1;
1501         }
1502         if ((ioc->buf_size < 1) &&
1503             (ioc->Request.Type.Direction != XFER_NONE)) {
1504                 status = -EINVAL;
1505                 goto cleanup1;
1506         }
1507         /* Check kmalloc limits  using all SGs */
1508         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1509                 status = -EINVAL;
1510                 goto cleanup1;
1511         }
1512         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1513                 status = -EINVAL;
1514                 goto cleanup1;
1515         }
1516         buff = kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1517         if (!buff) {
1518                 status = -ENOMEM;
1519                 goto cleanup1;
1520         }
1521         buff_size = kmalloc(MAXSGENTRIES * sizeof(int), GFP_KERNEL);
1522         if (!buff_size) {
1523                 status = -ENOMEM;
1524                 goto cleanup1;
1525         }
1526         left = ioc->buf_size;
1527         data_ptr = ioc->buf;
1528         while (left) {
1529                 sz = (left > ioc->malloc_size) ? ioc->malloc_size : left;
1530                 buff_size[sg_used] = sz;
1531                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1532                 if (buff[sg_used] == NULL) {
1533                         status = -ENOMEM;
1534                         goto cleanup1;
1535                 }
1536                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1537                         if (copy_from_user(buff[sg_used], data_ptr, sz)) {
1538                                 status = -EFAULT;
1539                                 goto cleanup1;
1540                         }
1541                 } else {
1542                         memset(buff[sg_used], 0, sz);
1543                 }
1544                 left -= sz;
1545                 data_ptr += sz;
1546                 sg_used++;
1547         }
1548         c = cmd_special_alloc(h);
1549         if (!c) {
1550                 status = -ENOMEM;
1551                 goto cleanup1;
1552         }
1553         c->cmd_type = CMD_IOCTL_PEND;
1554         c->Header.ReplyQueue = 0;
1555         c->Header.SGList = sg_used;
1556         c->Header.SGTotal = sg_used;
1557         c->Header.LUN = ioc->LUN_info;
1558         c->Header.Tag.lower = c->busaddr;
1559
1560         c->Request = ioc->Request;
1561         for (i = 0; i < sg_used; i++) {
1562                 temp64.val = pci_map_single(h->pdev, buff[i], buff_size[i],
1563                                     PCI_DMA_BIDIRECTIONAL);
1564                 c->SG[i].Addr.lower = temp64.val32.lower;
1565                 c->SG[i].Addr.upper = temp64.val32.upper;
1566                 c->SG[i].Len = buff_size[i];
1567                 c->SG[i].Ext = 0;       /* we are not chaining */
1568         }
1569         c->waiting = &wait;
1570         enqueue_cmd_and_start_io(h, c);
1571         wait_for_completion(&wait);
1572         /* unlock the buffers from DMA */
1573         for (i = 0; i < sg_used; i++) {
1574                 temp64.val32.lower = c->SG[i].Addr.lower;
1575                 temp64.val32.upper = c->SG[i].Addr.upper;
1576                 pci_unmap_single(h->pdev,
1577                         (dma_addr_t) temp64.val, buff_size[i],
1578                         PCI_DMA_BIDIRECTIONAL);
1579         }
1580         check_ioctl_unit_attention(h, c);
1581         /* Copy the error information out */
1582         ioc->error_info = *(c->err_info);
1583         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1584                 cmd_special_free(h, c);
1585                 status = -EFAULT;
1586                 goto cleanup1;
1587         }
1588         if (ioc->Request.Type.Direction == XFER_READ) {
1589                 /* Copy the data out of the buffer we created */
1590                 BYTE __user *ptr = ioc->buf;
1591                 for (i = 0; i < sg_used; i++) {
1592                         if (copy_to_user(ptr, buff[i], buff_size[i])) {
1593                                 cmd_special_free(h, c);
1594                                 status = -EFAULT;
1595                                 goto cleanup1;
1596                         }
1597                         ptr += buff_size[i];
1598                 }
1599         }
1600         cmd_special_free(h, c);
1601         status = 0;
1602 cleanup1:
1603         if (buff) {
1604                 for (i = 0; i < sg_used; i++)
1605                         kfree(buff[i]);
1606                 kfree(buff);
1607         }
1608         kfree(buff_size);
1609         kfree(ioc);
1610         return status;
1611 }
1612
1613 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1614         unsigned int cmd, unsigned long arg)
1615 {
1616         struct gendisk *disk = bdev->bd_disk;
1617         ctlr_info_t *h = get_host(disk);
1618         void __user *argp = (void __user *)arg;
1619
1620         dev_dbg(&h->pdev->dev, "cciss_ioctl: Called with cmd=%x %lx\n",
1621                 cmd, arg);
1622         switch (cmd) {
1623         case CCISS_GETPCIINFO:
1624                 return cciss_getpciinfo(h, argp);
1625         case CCISS_GETINTINFO:
1626                 return cciss_getintinfo(h, argp);
1627         case CCISS_SETINTINFO:
1628                 return cciss_setintinfo(h, argp);
1629         case CCISS_GETNODENAME:
1630                 return cciss_getnodename(h, argp);
1631         case CCISS_SETNODENAME:
1632                 return cciss_setnodename(h, argp);
1633         case CCISS_GETHEARTBEAT:
1634                 return cciss_getheartbeat(h, argp);
1635         case CCISS_GETBUSTYPES:
1636                 return cciss_getbustypes(h, argp);
1637         case CCISS_GETFIRMVER:
1638                 return cciss_getfirmver(h, argp);
1639         case CCISS_GETDRIVVER:
1640                 return cciss_getdrivver(h, argp);
1641         case CCISS_DEREGDISK:
1642         case CCISS_REGNEWD:
1643         case CCISS_REVALIDVOLS:
1644                 return rebuild_lun_table(h, 0, 1);
1645         case CCISS_GETLUNINFO:
1646                 return cciss_getluninfo(h, disk, argp);
1647         case CCISS_PASSTHRU:
1648                 return cciss_passthru(h, argp);
1649         case CCISS_BIG_PASSTHRU:
1650                 return cciss_bigpassthru(h, argp);
1651
1652         /* scsi_cmd_ioctl handles these, below, though some are not */
1653         /* very meaningful for cciss.  SG_IO is the main one people want. */
1654
1655         case SG_GET_VERSION_NUM:
1656         case SG_SET_TIMEOUT:
1657         case SG_GET_TIMEOUT:
1658         case SG_GET_RESERVED_SIZE:
1659         case SG_SET_RESERVED_SIZE:
1660         case SG_EMULATED_HOST:
1661         case SG_IO:
1662         case SCSI_IOCTL_SEND_COMMAND:
1663                 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
1664
1665         /* scsi_cmd_ioctl would normally handle these, below, but */
1666         /* they aren't a good fit for cciss, as CD-ROMs are */
1667         /* not supported, and we don't have any bus/target/lun */
1668         /* which we present to the kernel. */
1669
1670         case CDROM_SEND_PACKET:
1671         case CDROMCLOSETRAY:
1672         case CDROMEJECT:
1673         case SCSI_IOCTL_GET_IDLUN:
1674         case SCSI_IOCTL_GET_BUS_NUMBER:
1675         default:
1676                 return -ENOTTY;
1677         }
1678 }
1679
1680 static void cciss_check_queues(ctlr_info_t *h)
1681 {
1682         int start_queue = h->next_to_run;
1683         int i;
1684
1685         /* check to see if we have maxed out the number of commands that can
1686          * be placed on the queue.  If so then exit.  We do this check here
1687          * in case the interrupt we serviced was from an ioctl and did not
1688          * free any new commands.
1689          */
1690         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1691                 return;
1692
1693         /* We have room on the queue for more commands.  Now we need to queue
1694          * them up.  We will also keep track of the next queue to run so
1695          * that every queue gets a chance to be started first.
1696          */
1697         for (i = 0; i < h->highest_lun + 1; i++) {
1698                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1699                 /* make sure the disk has been added and the drive is real
1700                  * because this can be called from the middle of init_one.
1701                  */
1702                 if (!h->drv[curr_queue])
1703                         continue;
1704                 if (!(h->drv[curr_queue]->queue) ||
1705                         !(h->drv[curr_queue]->heads))
1706                         continue;
1707                 blk_start_queue(h->gendisk[curr_queue]->queue);
1708
1709                 /* check to see if we have maxed out the number of commands
1710                  * that can be placed on the queue.
1711                  */
1712                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1713                         if (curr_queue == start_queue) {
1714                                 h->next_to_run =
1715                                     (start_queue + 1) % (h->highest_lun + 1);
1716                                 break;
1717                         } else {
1718                                 h->next_to_run = curr_queue;
1719                                 break;
1720                         }
1721                 }
1722         }
1723 }
1724
1725 static void cciss_softirq_done(struct request *rq)
1726 {
1727         CommandList_struct *c = rq->completion_data;
1728         ctlr_info_t *h = hba[c->ctlr];
1729         SGDescriptor_struct *curr_sg = c->SG;
1730         u64bit temp64;
1731         unsigned long flags;
1732         int i, ddir;
1733         int sg_index = 0;
1734
1735         if (c->Request.Type.Direction == XFER_READ)
1736                 ddir = PCI_DMA_FROMDEVICE;
1737         else
1738                 ddir = PCI_DMA_TODEVICE;
1739
1740         /* command did not need to be retried */
1741         /* unmap the DMA mapping for all the scatter gather elements */
1742         for (i = 0; i < c->Header.SGList; i++) {
1743                 if (curr_sg[sg_index].Ext == CCISS_SG_CHAIN) {
1744                         cciss_unmap_sg_chain_block(h, c);
1745                         /* Point to the next block */
1746                         curr_sg = h->cmd_sg_list[c->cmdindex];
1747                         sg_index = 0;
1748                 }
1749                 temp64.val32.lower = curr_sg[sg_index].Addr.lower;
1750                 temp64.val32.upper = curr_sg[sg_index].Addr.upper;
1751                 pci_unmap_page(h->pdev, temp64.val, curr_sg[sg_index].Len,
1752                                 ddir);
1753                 ++sg_index;
1754         }
1755
1756         dev_dbg(&h->pdev->dev, "Done with %p\n", rq);
1757
1758         /* set the residual count for pc requests */
1759         if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
1760                 rq->resid_len = c->err_info->ResidualCnt;
1761
1762         blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1763
1764         spin_lock_irqsave(&h->lock, flags);
1765         cmd_free(h, c);
1766         cciss_check_queues(h);
1767         spin_unlock_irqrestore(&h->lock, flags);
1768 }
1769
1770 static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1771         unsigned char scsi3addr[], uint32_t log_unit)
1772 {
1773         memcpy(scsi3addr, h->drv[log_unit]->LunID,
1774                 sizeof(h->drv[log_unit]->LunID));
1775 }
1776
1777 /* This function gets the SCSI vendor, model, and revision of a logical drive
1778  * via the inquiry page 0.  Model, vendor, and rev are set to empty strings if
1779  * they cannot be read.
1780  */
1781 static void cciss_get_device_descr(ctlr_info_t *h, int logvol,
1782                                    char *vendor, char *model, char *rev)
1783 {
1784         int rc;
1785         InquiryData_struct *inq_buf;
1786         unsigned char scsi3addr[8];
1787
1788         *vendor = '\0';
1789         *model = '\0';
1790         *rev = '\0';
1791
1792         inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1793         if (!inq_buf)
1794                 return;
1795
1796         log_unit_to_scsi3addr(h, scsi3addr, logvol);
1797         rc = sendcmd_withirq(h, CISS_INQUIRY, inq_buf, sizeof(*inq_buf), 0,
1798                         scsi3addr, TYPE_CMD);
1799         if (rc == IO_OK) {
1800                 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1801                 vendor[VENDOR_LEN] = '\0';
1802                 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1803                 model[MODEL_LEN] = '\0';
1804                 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1805                 rev[REV_LEN] = '\0';
1806         }
1807
1808         kfree(inq_buf);
1809         return;
1810 }
1811
1812 /* This function gets the serial number of a logical drive via
1813  * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1814  * number cannot be had, for whatever reason, 16 bytes of 0xff
1815  * are returned instead.
1816  */
1817 static void cciss_get_serial_no(ctlr_info_t *h, int logvol,
1818                                 unsigned char *serial_no, int buflen)
1819 {
1820 #define PAGE_83_INQ_BYTES 64
1821         int rc;
1822         unsigned char *buf;
1823         unsigned char scsi3addr[8];
1824
1825         if (buflen > 16)
1826                 buflen = 16;
1827         memset(serial_no, 0xff, buflen);
1828         buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1829         if (!buf)
1830                 return;
1831         memset(serial_no, 0, buflen);
1832         log_unit_to_scsi3addr(h, scsi3addr, logvol);
1833         rc = sendcmd_withirq(h, CISS_INQUIRY, buf,
1834                 PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1835         if (rc == IO_OK)
1836                 memcpy(serial_no, &buf[8], buflen);
1837         kfree(buf);
1838         return;
1839 }
1840
1841 /*
1842  * cciss_add_disk sets up the block device queue for a logical drive
1843  */
1844 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1845                                 int drv_index)
1846 {
1847         disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1848         if (!disk->queue)
1849                 goto init_queue_failure;
1850         sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1851         disk->major = h->major;
1852         disk->first_minor = drv_index << NWD_SHIFT;
1853         disk->fops = &cciss_fops;
1854         if (cciss_create_ld_sysfs_entry(h, drv_index))
1855                 goto cleanup_queue;
1856         disk->private_data = h->drv[drv_index];
1857         disk->driverfs_dev = &h->drv[drv_index]->dev;
1858
1859         /* Set up queue information */
1860         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1861
1862         /* This is a hardware imposed limit. */
1863         blk_queue_max_segments(disk->queue, h->maxsgentries);
1864
1865         blk_queue_max_hw_sectors(disk->queue, h->cciss_max_sectors);
1866
1867         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1868
1869         disk->queue->queuedata = h;
1870
1871         blk_queue_logical_block_size(disk->queue,
1872                                      h->drv[drv_index]->block_size);
1873
1874         /* Make sure all queue data is written out before */
1875         /* setting h->drv[drv_index]->queue, as setting this */
1876         /* allows the interrupt handler to start the queue */
1877         wmb();
1878         h->drv[drv_index]->queue = disk->queue;
1879         add_disk(disk);
1880         return 0;
1881
1882 cleanup_queue:
1883         blk_cleanup_queue(disk->queue);
1884         disk->queue = NULL;
1885 init_queue_failure:
1886         return -1;
1887 }
1888
1889 /* This function will check the usage_count of the drive to be updated/added.
1890  * If the usage_count is zero and it is a heretofore unknown drive, or,
1891  * the drive's capacity, geometry, or serial number has changed,
1892  * then the drive information will be updated and the disk will be
1893  * re-registered with the kernel.  If these conditions don't hold,
1894  * then it will be left alone for the next reboot.  The exception to this
1895  * is disk 0 which will always be left registered with the kernel since it
1896  * is also the controller node.  Any changes to disk 0 will show up on
1897  * the next reboot.
1898  */
1899 static void cciss_update_drive_info(ctlr_info_t *h, int drv_index,
1900         int first_time, int via_ioctl)
1901 {
1902         struct gendisk *disk;
1903         InquiryData_struct *inq_buff = NULL;
1904         unsigned int block_size;
1905         sector_t total_size;
1906         unsigned long flags = 0;
1907         int ret = 0;
1908         drive_info_struct *drvinfo;
1909
1910         /* Get information about the disk and modify the driver structure */
1911         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1912         drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
1913         if (inq_buff == NULL || drvinfo == NULL)
1914                 goto mem_msg;
1915
1916         /* testing to see if 16-byte CDBs are already being used */
1917         if (h->cciss_read == CCISS_READ_16) {
1918                 cciss_read_capacity_16(h, drv_index,
1919                         &total_size, &block_size);
1920
1921         } else {
1922                 cciss_read_capacity(h, drv_index, &total_size, &block_size);
1923                 /* if read_capacity returns all F's this volume is >2TB */
1924                 /* in size so we switch to 16-byte CDB's for all */
1925                 /* read/write ops */
1926                 if (total_size == 0xFFFFFFFFULL) {
1927                         cciss_read_capacity_16(h, drv_index,
1928                         &total_size, &block_size);
1929                         h->cciss_read = CCISS_READ_16;
1930                         h->cciss_write = CCISS_WRITE_16;
1931                 } else {
1932                         h->cciss_read = CCISS_READ_10;
1933                         h->cciss_write = CCISS_WRITE_10;
1934                 }
1935         }
1936
1937         cciss_geometry_inquiry(h, drv_index, total_size, block_size,
1938                                inq_buff, drvinfo);
1939         drvinfo->block_size = block_size;
1940         drvinfo->nr_blocks = total_size + 1;
1941
1942         cciss_get_device_descr(h, drv_index, drvinfo->vendor,
1943                                 drvinfo->model, drvinfo->rev);
1944         cciss_get_serial_no(h, drv_index, drvinfo->serial_no,
1945                         sizeof(drvinfo->serial_no));
1946         /* Save the lunid in case we deregister the disk, below. */
1947         memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
1948                 sizeof(drvinfo->LunID));
1949
1950         /* Is it the same disk we already know, and nothing's changed? */
1951         if (h->drv[drv_index]->raid_level != -1 &&
1952                 ((memcmp(drvinfo->serial_no,
1953                                 h->drv[drv_index]->serial_no, 16) == 0) &&
1954                 drvinfo->block_size == h->drv[drv_index]->block_size &&
1955                 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
1956                 drvinfo->heads == h->drv[drv_index]->heads &&
1957                 drvinfo->sectors == h->drv[drv_index]->sectors &&
1958                 drvinfo->cylinders == h->drv[drv_index]->cylinders))
1959                         /* The disk is unchanged, nothing to update */
1960                         goto freeret;
1961
1962         /* If we get here it's not the same disk, or something's changed,
1963          * so we need to * deregister it, and re-register it, if it's not
1964          * in use.
1965          * If the disk already exists then deregister it before proceeding
1966          * (unless it's the first disk (for the controller node).
1967          */
1968         if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
1969                 dev_warn(&h->pdev->dev, "disk %d has changed.\n", drv_index);
1970                 spin_lock_irqsave(&h->lock, flags);
1971                 h->drv[drv_index]->busy_configuring = 1;
1972                 spin_unlock_irqrestore(&h->lock, flags);
1973
1974                 /* deregister_disk sets h->drv[drv_index]->queue = NULL
1975                  * which keeps the interrupt handler from starting
1976                  * the queue.
1977                  */
1978                 ret = deregister_disk(h, drv_index, 0, via_ioctl);
1979         }
1980
1981         /* If the disk is in use return */
1982         if (ret)
1983                 goto freeret;
1984
1985         /* Save the new information from cciss_geometry_inquiry
1986          * and serial number inquiry.  If the disk was deregistered
1987          * above, then h->drv[drv_index] will be NULL.
1988          */
1989         if (h->drv[drv_index] == NULL) {
1990                 drvinfo->device_initialized = 0;
1991                 h->drv[drv_index] = drvinfo;
1992                 drvinfo = NULL; /* so it won't be freed below. */
1993         } else {
1994                 /* special case for cxd0 */
1995                 h->drv[drv_index]->block_size = drvinfo->block_size;
1996                 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
1997                 h->drv[drv_index]->heads = drvinfo->heads;
1998                 h->drv[drv_index]->sectors = drvinfo->sectors;
1999                 h->drv[drv_index]->cylinders = drvinfo->cylinders;
2000                 h->drv[drv_index]->raid_level = drvinfo->raid_level;
2001                 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
2002                 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
2003                         VENDOR_LEN + 1);
2004                 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
2005                 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
2006         }
2007
2008         ++h->num_luns;
2009         disk = h->gendisk[drv_index];
2010         set_capacity(disk, h->drv[drv_index]->nr_blocks);
2011
2012         /* If it's not disk 0 (drv_index != 0)
2013          * or if it was disk 0, but there was previously
2014          * no actual corresponding configured logical drive
2015          * (raid_leve == -1) then we want to update the
2016          * logical drive's information.
2017          */
2018         if (drv_index || first_time) {
2019                 if (cciss_add_disk(h, disk, drv_index) != 0) {
2020                         cciss_free_gendisk(h, drv_index);
2021                         cciss_free_drive_info(h, drv_index);
2022                         dev_warn(&h->pdev->dev, "could not update disk %d\n",
2023                                 drv_index);
2024                         --h->num_luns;
2025                 }
2026         }
2027
2028 freeret:
2029         kfree(inq_buff);
2030         kfree(drvinfo);
2031         return;
2032 mem_msg:
2033         dev_err(&h->pdev->dev, "out of memory\n");
2034         goto freeret;
2035 }
2036
2037 /* This function will find the first index of the controllers drive array
2038  * that has a null drv pointer and allocate the drive info struct and
2039  * will return that index   This is where new drives will be added.
2040  * If the index to be returned is greater than the highest_lun index for
2041  * the controller then highest_lun is set * to this new index.
2042  * If there are no available indexes or if tha allocation fails, then -1
2043  * is returned.  * "controller_node" is used to know if this is a real
2044  * logical drive, or just the controller node, which determines if this
2045  * counts towards highest_lun.
2046  */
2047 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
2048 {
2049         int i;
2050         drive_info_struct *drv;
2051
2052         /* Search for an empty slot for our drive info */
2053         for (i = 0; i < CISS_MAX_LUN; i++) {
2054
2055                 /* if not cxd0 case, and it's occupied, skip it. */
2056                 if (h->drv[i] && i != 0)
2057                         continue;
2058                 /*
2059                  * If it's cxd0 case, and drv is alloc'ed already, and a
2060                  * disk is configured there, skip it.
2061                  */
2062                 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2063                         continue;
2064
2065                 /*
2066                  * We've found an empty slot.  Update highest_lun
2067                  * provided this isn't just the fake cxd0 controller node.
2068                  */
2069                 if (i > h->highest_lun && !controller_node)
2070                         h->highest_lun = i;
2071
2072                 /* If adding a real disk at cxd0, and it's already alloc'ed */
2073                 if (i == 0 && h->drv[i] != NULL)
2074                         return i;
2075
2076                 /*
2077                  * Found an empty slot, not already alloc'ed.  Allocate it.
2078                  * Mark it with raid_level == -1, so we know it's new later on.
2079                  */
2080                 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2081                 if (!drv)
2082                         return -1;
2083                 drv->raid_level = -1; /* so we know it's new */
2084                 h->drv[i] = drv;
2085                 return i;
2086         }
2087         return -1;
2088 }
2089
2090 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2091 {
2092         kfree(h->drv[drv_index]);
2093         h->drv[drv_index] = NULL;
2094 }
2095
2096 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2097 {
2098         put_disk(h->gendisk[drv_index]);
2099         h->gendisk[drv_index] = NULL;
2100 }
2101
2102 /* cciss_add_gendisk finds a free hba[]->drv structure
2103  * and allocates a gendisk if needed, and sets the lunid
2104  * in the drvinfo structure.   It returns the index into
2105  * the ->drv[] array, or -1 if none are free.
2106  * is_controller_node indicates whether highest_lun should
2107  * count this disk, or if it's only being added to provide
2108  * a means to talk to the controller in case no logical
2109  * drives have yet been configured.
2110  */
2111 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2112         int controller_node)
2113 {
2114         int drv_index;
2115
2116         drv_index = cciss_alloc_drive_info(h, controller_node);
2117         if (drv_index == -1)
2118                 return -1;
2119
2120         /*Check if the gendisk needs to be allocated */
2121         if (!h->gendisk[drv_index]) {
2122                 h->gendisk[drv_index] =
2123                         alloc_disk(1 << NWD_SHIFT);
2124                 if (!h->gendisk[drv_index]) {
2125                         dev_err(&h->pdev->dev,
2126                                 "could not allocate a new disk %d\n",
2127                                 drv_index);
2128                         goto err_free_drive_info;
2129                 }
2130         }
2131         memcpy(h->drv[drv_index]->LunID, lunid,
2132                 sizeof(h->drv[drv_index]->LunID));
2133         if (cciss_create_ld_sysfs_entry(h, drv_index))
2134                 goto err_free_disk;
2135         /* Don't need to mark this busy because nobody */
2136         /* else knows about this disk yet to contend */
2137         /* for access to it. */
2138         h->drv[drv_index]->busy_configuring = 0;
2139         wmb();
2140         return drv_index;
2141
2142 err_free_disk:
2143         cciss_free_gendisk(h, drv_index);
2144 err_free_drive_info:
2145         cciss_free_drive_info(h, drv_index);
2146         return -1;
2147 }
2148
2149 /* This is for the special case of a controller which
2150  * has no logical drives.  In this case, we still need
2151  * to register a disk so the controller can be accessed
2152  * by the Array Config Utility.
2153  */
2154 static void cciss_add_controller_node(ctlr_info_t *h)
2155 {
2156         struct gendisk *disk;
2157         int drv_index;
2158
2159         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2160                 return;
2161
2162         drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2163         if (drv_index == -1)
2164                 goto error;
2165         h->drv[drv_index]->block_size = 512;
2166         h->drv[drv_index]->nr_blocks = 0;
2167         h->drv[drv_index]->heads = 0;
2168         h->drv[drv_index]->sectors = 0;
2169         h->drv[drv_index]->cylinders = 0;
2170         h->drv[drv_index]->raid_level = -1;
2171         memset(h->drv[drv_index]->serial_no, 0, 16);
2172         disk = h->gendisk[drv_index];
2173         if (cciss_add_disk(h, disk, drv_index) == 0)
2174                 return;
2175         cciss_free_gendisk(h, drv_index);
2176         cciss_free_drive_info(h, drv_index);
2177 error:
2178         dev_warn(&h->pdev->dev, "could not add disk 0.\n");
2179         return;
2180 }
2181
2182 /* This function will add and remove logical drives from the Logical
2183  * drive array of the controller and maintain persistency of ordering
2184  * so that mount points are preserved until the next reboot.  This allows
2185  * for the removal of logical drives in the middle of the drive array
2186  * without a re-ordering of those drives.
2187  * INPUT
2188  * h            = The controller to perform the operations on
2189  */
2190 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2191         int via_ioctl)
2192 {
2193         int num_luns;
2194         ReportLunData_struct *ld_buff = NULL;
2195         int return_code;
2196         int listlength = 0;
2197         int i;
2198         int drv_found;
2199         int drv_index = 0;
2200         unsigned char lunid[8] = CTLR_LUNID;
2201         unsigned long flags;
2202
2203         if (!capable(CAP_SYS_RAWIO))
2204                 return -EPERM;
2205
2206         /* Set busy_configuring flag for this operation */
2207         spin_lock_irqsave(&h->lock, flags);
2208         if (h->busy_configuring) {
2209                 spin_unlock_irqrestore(&h->lock, flags);
2210                 return -EBUSY;
2211         }
2212         h->busy_configuring = 1;
2213         spin_unlock_irqrestore(&h->lock, flags);
2214
2215         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2216         if (ld_buff == NULL)
2217                 goto mem_msg;
2218
2219         return_code = sendcmd_withirq(h, CISS_REPORT_LOG, ld_buff,
2220                                       sizeof(ReportLunData_struct),
2221                                       0, CTLR_LUNID, TYPE_CMD);
2222
2223         if (return_code == IO_OK)
2224                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2225         else {  /* reading number of logical volumes failed */
2226                 dev_warn(&h->pdev->dev,
2227                         "report logical volume command failed\n");
2228                 listlength = 0;
2229                 goto freeret;
2230         }
2231
2232         num_luns = listlength / 8;      /* 8 bytes per entry */
2233         if (num_luns > CISS_MAX_LUN) {
2234                 num_luns = CISS_MAX_LUN;
2235                 dev_warn(&h->pdev->dev, "more luns configured"
2236                        " on controller than can be handled by"
2237                        " this driver.\n");
2238         }
2239
2240         if (num_luns == 0)
2241                 cciss_add_controller_node(h);
2242
2243         /* Compare controller drive array to driver's drive array
2244          * to see if any drives are missing on the controller due
2245          * to action of Array Config Utility (user deletes drive)
2246          * and deregister logical drives which have disappeared.
2247          */
2248         for (i = 0; i <= h->highest_lun; i++) {
2249                 int j;
2250                 drv_found = 0;
2251
2252                 /* skip holes in the array from already deleted drives */
2253                 if (h->drv[i] == NULL)
2254                         continue;
2255
2256                 for (j = 0; j < num_luns; j++) {
2257                         memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2258                         if (memcmp(h->drv[i]->LunID, lunid,
2259                                 sizeof(lunid)) == 0) {
2260                                 drv_found = 1;
2261                                 break;
2262                         }
2263                 }
2264                 if (!drv_found) {
2265                         /* Deregister it from the OS, it's gone. */
2266                         spin_lock_irqsave(&h->lock, flags);
2267                         h->drv[i]->busy_configuring = 1;
2268                         spin_unlock_irqrestore(&h->lock, flags);
2269                         return_code = deregister_disk(h, i, 1, via_ioctl);
2270                         if (h->drv[i] != NULL)
2271                                 h->drv[i]->busy_configuring = 0;
2272                 }
2273         }
2274
2275         /* Compare controller drive array to driver's drive array.
2276          * Check for updates in the drive information and any new drives
2277          * on the controller due to ACU adding logical drives, or changing
2278          * a logical drive's size, etc.  Reregister any new/changed drives
2279          */
2280         for (i = 0; i < num_luns; i++) {
2281                 int j;
2282
2283                 drv_found = 0;
2284
2285                 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2286                 /* Find if the LUN is already in the drive array
2287                  * of the driver.  If so then update its info
2288                  * if not in use.  If it does not exist then find
2289                  * the first free index and add it.
2290                  */
2291                 for (j = 0; j <= h->highest_lun; j++) {
2292                         if (h->drv[j] != NULL &&
2293                                 memcmp(h->drv[j]->LunID, lunid,
2294                                         sizeof(h->drv[j]->LunID)) == 0) {
2295                                 drv_index = j;
2296                                 drv_found = 1;
2297                                 break;
2298                         }
2299                 }
2300
2301                 /* check if the drive was found already in the array */
2302                 if (!drv_found) {
2303                         drv_index = cciss_add_gendisk(h, lunid, 0);
2304                         if (drv_index == -1)
2305                                 goto freeret;
2306                 }
2307                 cciss_update_drive_info(h, drv_index, first_time, via_ioctl);
2308         }               /* end for */
2309
2310 freeret:
2311         kfree(ld_buff);
2312         h->busy_configuring = 0;
2313         /* We return -1 here to tell the ACU that we have registered/updated
2314          * all of the drives that we can and to keep it from calling us
2315          * additional times.
2316          */
2317         return -1;
2318 mem_msg:
2319         dev_err(&h->pdev->dev, "out of memory\n");
2320         h->busy_configuring = 0;
2321         goto freeret;
2322 }
2323
2324 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2325 {
2326         /* zero out the disk size info */
2327         drive_info->nr_blocks = 0;
2328         drive_info->block_size = 0;
2329         drive_info->heads = 0;
2330         drive_info->sectors = 0;
2331         drive_info->cylinders = 0;
2332         drive_info->raid_level = -1;
2333         memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2334         memset(drive_info->model, 0, sizeof(drive_info->model));
2335         memset(drive_info->rev, 0, sizeof(drive_info->rev));
2336         memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2337         /*
2338          * don't clear the LUNID though, we need to remember which
2339          * one this one is.
2340          */
2341 }
2342
2343 /* This function will deregister the disk and it's queue from the
2344  * kernel.  It must be called with the controller lock held and the
2345  * drv structures busy_configuring flag set.  It's parameters are:
2346  *
2347  * disk = This is the disk to be deregistered
2348  * drv  = This is the drive_info_struct associated with the disk to be
2349  *        deregistered.  It contains information about the disk used
2350  *        by the driver.
2351  * clear_all = This flag determines whether or not the disk information
2352  *             is going to be completely cleared out and the highest_lun
2353  *             reset.  Sometimes we want to clear out information about
2354  *             the disk in preparation for re-adding it.  In this case
2355  *             the highest_lun should be left unchanged and the LunID
2356  *             should not be cleared.
2357  * via_ioctl
2358  *    This indicates whether we've reached this path via ioctl.
2359  *    This affects the maximum usage count allowed for c0d0 to be messed with.
2360  *    If this path is reached via ioctl(), then the max_usage_count will
2361  *    be 1, as the process calling ioctl() has got to have the device open.
2362  *    If we get here via sysfs, then the max usage count will be zero.
2363 */
2364 static int deregister_disk(ctlr_info_t *h, int drv_index,
2365                            int clear_all, int via_ioctl)
2366 {
2367         int i;
2368         struct gendisk *disk;
2369         drive_info_struct *drv;
2370         int recalculate_highest_lun;
2371
2372         if (!capable(CAP_SYS_RAWIO))
2373                 return -EPERM;
2374
2375         drv = h->drv[drv_index];
2376         disk = h->gendisk[drv_index];
2377
2378         /* make sure logical volume is NOT is use */
2379         if (clear_all || (h->gendisk[0] == disk)) {
2380                 if (drv->usage_count > via_ioctl)
2381                         return -EBUSY;
2382         } else if (drv->usage_count > 0)
2383                 return -EBUSY;
2384
2385         recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2386
2387         /* invalidate the devices and deregister the disk.  If it is disk
2388          * zero do not deregister it but just zero out it's values.  This
2389          * allows us to delete disk zero but keep the controller registered.
2390          */
2391         if (h->gendisk[0] != disk) {
2392                 struct request_queue *q = disk->queue;
2393                 if (disk->flags & GENHD_FL_UP) {
2394                         cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2395                         del_gendisk(disk);
2396                 }
2397                 if (q)
2398                         blk_cleanup_queue(q);
2399                 /* If clear_all is set then we are deleting the logical
2400                  * drive, not just refreshing its info.  For drives
2401                  * other than disk 0 we will call put_disk.  We do not
2402                  * do this for disk 0 as we need it to be able to
2403                  * configure the controller.
2404                  */
2405                 if (clear_all){
2406                         /* This isn't pretty, but we need to find the
2407                          * disk in our array and NULL our the pointer.
2408                          * This is so that we will call alloc_disk if
2409                          * this index is used again later.
2410                          */
2411                         for (i=0; i < CISS_MAX_LUN; i++){
2412                                 if (h->gendisk[i] == disk) {
2413                                         h->gendisk[i] = NULL;
2414                                         break;
2415                                 }
2416                         }
2417                         put_disk(disk);
2418                 }
2419         } else {
2420                 set_capacity(disk, 0);
2421                 cciss_clear_drive_info(drv);
2422         }
2423
2424         --h->num_luns;
2425
2426         /* if it was the last disk, find the new hightest lun */
2427         if (clear_all && recalculate_highest_lun) {
2428                 int newhighest = -1;
2429                 for (i = 0; i <= h->highest_lun; i++) {
2430                         /* if the disk has size > 0, it is available */
2431                         if (h->drv[i] && h->drv[i]->heads)
2432                                 newhighest = i;
2433                 }
2434                 h->highest_lun = newhighest;
2435         }
2436         return 0;
2437 }
2438
2439 static int fill_cmd(ctlr_info_t *h, CommandList_struct *c, __u8 cmd, void *buff,
2440                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2441                 int cmd_type)
2442 {
2443         u64bit buff_dma_handle;
2444         int status = IO_OK;
2445
2446         c->cmd_type = CMD_IOCTL_PEND;
2447         c->Header.ReplyQueue = 0;
2448         if (buff != NULL) {
2449                 c->Header.SGList = 1;
2450                 c->Header.SGTotal = 1;
2451         } else {
2452                 c->Header.SGList = 0;
2453                 c->Header.SGTotal = 0;
2454         }
2455         c->Header.Tag.lower = c->busaddr;
2456         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2457
2458         c->Request.Type.Type = cmd_type;
2459         if (cmd_type == TYPE_CMD) {
2460                 switch (cmd) {
2461                 case CISS_INQUIRY:
2462                         /* are we trying to read a vital product page */
2463                         if (page_code != 0) {
2464                                 c->Request.CDB[1] = 0x01;
2465                                 c->Request.CDB[2] = page_code;
2466                         }
2467                         c->Request.CDBLen = 6;
2468                         c->Request.Type.Attribute = ATTR_SIMPLE;
2469                         c->Request.Type.Direction = XFER_READ;
2470                         c->Request.Timeout = 0;
2471                         c->Request.CDB[0] = CISS_INQUIRY;
2472                         c->Request.CDB[4] = size & 0xFF;
2473                         break;
2474                 case CISS_REPORT_LOG:
2475                 case CISS_REPORT_PHYS:
2476                         /* Talking to controller so It's a physical command
2477                            mode = 00 target = 0.  Nothing to write.
2478                          */
2479                         c->Request.CDBLen = 12;
2480                         c->Request.Type.Attribute = ATTR_SIMPLE;
2481                         c->Request.Type.Direction = XFER_READ;
2482                         c->Request.Timeout = 0;
2483                         c->Request.CDB[0] = cmd;
2484                         c->Request.CDB[6] = (size >> 24) & 0xFF; /* MSB */
2485                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2486                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2487                         c->Request.CDB[9] = size & 0xFF;
2488                         break;
2489
2490                 case CCISS_READ_CAPACITY:
2491                         c->Request.CDBLen = 10;
2492                         c->Request.Type.Attribute = ATTR_SIMPLE;
2493                         c->Request.Type.Direction = XFER_READ;
2494                         c->Request.Timeout = 0;
2495                         c->Request.CDB[0] = cmd;
2496                         break;
2497                 case CCISS_READ_CAPACITY_16:
2498                         c->Request.CDBLen = 16;
2499                         c->Request.Type.Attribute = ATTR_SIMPLE;
2500                         c->Request.Type.Direction = XFER_READ;
2501                         c->Request.Timeout = 0;
2502                         c->Request.CDB[0] = cmd;
2503                         c->Request.CDB[1] = 0x10;
2504                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2505                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2506                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2507                         c->Request.CDB[13] = size & 0xFF;
2508                         c->Request.Timeout = 0;
2509                         c->Request.CDB[0] = cmd;
2510                         break;
2511                 case CCISS_CACHE_FLUSH:
2512                         c->Request.CDBLen = 12;
2513                         c->Request.Type.Attribute = ATTR_SIMPLE;
2514                         c->Request.Type.Direction = XFER_WRITE;
2515                         c->Request.Timeout = 0;
2516                         c->Request.CDB[0] = BMIC_WRITE;
2517                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2518                         break;
2519                 case TEST_UNIT_READY:
2520                         c->Request.CDBLen = 6;
2521                         c->Request.Type.Attribute = ATTR_SIMPLE;
2522                         c->Request.Type.Direction = XFER_NONE;
2523                         c->Request.Timeout = 0;
2524                         break;
2525                 default:
2526                         dev_warn(&h->pdev->dev, "Unknown Command 0x%c\n", cmd);
2527                         return IO_ERROR;
2528                 }
2529         } else if (cmd_type == TYPE_MSG) {
2530                 switch (cmd) {
2531                 case 0: /* ABORT message */
2532                         c->Request.CDBLen = 12;
2533                         c->Request.Type.Attribute = ATTR_SIMPLE;
2534                         c->Request.Type.Direction = XFER_WRITE;
2535                         c->Request.Timeout = 0;
2536                         c->Request.CDB[0] = cmd;        /* abort */
2537                         c->Request.CDB[1] = 0;  /* abort a command */
2538                         /* buff contains the tag of the command to abort */
2539                         memcpy(&c->Request.CDB[4], buff, 8);
2540                         break;
2541                 case 1: /* RESET message */
2542                         c->Request.CDBLen = 16;
2543                         c->Request.Type.Attribute = ATTR_SIMPLE;
2544                         c->Request.Type.Direction = XFER_NONE;
2545                         c->Request.Timeout = 0;
2546                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2547                         c->Request.CDB[0] = cmd;        /* reset */
2548                         c->Request.CDB[1] = 0x03;       /* reset a target */
2549                         break;
2550                 case 3: /* No-Op message */
2551                         c->Request.CDBLen = 1;
2552                         c->Request.Type.Attribute = ATTR_SIMPLE;
2553                         c->Request.Type.Direction = XFER_WRITE;
2554                         c->Request.Timeout = 0;
2555                         c->Request.CDB[0] = cmd;
2556                         break;
2557                 default:
2558                         dev_warn(&h->pdev->dev,
2559                                 "unknown message type %d\n", cmd);
2560                         return IO_ERROR;
2561                 }
2562         } else {
2563                 dev_warn(&h->pdev->dev, "unknown command type %d\n", cmd_type);
2564                 return IO_ERROR;
2565         }
2566         /* Fill in the scatter gather information */
2567         if (size > 0) {
2568                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2569                                                              buff, size,
2570                                                              PCI_DMA_BIDIRECTIONAL);
2571                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2572                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2573                 c->SG[0].Len = size;
2574                 c->SG[0].Ext = 0;       /* we are not chaining */
2575         }
2576         return status;
2577 }
2578
2579 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2580 {
2581         switch (c->err_info->ScsiStatus) {
2582         case SAM_STAT_GOOD:
2583                 return IO_OK;
2584         case SAM_STAT_CHECK_CONDITION:
2585                 switch (0xf & c->err_info->SenseInfo[2]) {
2586                 case 0: return IO_OK; /* no sense */
2587                 case 1: return IO_OK; /* recovered error */
2588                 default:
2589                         if (check_for_unit_attention(h, c))
2590                                 return IO_NEEDS_RETRY;
2591                         dev_warn(&h->pdev->dev, "cmd 0x%02x "
2592                                 "check condition, sense key = 0x%02x\n",
2593                                 c->Request.CDB[0], c->err_info->SenseInfo[2]);
2594                 }
2595                 break;
2596         default:
2597                 dev_warn(&h->pdev->dev, "cmd 0x%02x"
2598                         "scsi status = 0x%02x\n",
2599                         c->Request.CDB[0], c->err_info->ScsiStatus);
2600                 break;
2601         }
2602         return IO_ERROR;
2603 }
2604
2605 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2606 {
2607         int return_status = IO_OK;
2608
2609         if (c->err_info->CommandStatus == CMD_SUCCESS)
2610                 return IO_OK;
2611
2612         switch (c->err_info->CommandStatus) {
2613         case CMD_TARGET_STATUS:
2614                 return_status = check_target_status(h, c);
2615                 break;
2616         case CMD_DATA_UNDERRUN:
2617         case CMD_DATA_OVERRUN:
2618                 /* expected for inquiry and report lun commands */
2619                 break;
2620         case CMD_INVALID:
2621                 dev_warn(&h->pdev->dev, "cmd 0x%02x is "
2622                        "reported invalid\n", c->Request.CDB[0]);
2623                 return_status = IO_ERROR;
2624                 break;
2625         case CMD_PROTOCOL_ERR:
2626                 dev_warn(&h->pdev->dev, "cmd 0x%02x has "
2627                        "protocol error\n", c->Request.CDB[0]);
2628                 return_status = IO_ERROR;
2629                 break;
2630         case CMD_HARDWARE_ERR:
2631                 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2632                        " hardware error\n", c->Request.CDB[0]);
2633                 return_status = IO_ERROR;
2634                 break;
2635         case CMD_CONNECTION_LOST:
2636                 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2637                        "connection lost\n", c->Request.CDB[0]);
2638                 return_status = IO_ERROR;
2639                 break;
2640         case CMD_ABORTED:
2641                 dev_warn(&h->pdev->dev, "cmd 0x%02x was "
2642                        "aborted\n", c->Request.CDB[0]);
2643                 return_status = IO_ERROR;
2644                 break;
2645         case CMD_ABORT_FAILED:
2646                 dev_warn(&h->pdev->dev, "cmd 0x%02x reports "
2647                        "abort failed\n", c->Request.CDB[0]);
2648                 return_status = IO_ERROR;
2649                 break;
2650         case CMD_UNSOLICITED_ABORT:
2651                 dev_warn(&h->pdev->dev, "unsolicited abort 0x%02x\n",
2652                         c->Request.CDB[0]);
2653                 return_status = IO_NEEDS_RETRY;
2654                 break;
2655         default:
2656                 dev_warn(&h->pdev->dev, "cmd 0x%02x returned "
2657                        "unknown status %x\n", c->Request.CDB[0],
2658                        c->err_info->CommandStatus);
2659                 return_status = IO_ERROR;
2660         }
2661         return return_status;
2662 }
2663
2664 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2665         int attempt_retry)
2666 {
2667         DECLARE_COMPLETION_ONSTACK(wait);
2668         u64bit buff_dma_handle;
2669         int return_status = IO_OK;
2670
2671 resend_cmd2:
2672         c->waiting = &wait;
2673         enqueue_cmd_and_start_io(h, c);
2674
2675         wait_for_completion(&wait);
2676
2677         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2678                 goto command_done;
2679
2680         return_status = process_sendcmd_error(h, c);
2681
2682         if (return_status == IO_NEEDS_RETRY &&
2683                 c->retry_count < MAX_CMD_RETRIES) {
2684                 dev_warn(&h->pdev->dev, "retrying 0x%02x\n",
2685                         c->Request.CDB[0]);
2686                 c->retry_count++;
2687                 /* erase the old error information */
2688                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2689                 return_status = IO_OK;
2690                 INIT_COMPLETION(wait);
2691                 goto resend_cmd2;
2692         }
2693
2694 command_done:
2695         /* unlock the buffers from DMA */
2696         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2697         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2698         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2699                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2700         return return_status;
2701 }
2702
2703 static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
2704                            __u8 page_code, unsigned char scsi3addr[],
2705                         int cmd_type)
2706 {
2707         CommandList_struct *c;
2708         int return_status;
2709
2710         c = cmd_special_alloc(h);
2711         if (!c)
2712                 return -ENOMEM;
2713         return_status = fill_cmd(h, c, cmd, buff, size, page_code,
2714                 scsi3addr, cmd_type);
2715         if (return_status == IO_OK)
2716                 return_status = sendcmd_withirq_core(h, c, 1);
2717
2718         cmd_special_free(h, c);
2719         return return_status;
2720 }
2721
2722 static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
2723                                    sector_t total_size,
2724                                    unsigned int block_size,
2725                                    InquiryData_struct *inq_buff,
2726                                    drive_info_struct *drv)
2727 {
2728         int return_code;
2729         unsigned long t;
2730         unsigned char scsi3addr[8];
2731
2732         memset(inq_buff, 0, sizeof(InquiryData_struct));
2733         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2734         return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
2735                         sizeof(*inq_buff), 0xC1, scsi3addr, TYPE_CMD);
2736         if (return_code == IO_OK) {
2737                 if (inq_buff->data_byte[8] == 0xFF) {
2738                         dev_warn(&h->pdev->dev,
2739                                "reading geometry failed, volume "
2740                                "does not support reading geometry\n");
2741                         drv->heads = 255;
2742                         drv->sectors = 32;      /* Sectors per track */
2743                         drv->cylinders = total_size + 1;
2744                         drv->raid_level = RAID_UNKNOWN;
2745                 } else {
2746                         drv->heads = inq_buff->data_byte[6];
2747                         drv->sectors = inq_buff->data_byte[7];
2748                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2749                         drv->cylinders += inq_buff->data_byte[5];
2750                         drv->raid_level = inq_buff->data_byte[8];
2751                 }
2752                 drv->block_size = block_size;
2753                 drv->nr_blocks = total_size + 1;
2754                 t = drv->heads * drv->sectors;
2755                 if (t > 1) {
2756                         sector_t real_size = total_size + 1;
2757                         unsigned long rem = sector_div(real_size, t);
2758                         if (rem)
2759                                 real_size++;
2760                         drv->cylinders = real_size;
2761                 }
2762         } else {                /* Get geometry failed */
2763                 dev_warn(&h->pdev->dev, "reading geometry failed\n");
2764         }
2765 }
2766
2767 static void
2768 cciss_read_capacity(ctlr_info_t *h, int logvol, sector_t *total_size,
2769                     unsigned int *block_size)
2770 {
2771         ReadCapdata_struct *buf;
2772         int return_code;
2773         unsigned char scsi3addr[8];
2774
2775         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2776         if (!buf) {
2777                 dev_warn(&h->pdev->dev, "out of memory\n");
2778                 return;
2779         }
2780
2781         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2782         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY, buf,
2783                 sizeof(ReadCapdata_struct), 0, scsi3addr, TYPE_CMD);
2784         if (return_code == IO_OK) {
2785                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2786                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2787         } else {                /* read capacity command failed */
2788                 dev_warn(&h->pdev->dev, "read capacity failed\n");
2789                 *total_size = 0;
2790                 *block_size = BLOCK_SIZE;
2791         }
2792         kfree(buf);
2793 }
2794
2795 static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
2796         sector_t *total_size, unsigned int *block_size)
2797 {
2798         ReadCapdata_struct_16 *buf;
2799         int return_code;
2800         unsigned char scsi3addr[8];
2801
2802         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2803         if (!buf) {
2804                 dev_warn(&h->pdev->dev, "out of memory\n");
2805                 return;
2806         }
2807
2808         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2809         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY_16,
2810                 buf, sizeof(ReadCapdata_struct_16),
2811                         0, scsi3addr, TYPE_CMD);
2812         if (return_code == IO_OK) {
2813                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2814                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2815         } else {                /* read capacity command failed */
2816                 dev_warn(&h->pdev->dev, "read capacity failed\n");
2817                 *total_size = 0;
2818                 *block_size = BLOCK_SIZE;
2819         }
2820         dev_info(&h->pdev->dev, "      blocks= %llu block_size= %d\n",
2821                (unsigned long long)*total_size+1, *block_size);
2822         kfree(buf);
2823 }
2824
2825 static int cciss_revalidate(struct gendisk *disk)
2826 {
2827         ctlr_info_t *h = get_host(disk);
2828         drive_info_struct *drv = get_drv(disk);
2829         int logvol;
2830         int FOUND = 0;
2831         unsigned int block_size;
2832         sector_t total_size;
2833         InquiryData_struct *inq_buff = NULL;
2834
2835         for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
2836                 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2837                         sizeof(drv->LunID)) == 0) {
2838                         FOUND = 1;
2839                         break;
2840                 }
2841         }
2842
2843         if (!FOUND)
2844                 return 1;
2845
2846         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2847         if (inq_buff == NULL) {
2848                 dev_warn(&h->pdev->dev, "out of memory\n");
2849                 return 1;
2850         }
2851         if (h->cciss_read == CCISS_READ_10) {
2852                 cciss_read_capacity(h, logvol,
2853                                         &total_size, &block_size);
2854         } else {
2855                 cciss_read_capacity_16(h, logvol,
2856                                         &total_size, &block_size);
2857         }
2858         cciss_geometry_inquiry(h, logvol, total_size, block_size,
2859                                inq_buff, drv);
2860
2861         blk_queue_logical_block_size(drv->queue, drv->block_size);
2862         set_capacity(disk, drv->nr_blocks);
2863
2864         kfree(inq_buff);
2865         return 0;
2866 }
2867
2868 /*
2869  * Map (physical) PCI mem into (virtual) kernel space
2870  */
2871 static void __iomem *remap_pci_mem(ulong base, ulong size)
2872 {
2873         ulong page_base = ((ulong) base) & PAGE_MASK;
2874         ulong page_offs = ((ulong) base) - page_base;
2875         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2876
2877         return page_remapped ? (page_remapped + page_offs) : NULL;
2878 }
2879
2880 /*
2881  * Takes jobs of the Q and sends them to the hardware, then puts it on
2882  * the Q to wait for completion.
2883  */
2884 static void start_io(ctlr_info_t *h)
2885 {
2886         CommandList_struct *c;
2887
2888         while (!hlist_empty(&h->reqQ)) {
2889                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
2890                 /* can't do anything if fifo is full */
2891                 if ((h->access.fifo_full(h))) {
2892                         dev_warn(&h->pdev->dev, "fifo full\n");
2893                         break;
2894                 }
2895
2896                 /* Get the first entry from the Request Q */
2897                 removeQ(c);
2898                 h->Qdepth--;
2899
2900                 /* Tell the controller execute command */
2901                 h->access.submit_command(h, c);
2902
2903                 /* Put job onto the completed Q */
2904                 addQ(&h->cmpQ, c);
2905         }
2906 }
2907
2908 /* Assumes that h->lock is held. */
2909 /* Zeros out the error record and then resends the command back */
2910 /* to the controller */
2911 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2912 {
2913         /* erase the old error information */
2914         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2915
2916         /* add it to software queue and then send it to the controller */
2917         addQ(&h->reqQ, c);
2918         h->Qdepth++;
2919         if (h->Qdepth > h->maxQsinceinit)
2920                 h->maxQsinceinit = h->Qdepth;
2921
2922         start_io(h);
2923 }
2924
2925 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
2926         unsigned int msg_byte, unsigned int host_byte,
2927         unsigned int driver_byte)
2928 {
2929         /* inverse of macros in scsi.h */
2930         return (scsi_status_byte & 0xff) |
2931                 ((msg_byte & 0xff) << 8) |
2932                 ((host_byte & 0xff) << 16) |
2933                 ((driver_byte & 0xff) << 24);
2934 }
2935
2936 static inline int evaluate_target_status(ctlr_info_t *h,
2937                         CommandList_struct *cmd, int *retry_cmd)
2938 {
2939         unsigned char sense_key;
2940         unsigned char status_byte, msg_byte, host_byte, driver_byte;
2941         int error_value;
2942
2943         *retry_cmd = 0;
2944         /* If we get in here, it means we got "target status", that is, scsi status */
2945         status_byte = cmd->err_info->ScsiStatus;
2946         driver_byte = DRIVER_OK;
2947         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
2948
2949         if (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC)
2950                 host_byte = DID_PASSTHROUGH;
2951         else
2952                 host_byte = DID_OK;
2953
2954         error_value = make_status_bytes(status_byte, msg_byte,
2955                 host_byte, driver_byte);
2956
2957         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
2958                 if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC)
2959                         dev_warn(&h->pdev->dev, "cmd %p "
2960                                "has SCSI Status 0x%x\n",
2961                                cmd, cmd->err_info->ScsiStatus);
2962                 return error_value;
2963         }
2964
2965         /* check the sense key */
2966         sense_key = 0xf & cmd->err_info->SenseInfo[2];
2967         /* no status or recovered error */
2968         if (((sense_key == 0x0) || (sense_key == 0x1)) &&
2969             (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC))
2970                 error_value = 0;
2971
2972         if (check_for_unit_attention(h, cmd)) {
2973                 *retry_cmd = !(cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC);
2974                 return 0;
2975         }
2976
2977         /* Not SG_IO or similar? */
2978         if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC) {
2979                 if (error_value != 0)
2980                         dev_warn(&h->pdev->dev, "cmd %p has CHECK CONDITION"
2981                                " sense key = 0x%x\n", cmd, sense_key);
2982                 return error_value;
2983         }
2984
2985         /* SG_IO or similar, copy sense data back */
2986         if (cmd->rq->sense) {
2987                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
2988                         cmd->rq->sense_len = cmd->err_info->SenseLen;
2989                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
2990                         cmd->rq->sense_len);
2991         } else
2992                 cmd->rq->sense_len = 0;
2993
2994         return error_value;
2995 }
2996
2997 /* checks the status of the job and calls complete buffers to mark all
2998  * buffers for the completed job. Note that this function does not need
2999  * to hold the hba/queue lock.
3000  */
3001 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
3002                                     int timeout)
3003 {
3004         int retry_cmd = 0;
3005         struct request *rq = cmd->rq;
3006
3007         rq->errors = 0;
3008
3009         if (timeout)
3010                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
3011
3012         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
3013                 goto after_error_processing;
3014
3015         switch (cmd->err_info->CommandStatus) {
3016         case CMD_TARGET_STATUS:
3017                 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
3018                 break;
3019         case CMD_DATA_UNDERRUN:
3020                 if (cmd->rq->cmd_type == REQ_TYPE_FS) {
3021                         dev_warn(&h->pdev->dev, "cmd %p has"
3022                                " completed with data underrun "
3023                                "reported\n", cmd);
3024                         cmd->rq->resid_len = cmd->err_info->ResidualCnt;
3025                 }
3026                 break;
3027         case CMD_DATA_OVERRUN:
3028                 if (cmd->rq->cmd_type == REQ_TYPE_FS)
3029                         dev_warn(&h->pdev->dev, "cciss: cmd %p has"
3030                                " completed with data overrun "
3031                                "reported\n", cmd);
3032                 break;
3033         case CMD_INVALID:
3034                 dev_warn(&h->pdev->dev, "cciss: cmd %p is "
3035                        "reported invalid\n", cmd);
3036                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3037                         cmd->err_info->CommandStatus, DRIVER_OK,
3038                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3039                                 DID_PASSTHROUGH : DID_ERROR);
3040                 break;
3041         case CMD_PROTOCOL_ERR:
3042                 dev_warn(&h->pdev->dev, "cciss: cmd %p has "
3043                        "protocol error\n", cmd);
3044                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3045                         cmd->err_info->CommandStatus, DRIVER_OK,
3046                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3047                                 DID_PASSTHROUGH : DID_ERROR);
3048                 break;
3049         case CMD_HARDWARE_ERR:
3050                 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3051                        " hardware error\n", cmd);
3052                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3053                         cmd->err_info->CommandStatus, DRIVER_OK,
3054                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3055                                 DID_PASSTHROUGH : DID_ERROR);
3056                 break;
3057         case CMD_CONNECTION_LOST:
3058                 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3059                        "connection lost\n", cmd);
3060                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3061                         cmd->err_info->CommandStatus, DRIVER_OK,
3062                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3063                                 DID_PASSTHROUGH : DID_ERROR);
3064                 break;
3065         case CMD_ABORTED:
3066                 dev_warn(&h->pdev->dev, "cciss: cmd %p was "
3067                        "aborted\n", cmd);
3068                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3069                         cmd->err_info->CommandStatus, DRIVER_OK,
3070                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3071                                 DID_PASSTHROUGH : DID_ABORT);
3072                 break;
3073         case CMD_ABORT_FAILED:
3074                 dev_warn(&h->pdev->dev, "cciss: cmd %p reports "
3075                        "abort failed\n", cmd);
3076                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3077                         cmd->err_info->CommandStatus, DRIVER_OK,
3078                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3079                                 DID_PASSTHROUGH : DID_ERROR);
3080                 break;
3081         case CMD_UNSOLICITED_ABORT:
3082                 dev_warn(&h->pdev->dev, "cciss%d: unsolicited "
3083                        "abort %p\n", h->ctlr, cmd);
3084                 if (cmd->retry_count < MAX_CMD_RETRIES) {
3085                         retry_cmd = 1;
3086                         dev_warn(&h->pdev->dev, "retrying %p\n", cmd);
3087                         cmd->retry_count++;
3088                 } else
3089                         dev_warn(&h->pdev->dev,
3090                                 "%p retried too many times\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_ABORT);
3095                 break;
3096         case CMD_TIMEOUT:
3097                 dev_warn(&h->pdev->dev, "cmd %p timedout\n", cmd);
3098                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3099                         cmd->err_info->CommandStatus, DRIVER_OK,
3100                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3101                                 DID_PASSTHROUGH : DID_ERROR);
3102                 break;
3103         default:
3104                 dev_warn(&h->pdev->dev, "cmd %p returned "
3105                        "unknown status %x\n", cmd,
3106                        cmd->err_info->CommandStatus);
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_ERROR);
3111         }
3112
3113 after_error_processing:
3114
3115         /* We need to return this command */
3116         if (retry_cmd) {
3117                 resend_cciss_cmd(h, cmd);
3118                 return;
3119         }
3120         cmd->rq->completion_data = cmd;
3121         blk_complete_request(cmd->rq);
3122 }
3123
3124 static inline u32 cciss_tag_contains_index(u32 tag)
3125 {
3126 #define DIRECT_LOOKUP_BIT 0x10
3127         return tag & DIRECT_LOOKUP_BIT;
3128 }
3129
3130 static inline u32 cciss_tag_to_index(u32 tag)
3131 {
3132 #define DIRECT_LOOKUP_SHIFT 5
3133         return tag >> DIRECT_LOOKUP_SHIFT;
3134 }
3135
3136 static inline u32 cciss_tag_discard_error_bits(u32 tag)
3137 {
3138 #define CCISS_ERROR_BITS 0x03
3139         return tag & ~CCISS_ERROR_BITS;
3140 }
3141
3142 static inline void cciss_mark_tag_indexed(u32 *tag)
3143 {
3144         *tag |= DIRECT_LOOKUP_BIT;
3145 }
3146
3147 static inline void cciss_set_tag_index(u32 *tag, u32 index)
3148 {
3149         *tag |= (index << DIRECT_LOOKUP_SHIFT);
3150 }
3151
3152 /*
3153  * Get a request and submit it to the controller.
3154  */
3155 static void do_cciss_request(struct request_queue *q)
3156 {
3157         ctlr_info_t *h = q->queuedata;
3158         CommandList_struct *c;
3159         sector_t start_blk;
3160         int seg;
3161         struct request *creq;
3162         u64bit temp64;
3163         struct scatterlist *tmp_sg;
3164         SGDescriptor_struct *curr_sg;
3165         drive_info_struct *drv;
3166         int i, dir;
3167         int sg_index = 0;
3168         int chained = 0;
3169
3170         /* We call start_io here in case there is a command waiting on the
3171          * queue that has not been sent.
3172          */
3173         if (blk_queue_plugged(q))
3174                 goto startio;
3175
3176       queue:
3177         creq = blk_peek_request(q);
3178         if (!creq)
3179                 goto startio;
3180
3181         BUG_ON(creq->nr_phys_segments > h->maxsgentries);
3182
3183         c = cmd_alloc(h);
3184         if (!c)
3185                 goto full;
3186
3187         blk_start_request(creq);
3188
3189         tmp_sg = h->scatter_list[c->cmdindex];
3190         spin_unlock_irq(q->queue_lock);
3191
3192         c->cmd_type = CMD_RWREQ;
3193         c->rq = creq;
3194
3195         /* fill in the request */
3196         drv = creq->rq_disk->private_data;
3197         c->Header.ReplyQueue = 0;       /* unused in simple mode */
3198         /* got command from pool, so use the command block index instead */
3199         /* for direct lookups. */
3200         /* The first 2 bits are reserved for controller error reporting. */
3201         cciss_set_tag_index(&c->Header.Tag.lower, c->cmdindex);
3202         cciss_mark_tag_indexed(&c->Header.Tag.lower);
3203         memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3204         c->Request.CDBLen = 10; /* 12 byte commands not in FW yet; */
3205         c->Request.Type.Type = TYPE_CMD;        /* It is a command. */
3206         c->Request.Type.Attribute = ATTR_SIMPLE;
3207         c->Request.Type.Direction =
3208             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3209         c->Request.Timeout = 0; /* Don't time out */
3210         c->Request.CDB[0] =
3211             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3212         start_blk = blk_rq_pos(creq);
3213         dev_dbg(&h->pdev->dev, "sector =%d nr_sectors=%d\n",
3214                (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3215         sg_init_table(tmp_sg, h->maxsgentries);
3216         seg = blk_rq_map_sg(q, creq, tmp_sg);
3217
3218         /* get the DMA records for the setup */
3219         if (c->Request.Type.Direction == XFER_READ)
3220                 dir = PCI_DMA_FROMDEVICE;
3221         else
3222                 dir = PCI_DMA_TODEVICE;
3223
3224         curr_sg = c->SG;
3225         sg_index = 0;
3226         chained = 0;
3227
3228         for (i = 0; i < seg; i++) {
3229                 if (((sg_index+1) == (h->max_cmd_sgentries)) &&
3230                         !chained && ((seg - i) > 1)) {
3231                         /* Point to next chain block. */
3232                         curr_sg = h->cmd_sg_list[c->cmdindex];
3233                         sg_index = 0;
3234                         chained = 1;
3235                 }
3236                 curr_sg[sg_index].Len = tmp_sg[i].length;
3237                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3238                                                 tmp_sg[i].offset,
3239                                                 tmp_sg[i].length, dir);
3240                 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3241                 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3242                 curr_sg[sg_index].Ext = 0;  /* we are not chaining */
3243                 ++sg_index;
3244         }
3245         if (chained)
3246                 cciss_map_sg_chain_block(h, c, h->cmd_sg_list[c->cmdindex],
3247                         (seg - (h->max_cmd_sgentries - 1)) *
3248                                 sizeof(SGDescriptor_struct));
3249
3250         /* track how many SG entries we are using */
3251         if (seg > h->maxSG)
3252                 h->maxSG = seg;
3253
3254         dev_dbg(&h->pdev->dev, "Submitting %u sectors in %d segments "
3255                         "chained[%d]\n",
3256                         blk_rq_sectors(creq), seg, chained);
3257
3258         c->Header.SGTotal = seg + chained;
3259         if (seg <= h->max_cmd_sgentries)
3260                 c->Header.SGList = c->Header.SGTotal;
3261         else
3262                 c->Header.SGList = h->max_cmd_sgentries;
3263         set_performant_mode(h, c);
3264
3265         if (likely(creq->cmd_type == REQ_TYPE_FS)) {
3266                 if(h->cciss_read == CCISS_READ_10) {
3267                         c->Request.CDB[1] = 0;
3268                         c->Request.CDB[2] = (start_blk >> 24) & 0xff; /* MSB */
3269                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3270                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3271                         c->Request.CDB[5] = start_blk & 0xff;
3272                         c->Request.CDB[6] = 0; /* (sect >> 24) & 0xff; MSB */
3273                         c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3274                         c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3275                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3276                 } else {
3277                         u32 upper32 = upper_32_bits(start_blk);
3278
3279                         c->Request.CDBLen = 16;
3280                         c->Request.CDB[1]= 0;
3281                         c->Request.CDB[2]= (upper32 >> 24) & 0xff; /* MSB */
3282                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3283                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3284                         c->Request.CDB[5]= upper32 & 0xff;
3285                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3286                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3287                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3288                         c->Request.CDB[9]= start_blk & 0xff;
3289                         c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3290                         c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3291                         c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3292                         c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3293                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
3294                 }
3295         } else if (creq->cmd_type == REQ_TYPE_BLOCK_PC) {
3296                 c->Request.CDBLen = creq->cmd_len;
3297                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3298         } else {
3299                 dev_warn(&h->pdev->dev, "bad request type %d\n",
3300                         creq->cmd_type);
3301                 BUG();
3302         }
3303
3304         spin_lock_irq(q->queue_lock);
3305
3306         addQ(&h->reqQ, c);
3307         h->Qdepth++;
3308         if (h->Qdepth > h->maxQsinceinit)
3309                 h->maxQsinceinit = h->Qdepth;
3310
3311         goto queue;
3312 full:
3313         blk_stop_queue(q);
3314 startio:
3315         /* We will already have the driver lock here so not need
3316          * to lock it.
3317          */