IRQ: Maintain regs pointer globally rather than passing to IRQ handlers
[sfrench/cifs-2.6.git] / drivers / block / cciss.c
1 /*
2  *    Disk Array driver for HP SA 5xxx and 6xxx Controllers
3  *    Copyright 2000, 2006 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; either version 2 of the License, or
8  *    (at your option) any later version.
9  *
10  *    This program is distributed in the hope that it will be useful,
11  *    but WITHOUT ANY WARRANTY; without even the implied warranty of
12  *    MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
13  *    NON INFRINGEMENT.  See the GNU General Public License for more details.
14  *
15  *    You should have received a copy of the GNU General Public License
16  *    along with this program; if not, write to the Free Software
17  *    Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, 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/init.h>
37 #include <linux/hdreg.h>
38 #include <linux/spinlock.h>
39 #include <linux/compat.h>
40 #include <linux/blktrace_api.h>
41 #include <asm/uaccess.h>
42 #include <asm/io.h>
43
44 #include <linux/dma-mapping.h>
45 #include <linux/blkdev.h>
46 #include <linux/genhd.h>
47 #include <linux/completion.h>
48
49 #define CCISS_DRIVER_VERSION(maj,min,submin) ((maj<<16)|(min<<8)|(submin))
50 #define DRIVER_NAME "HP CISS Driver (v 3.6.10)"
51 #define DRIVER_VERSION CCISS_DRIVER_VERSION(3,6,10)
52
53 /* Embedded module documentation macros - see modules.h */
54 MODULE_AUTHOR("Hewlett-Packard Company");
55 MODULE_DESCRIPTION("Driver for HP Controller SA5xxx SA6xxx version 3.6.10");
56 MODULE_SUPPORTED_DEVICE("HP SA5i SA5i+ SA532 SA5300 SA5312 SA641 SA642 SA6400"
57                         " SA6i P600 P800 P400 P400i E200 E200i E500");
58 MODULE_LICENSE("GPL");
59
60 #include "cciss_cmd.h"
61 #include "cciss.h"
62 #include <linux/cciss_ioctl.h>
63
64 /* define the PCI info for the cards we can control */
65 static const struct pci_device_id cciss_pci_device_id[] = {
66         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISS,  0x0E11, 0x4070},
67         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4080},
68         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4082},
69         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4083},
70         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x4091},
71         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409A},
72         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409B},
73         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409C},
74         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409D},
75         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSA,     0x103C, 0x3225},
76         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3223},
77         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3234},
78         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3235},
79         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3211},
80         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3212},
81         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3213},
82         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3214},
83         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3215},
84         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3233},
85         {0,}
86 };
87
88 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
89
90 /*  board_id = Subsystem Device ID & Vendor ID
91  *  product = Marketing Name for the board
92  *  access = Address of the struct of function pointers
93  */
94 static struct board_type products[] = {
95         {0x40700E11, "Smart Array 5300", &SA5_access},
96         {0x40800E11, "Smart Array 5i", &SA5B_access},
97         {0x40820E11, "Smart Array 532", &SA5B_access},
98         {0x40830E11, "Smart Array 5312", &SA5B_access},
99         {0x409A0E11, "Smart Array 641", &SA5_access},
100         {0x409B0E11, "Smart Array 642", &SA5_access},
101         {0x409C0E11, "Smart Array 6400", &SA5_access},
102         {0x409D0E11, "Smart Array 6400 EM", &SA5_access},
103         {0x40910E11, "Smart Array 6i", &SA5_access},
104         {0x3225103C, "Smart Array P600", &SA5_access},
105         {0x3223103C, "Smart Array P800", &SA5_access},
106         {0x3234103C, "Smart Array P400", &SA5_access},
107         {0x3235103C, "Smart Array P400i", &SA5_access},
108         {0x3211103C, "Smart Array E200i", &SA5_access},
109         {0x3212103C, "Smart Array E200", &SA5_access},
110         {0x3213103C, "Smart Array E200i", &SA5_access},
111         {0x3214103C, "Smart Array E200i", &SA5_access},
112         {0x3215103C, "Smart Array E200i", &SA5_access},
113         {0x3233103C, "Smart Array E500", &SA5_access},
114 };
115
116 /* How long to wait (in milliseconds) for board to go into simple mode */
117 #define MAX_CONFIG_WAIT 30000
118 #define MAX_IOCTL_CONFIG_WAIT 1000
119
120 /*define how many times we will try a command because of bus resets */
121 #define MAX_CMD_RETRIES 3
122
123 #define READ_AHEAD       1024
124 #define NR_CMDS          384    /* #commands that can be outstanding */
125 #define MAX_CTLR        32
126
127 /* Originally cciss driver only supports 8 major numbers */
128 #define MAX_CTLR_ORIG   8
129
130 static ctlr_info_t *hba[MAX_CTLR];
131
132 static void do_cciss_request(request_queue_t *q);
133 static irqreturn_t do_cciss_intr(int irq, void *dev_id);
134 static int cciss_open(struct inode *inode, struct file *filep);
135 static int cciss_release(struct inode *inode, struct file *filep);
136 static int cciss_ioctl(struct inode *inode, struct file *filep,
137                        unsigned int cmd, unsigned long arg);
138 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
139
140 static int revalidate_allvol(ctlr_info_t *host);
141 static int cciss_revalidate(struct gendisk *disk);
142 static int rebuild_lun_table(ctlr_info_t *h, struct gendisk *del_disk);
143 static int deregister_disk(struct gendisk *disk, drive_info_struct *drv,
144                            int clear_all);
145
146 static void cciss_read_capacity(int ctlr, int logvol, int withirq,
147                         sector_t *total_size, unsigned int *block_size);
148 static void cciss_read_capacity_16(int ctlr, int logvol, int withirq,
149                         sector_t *total_size, unsigned int *block_size);
150 static void cciss_geometry_inquiry(int ctlr, int logvol,
151                         int withirq, sector_t total_size,
152                         unsigned int block_size, InquiryData_struct *inq_buff,
153                                    drive_info_struct *drv);
154 static void cciss_getgeometry(int cntl_num);
155 static void __devinit cciss_interrupt_mode(ctlr_info_t *, struct pci_dev *,
156                                            __u32);
157 static void start_io(ctlr_info_t *h);
158 static int sendcmd(__u8 cmd, int ctlr, void *buff, size_t size,
159                    unsigned int use_unit_num, unsigned int log_unit,
160                    __u8 page_code, unsigned char *scsi3addr, int cmd_type);
161 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
162                            unsigned int use_unit_num, unsigned int log_unit,
163                            __u8 page_code, int cmd_type);
164
165 static void fail_all_cmds(unsigned long ctlr);
166
167 #ifdef CONFIG_PROC_FS
168 static int cciss_proc_get_info(char *buffer, char **start, off_t offset,
169                                int length, int *eof, void *data);
170 static void cciss_procinit(int i);
171 #else
172 static void cciss_procinit(int i)
173 {
174 }
175 #endif                          /* CONFIG_PROC_FS */
176
177 #ifdef CONFIG_COMPAT
178 static long cciss_compat_ioctl(struct file *f, unsigned cmd, unsigned long arg);
179 #endif
180
181 static struct block_device_operations cciss_fops = {
182         .owner = THIS_MODULE,
183         .open = cciss_open,
184         .release = cciss_release,
185         .ioctl = cciss_ioctl,
186         .getgeo = cciss_getgeo,
187 #ifdef CONFIG_COMPAT
188         .compat_ioctl = cciss_compat_ioctl,
189 #endif
190         .revalidate_disk = cciss_revalidate,
191 };
192
193 /*
194  * Enqueuing and dequeuing functions for cmdlists.
195  */
196 static inline void addQ(CommandList_struct **Qptr, CommandList_struct *c)
197 {
198         if (*Qptr == NULL) {
199                 *Qptr = c;
200                 c->next = c->prev = c;
201         } else {
202                 c->prev = (*Qptr)->prev;
203                 c->next = (*Qptr);
204                 (*Qptr)->prev->next = c;
205                 (*Qptr)->prev = c;
206         }
207 }
208
209 static inline CommandList_struct *removeQ(CommandList_struct **Qptr,
210                                           CommandList_struct *c)
211 {
212         if (c && c->next != c) {
213                 if (*Qptr == c)
214                         *Qptr = c->next;
215                 c->prev->next = c->next;
216                 c->next->prev = c->prev;
217         } else {
218                 *Qptr = NULL;
219         }
220         return c;
221 }
222
223 #include "cciss_scsi.c"         /* For SCSI tape support */
224
225 #ifdef CONFIG_PROC_FS
226
227 /*
228  * Report information about this controller.
229  */
230 #define ENG_GIG 1000000000
231 #define ENG_GIG_FACTOR (ENG_GIG/512)
232 #define RAID_UNKNOWN 6
233 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
234         "UNKNOWN"
235 };
236
237 static struct proc_dir_entry *proc_cciss;
238
239 static int cciss_proc_get_info(char *buffer, char **start, off_t offset,
240                                int length, int *eof, void *data)
241 {
242         off_t pos = 0;
243         off_t len = 0;
244         int size, i, ctlr;
245         ctlr_info_t *h = (ctlr_info_t *) data;
246         drive_info_struct *drv;
247         unsigned long flags;
248         sector_t vol_sz, vol_sz_frac;
249
250         ctlr = h->ctlr;
251
252         /* prevent displaying bogus info during configuration
253          * or deconfiguration of a logical volume
254          */
255         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
256         if (h->busy_configuring) {
257                 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
258                 return -EBUSY;
259         }
260         h->busy_configuring = 1;
261         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
262
263         size = sprintf(buffer, "%s: HP %s Controller\n"
264                        "Board ID: 0x%08lx\n"
265                        "Firmware Version: %c%c%c%c\n"
266                        "IRQ: %d\n"
267                        "Logical drives: %d\n"
268                        "Current Q depth: %d\n"
269                        "Current # commands on controller: %d\n"
270                        "Max Q depth since init: %d\n"
271                        "Max # commands on controller since init: %d\n"
272                        "Max SG entries since init: %d\n\n",
273                        h->devname,
274                        h->product_name,
275                        (unsigned long)h->board_id,
276                        h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
277                        h->firm_ver[3], (unsigned int)h->intr[SIMPLE_MODE_INT],
278                        h->num_luns, h->Qdepth, h->commands_outstanding,
279                        h->maxQsinceinit, h->max_outstanding, h->maxSG);
280
281         pos += size;
282         len += size;
283         cciss_proc_tape_report(ctlr, buffer, &pos, &len);
284         for (i = 0; i <= h->highest_lun; i++) {
285
286                 drv = &h->drv[i];
287                 if (drv->heads == 0)
288                         continue;
289
290                 vol_sz = drv->nr_blocks;
291                 vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
292                 vol_sz_frac *= 100;
293                 sector_div(vol_sz_frac, ENG_GIG_FACTOR);
294
295                 if (drv->raid_level > 5)
296                         drv->raid_level = RAID_UNKNOWN;
297                 size = sprintf(buffer + len, "cciss/c%dd%d:"
298                                "\t%4u.%02uGB\tRAID %s\n",
299                                ctlr, i, (int)vol_sz, (int)vol_sz_frac,
300                                raid_label[drv->raid_level]);
301                 pos += size;
302                 len += size;
303         }
304
305         *eof = 1;
306         *start = buffer + offset;
307         len -= offset;
308         if (len > length)
309                 len = length;
310         h->busy_configuring = 0;
311         return len;
312 }
313
314 static int
315 cciss_proc_write(struct file *file, const char __user *buffer,
316                  unsigned long count, void *data)
317 {
318         unsigned char cmd[80];
319         int len;
320 #ifdef CONFIG_CISS_SCSI_TAPE
321         ctlr_info_t *h = (ctlr_info_t *) data;
322         int rc;
323 #endif
324
325         if (count > sizeof(cmd) - 1)
326                 return -EINVAL;
327         if (copy_from_user(cmd, buffer, count))
328                 return -EFAULT;
329         cmd[count] = '\0';
330         len = strlen(cmd);      // above 3 lines ensure safety
331         if (len && cmd[len - 1] == '\n')
332                 cmd[--len] = '\0';
333 #       ifdef CONFIG_CISS_SCSI_TAPE
334         if (strcmp("engage scsi", cmd) == 0) {
335                 rc = cciss_engage_scsi(h->ctlr);
336                 if (rc != 0)
337                         return -rc;
338                 return count;
339         }
340         /* might be nice to have "disengage" too, but it's not
341            safely possible. (only 1 module use count, lock issues.) */
342 #       endif
343         return -EINVAL;
344 }
345
346 /*
347  * Get us a file in /proc/cciss that says something about each controller.
348  * Create /proc/cciss if it doesn't exist yet.
349  */
350 static void __devinit cciss_procinit(int i)
351 {
352         struct proc_dir_entry *pde;
353
354         if (proc_cciss == NULL) {
355                 proc_cciss = proc_mkdir("cciss", proc_root_driver);
356                 if (!proc_cciss)
357                         return;
358         }
359
360         pde = create_proc_read_entry(hba[i]->devname,
361                                      S_IWUSR | S_IRUSR | S_IRGRP | S_IROTH,
362                                      proc_cciss, cciss_proc_get_info, hba[i]);
363         pde->write_proc = cciss_proc_write;
364 }
365 #endif                          /* CONFIG_PROC_FS */
366
367 /*
368  * For operations that cannot sleep, a command block is allocated at init,
369  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
370  * which ones are free or in use.  For operations that can wait for kmalloc
371  * to possible sleep, this routine can be called with get_from_pool set to 0.
372  * cmd_free() MUST be called with a got_from_pool set to 0 if cmd_alloc was.
373  */
374 static CommandList_struct *cmd_alloc(ctlr_info_t *h, int get_from_pool)
375 {
376         CommandList_struct *c;
377         int i;
378         u64bit temp64;
379         dma_addr_t cmd_dma_handle, err_dma_handle;
380
381         if (!get_from_pool) {
382                 c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
383                         sizeof(CommandList_struct), &cmd_dma_handle);
384                 if (c == NULL)
385                         return NULL;
386                 memset(c, 0, sizeof(CommandList_struct));
387
388                 c->cmdindex = -1;
389
390                 c->err_info = (ErrorInfo_struct *)
391                     pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
392                             &err_dma_handle);
393
394                 if (c->err_info == NULL) {
395                         pci_free_consistent(h->pdev,
396                                 sizeof(CommandList_struct), c, cmd_dma_handle);
397                         return NULL;
398                 }
399                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
400         } else {                /* get it out of the controllers pool */
401
402                 do {
403                         i = find_first_zero_bit(h->cmd_pool_bits, NR_CMDS);
404                         if (i == NR_CMDS)
405                                 return NULL;
406                 } while (test_and_set_bit
407                          (i & (BITS_PER_LONG - 1),
408                           h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
409 #ifdef CCISS_DEBUG
410                 printk(KERN_DEBUG "cciss: using command buffer %d\n", i);
411 #endif
412                 c = h->cmd_pool + i;
413                 memset(c, 0, sizeof(CommandList_struct));
414                 cmd_dma_handle = h->cmd_pool_dhandle
415                     + i * sizeof(CommandList_struct);
416                 c->err_info = h->errinfo_pool + i;
417                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
418                 err_dma_handle = h->errinfo_pool_dhandle
419                     + i * sizeof(ErrorInfo_struct);
420                 h->nr_allocs++;
421
422                 c->cmdindex = i;
423         }
424
425         c->busaddr = (__u32) cmd_dma_handle;
426         temp64.val = (__u64) err_dma_handle;
427         c->ErrDesc.Addr.lower = temp64.val32.lower;
428         c->ErrDesc.Addr.upper = temp64.val32.upper;
429         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
430
431         c->ctlr = h->ctlr;
432         return c;
433 }
434
435 /*
436  * Frees a command block that was previously allocated with cmd_alloc().
437  */
438 static void cmd_free(ctlr_info_t *h, CommandList_struct *c, int got_from_pool)
439 {
440         int i;
441         u64bit temp64;
442
443         if (!got_from_pool) {
444                 temp64.val32.lower = c->ErrDesc.Addr.lower;
445                 temp64.val32.upper = c->ErrDesc.Addr.upper;
446                 pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
447                                     c->err_info, (dma_addr_t) temp64.val);
448                 pci_free_consistent(h->pdev, sizeof(CommandList_struct),
449                                     c, (dma_addr_t) c->busaddr);
450         } else {
451                 i = c - h->cmd_pool;
452                 clear_bit(i & (BITS_PER_LONG - 1),
453                           h->cmd_pool_bits + (i / BITS_PER_LONG));
454                 h->nr_frees++;
455         }
456 }
457
458 static inline ctlr_info_t *get_host(struct gendisk *disk)
459 {
460         return disk->queue->queuedata;
461 }
462
463 static inline drive_info_struct *get_drv(struct gendisk *disk)
464 {
465         return disk->private_data;
466 }
467
468 /*
469  * Open.  Make sure the device is really there.
470  */
471 static int cciss_open(struct inode *inode, struct file *filep)
472 {
473         ctlr_info_t *host = get_host(inode->i_bdev->bd_disk);
474         drive_info_struct *drv = get_drv(inode->i_bdev->bd_disk);
475
476 #ifdef CCISS_DEBUG
477         printk(KERN_DEBUG "cciss_open %s\n", inode->i_bdev->bd_disk->disk_name);
478 #endif                          /* CCISS_DEBUG */
479
480         if (host->busy_initializing || drv->busy_configuring)
481                 return -EBUSY;
482         /*
483          * Root is allowed to open raw volume zero even if it's not configured
484          * so array config can still work. Root is also allowed to open any
485          * volume that has a LUN ID, so it can issue IOCTL to reread the
486          * disk information.  I don't think I really like this
487          * but I'm already using way to many device nodes to claim another one
488          * for "raw controller".
489          */
490         if (drv->nr_blocks == 0) {
491                 if (iminor(inode) != 0) {       /* not node 0? */
492                         /* if not node 0 make sure it is a partition = 0 */
493                         if (iminor(inode) & 0x0f) {
494                                 return -ENXIO;
495                                 /* if it is, make sure we have a LUN ID */
496                         } else if (drv->LunID == 0) {
497                                 return -ENXIO;
498                         }
499                 }
500                 if (!capable(CAP_SYS_ADMIN))
501                         return -EPERM;
502         }
503         drv->usage_count++;
504         host->usage_count++;
505         return 0;
506 }
507
508 /*
509  * Close.  Sync first.
510  */
511 static int cciss_release(struct inode *inode, struct file *filep)
512 {
513         ctlr_info_t *host = get_host(inode->i_bdev->bd_disk);
514         drive_info_struct *drv = get_drv(inode->i_bdev->bd_disk);
515
516 #ifdef CCISS_DEBUG
517         printk(KERN_DEBUG "cciss_release %s\n",
518                inode->i_bdev->bd_disk->disk_name);
519 #endif                          /* CCISS_DEBUG */
520
521         drv->usage_count--;
522         host->usage_count--;
523         return 0;
524 }
525
526 #ifdef CONFIG_COMPAT
527
528 static int do_ioctl(struct file *f, unsigned cmd, unsigned long arg)
529 {
530         int ret;
531         lock_kernel();
532         ret = cciss_ioctl(f->f_dentry->d_inode, f, cmd, arg);
533         unlock_kernel();
534         return ret;
535 }
536
537 static int cciss_ioctl32_passthru(struct file *f, unsigned cmd,
538                                   unsigned long arg);
539 static int cciss_ioctl32_big_passthru(struct file *f, unsigned cmd,
540                                       unsigned long arg);
541
542 static long cciss_compat_ioctl(struct file *f, unsigned cmd, unsigned long arg)
543 {
544         switch (cmd) {
545         case CCISS_GETPCIINFO:
546         case CCISS_GETINTINFO:
547         case CCISS_SETINTINFO:
548         case CCISS_GETNODENAME:
549         case CCISS_SETNODENAME:
550         case CCISS_GETHEARTBEAT:
551         case CCISS_GETBUSTYPES:
552         case CCISS_GETFIRMVER:
553         case CCISS_GETDRIVVER:
554         case CCISS_REVALIDVOLS:
555         case CCISS_DEREGDISK:
556         case CCISS_REGNEWDISK:
557         case CCISS_REGNEWD:
558         case CCISS_RESCANDISK:
559         case CCISS_GETLUNINFO:
560                 return do_ioctl(f, cmd, arg);
561
562         case CCISS_PASSTHRU32:
563                 return cciss_ioctl32_passthru(f, cmd, arg);
564         case CCISS_BIG_PASSTHRU32:
565                 return cciss_ioctl32_big_passthru(f, cmd, arg);
566
567         default:
568                 return -ENOIOCTLCMD;
569         }
570 }
571
572 static int cciss_ioctl32_passthru(struct file *f, unsigned cmd,
573                                   unsigned long arg)
574 {
575         IOCTL32_Command_struct __user *arg32 =
576             (IOCTL32_Command_struct __user *) arg;
577         IOCTL_Command_struct arg64;
578         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
579         int err;
580         u32 cp;
581
582         err = 0;
583         err |=
584             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
585                            sizeof(arg64.LUN_info));
586         err |=
587             copy_from_user(&arg64.Request, &arg32->Request,
588                            sizeof(arg64.Request));
589         err |=
590             copy_from_user(&arg64.error_info, &arg32->error_info,
591                            sizeof(arg64.error_info));
592         err |= get_user(arg64.buf_size, &arg32->buf_size);
593         err |= get_user(cp, &arg32->buf);
594         arg64.buf = compat_ptr(cp);
595         err |= copy_to_user(p, &arg64, sizeof(arg64));
596
597         if (err)
598                 return -EFAULT;
599
600         err = do_ioctl(f, CCISS_PASSTHRU, (unsigned long)p);
601         if (err)
602                 return err;
603         err |=
604             copy_in_user(&arg32->error_info, &p->error_info,
605                          sizeof(arg32->error_info));
606         if (err)
607                 return -EFAULT;
608         return err;
609 }
610
611 static int cciss_ioctl32_big_passthru(struct file *file, unsigned cmd,
612                                       unsigned long arg)
613 {
614         BIG_IOCTL32_Command_struct __user *arg32 =
615             (BIG_IOCTL32_Command_struct __user *) arg;
616         BIG_IOCTL_Command_struct arg64;
617         BIG_IOCTL_Command_struct __user *p =
618             compat_alloc_user_space(sizeof(arg64));
619         int err;
620         u32 cp;
621
622         err = 0;
623         err |=
624             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
625                            sizeof(arg64.LUN_info));
626         err |=
627             copy_from_user(&arg64.Request, &arg32->Request,
628                            sizeof(arg64.Request));
629         err |=
630             copy_from_user(&arg64.error_info, &arg32->error_info,
631                            sizeof(arg64.error_info));
632         err |= get_user(arg64.buf_size, &arg32->buf_size);
633         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
634         err |= get_user(cp, &arg32->buf);
635         arg64.buf = compat_ptr(cp);
636         err |= copy_to_user(p, &arg64, sizeof(arg64));
637
638         if (err)
639                 return -EFAULT;
640
641         err = do_ioctl(file, CCISS_BIG_PASSTHRU, (unsigned long)p);
642         if (err)
643                 return err;
644         err |=
645             copy_in_user(&arg32->error_info, &p->error_info,
646                          sizeof(arg32->error_info));
647         if (err)
648                 return -EFAULT;
649         return err;
650 }
651 #endif
652
653 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
654 {
655         drive_info_struct *drv = get_drv(bdev->bd_disk);
656
657         if (!drv->cylinders)
658                 return -ENXIO;
659
660         geo->heads = drv->heads;
661         geo->sectors = drv->sectors;
662         geo->cylinders = drv->cylinders;
663         return 0;
664 }
665
666 /*
667  * ioctl
668  */
669 static int cciss_ioctl(struct inode *inode, struct file *filep,
670                        unsigned int cmd, unsigned long arg)
671 {
672         struct block_device *bdev = inode->i_bdev;
673         struct gendisk *disk = bdev->bd_disk;
674         ctlr_info_t *host = get_host(disk);
675         drive_info_struct *drv = get_drv(disk);
676         int ctlr = host->ctlr;
677         void __user *argp = (void __user *)arg;
678
679 #ifdef CCISS_DEBUG
680         printk(KERN_DEBUG "cciss_ioctl: Called with cmd=%x %lx\n", cmd, arg);
681 #endif                          /* CCISS_DEBUG */
682
683         switch (cmd) {
684         case CCISS_GETPCIINFO:
685                 {
686                         cciss_pci_info_struct pciinfo;
687
688                         if (!arg)
689                                 return -EINVAL;
690                         pciinfo.domain = pci_domain_nr(host->pdev->bus);
691                         pciinfo.bus = host->pdev->bus->number;
692                         pciinfo.dev_fn = host->pdev->devfn;
693                         pciinfo.board_id = host->board_id;
694                         if (copy_to_user
695                             (argp, &pciinfo, sizeof(cciss_pci_info_struct)))
696                                 return -EFAULT;
697                         return 0;
698                 }
699         case CCISS_GETINTINFO:
700                 {
701                         cciss_coalint_struct intinfo;
702                         if (!arg)
703                                 return -EINVAL;
704                         intinfo.delay =
705                             readl(&host->cfgtable->HostWrite.CoalIntDelay);
706                         intinfo.count =
707                             readl(&host->cfgtable->HostWrite.CoalIntCount);
708                         if (copy_to_user
709                             (argp, &intinfo, sizeof(cciss_coalint_struct)))
710                                 return -EFAULT;
711                         return 0;
712                 }
713         case CCISS_SETINTINFO:
714                 {
715                         cciss_coalint_struct intinfo;
716                         unsigned long flags;
717                         int i;
718
719                         if (!arg)
720                                 return -EINVAL;
721                         if (!capable(CAP_SYS_ADMIN))
722                                 return -EPERM;
723                         if (copy_from_user
724                             (&intinfo, argp, sizeof(cciss_coalint_struct)))
725                                 return -EFAULT;
726                         if ((intinfo.delay == 0) && (intinfo.count == 0))
727                         {
728 //                      printk("cciss_ioctl: delay and count cannot be 0\n");
729                                 return -EINVAL;
730                         }
731                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
732                         /* Update the field, and then ring the doorbell */
733                         writel(intinfo.delay,
734                                &(host->cfgtable->HostWrite.CoalIntDelay));
735                         writel(intinfo.count,
736                                &(host->cfgtable->HostWrite.CoalIntCount));
737                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
738
739                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
740                                 if (!(readl(host->vaddr + SA5_DOORBELL)
741                                       & CFGTBL_ChangeReq))
742                                         break;
743                                 /* delay and try again */
744                                 udelay(1000);
745                         }
746                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
747                         if (i >= MAX_IOCTL_CONFIG_WAIT)
748                                 return -EAGAIN;
749                         return 0;
750                 }
751         case CCISS_GETNODENAME:
752                 {
753                         NodeName_type NodeName;
754                         int i;
755
756                         if (!arg)
757                                 return -EINVAL;
758                         for (i = 0; i < 16; i++)
759                                 NodeName[i] =
760                                     readb(&host->cfgtable->ServerName[i]);
761                         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
762                                 return -EFAULT;
763                         return 0;
764                 }
765         case CCISS_SETNODENAME:
766                 {
767                         NodeName_type NodeName;
768                         unsigned long flags;
769                         int i;
770
771                         if (!arg)
772                                 return -EINVAL;
773                         if (!capable(CAP_SYS_ADMIN))
774                                 return -EPERM;
775
776                         if (copy_from_user
777                             (NodeName, argp, sizeof(NodeName_type)))
778                                 return -EFAULT;
779
780                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
781
782                         /* Update the field, and then ring the doorbell */
783                         for (i = 0; i < 16; i++)
784                                 writeb(NodeName[i],
785                                        &host->cfgtable->ServerName[i]);
786
787                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
788
789                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
790                                 if (!(readl(host->vaddr + SA5_DOORBELL)
791                                       & CFGTBL_ChangeReq))
792                                         break;
793                                 /* delay and try again */
794                                 udelay(1000);
795                         }
796                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
797                         if (i >= MAX_IOCTL_CONFIG_WAIT)
798                                 return -EAGAIN;
799                         return 0;
800                 }
801
802         case CCISS_GETHEARTBEAT:
803                 {
804                         Heartbeat_type heartbeat;
805
806                         if (!arg)
807                                 return -EINVAL;
808                         heartbeat = readl(&host->cfgtable->HeartBeat);
809                         if (copy_to_user
810                             (argp, &heartbeat, sizeof(Heartbeat_type)))
811                                 return -EFAULT;
812                         return 0;
813                 }
814         case CCISS_GETBUSTYPES:
815                 {
816                         BusTypes_type BusTypes;
817
818                         if (!arg)
819                                 return -EINVAL;
820                         BusTypes = readl(&host->cfgtable->BusTypes);
821                         if (copy_to_user
822                             (argp, &BusTypes, sizeof(BusTypes_type)))
823                                 return -EFAULT;
824                         return 0;
825                 }
826         case CCISS_GETFIRMVER:
827                 {
828                         FirmwareVer_type firmware;
829
830                         if (!arg)
831                                 return -EINVAL;
832                         memcpy(firmware, host->firm_ver, 4);
833
834                         if (copy_to_user
835                             (argp, firmware, sizeof(FirmwareVer_type)))
836                                 return -EFAULT;
837                         return 0;
838                 }
839         case CCISS_GETDRIVVER:
840                 {
841                         DriverVer_type DriverVer = DRIVER_VERSION;
842
843                         if (!arg)
844                                 return -EINVAL;
845
846                         if (copy_to_user
847                             (argp, &DriverVer, sizeof(DriverVer_type)))
848                                 return -EFAULT;
849                         return 0;
850                 }
851
852         case CCISS_REVALIDVOLS:
853                 if (bdev != bdev->bd_contains || drv != host->drv)
854                         return -ENXIO;
855                 return revalidate_allvol(host);
856
857         case CCISS_GETLUNINFO:{
858                         LogvolInfo_struct luninfo;
859
860                         luninfo.LunID = drv->LunID;
861                         luninfo.num_opens = drv->usage_count;
862                         luninfo.num_parts = 0;
863                         if (copy_to_user(argp, &luninfo,
864                                          sizeof(LogvolInfo_struct)))
865                                 return -EFAULT;
866                         return 0;
867                 }
868         case CCISS_DEREGDISK:
869                 return rebuild_lun_table(host, disk);
870
871         case CCISS_REGNEWD:
872                 return rebuild_lun_table(host, NULL);
873
874         case CCISS_PASSTHRU:
875                 {
876                         IOCTL_Command_struct iocommand;
877                         CommandList_struct *c;
878                         char *buff = NULL;
879                         u64bit temp64;
880                         unsigned long flags;
881                         DECLARE_COMPLETION_ONSTACK(wait);
882
883                         if (!arg)
884                                 return -EINVAL;
885
886                         if (!capable(CAP_SYS_RAWIO))
887                                 return -EPERM;
888
889                         if (copy_from_user
890                             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
891                                 return -EFAULT;
892                         if ((iocommand.buf_size < 1) &&
893                             (iocommand.Request.Type.Direction != XFER_NONE)) {
894                                 return -EINVAL;
895                         }
896 #if 0                           /* 'buf_size' member is 16-bits, and always smaller than kmalloc limit */
897                         /* Check kmalloc limits */
898                         if (iocommand.buf_size > 128000)
899                                 return -EINVAL;
900 #endif
901                         if (iocommand.buf_size > 0) {
902                                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
903                                 if (buff == NULL)
904                                         return -EFAULT;
905                         }
906                         if (iocommand.Request.Type.Direction == XFER_WRITE) {
907                                 /* Copy the data into the buffer we created */
908                                 if (copy_from_user
909                                     (buff, iocommand.buf, iocommand.buf_size)) {
910                                         kfree(buff);
911                                         return -EFAULT;
912                                 }
913                         } else {
914                                 memset(buff, 0, iocommand.buf_size);
915                         }
916                         if ((c = cmd_alloc(host, 0)) == NULL) {
917                                 kfree(buff);
918                                 return -ENOMEM;
919                         }
920                         // Fill in the command type
921                         c->cmd_type = CMD_IOCTL_PEND;
922                         // Fill in Command Header
923                         c->Header.ReplyQueue = 0;       // unused in simple mode
924                         if (iocommand.buf_size > 0)     // buffer to fill
925                         {
926                                 c->Header.SGList = 1;
927                                 c->Header.SGTotal = 1;
928                         } else  // no buffers to fill
929                         {
930                                 c->Header.SGList = 0;
931                                 c->Header.SGTotal = 0;
932                         }
933                         c->Header.LUN = iocommand.LUN_info;
934                         c->Header.Tag.lower = c->busaddr;       // use the kernel address the cmd block for tag
935
936                         // Fill in Request block
937                         c->Request = iocommand.Request;
938
939                         // Fill in the scatter gather information
940                         if (iocommand.buf_size > 0) {
941                                 temp64.val = pci_map_single(host->pdev, buff,
942                                         iocommand.buf_size,
943                                         PCI_DMA_BIDIRECTIONAL);
944                                 c->SG[0].Addr.lower = temp64.val32.lower;
945                                 c->SG[0].Addr.upper = temp64.val32.upper;
946                                 c->SG[0].Len = iocommand.buf_size;
947                                 c->SG[0].Ext = 0;       // we are not chaining
948                         }
949                         c->waiting = &wait;
950
951                         /* Put the request on the tail of the request queue */
952                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
953                         addQ(&host->reqQ, c);
954                         host->Qdepth++;
955                         start_io(host);
956                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
957
958                         wait_for_completion(&wait);
959
960                         /* unlock the buffers from DMA */
961                         temp64.val32.lower = c->SG[0].Addr.lower;
962                         temp64.val32.upper = c->SG[0].Addr.upper;
963                         pci_unmap_single(host->pdev, (dma_addr_t) temp64.val,
964                                          iocommand.buf_size,
965                                          PCI_DMA_BIDIRECTIONAL);
966
967                         /* Copy the error information out */
968                         iocommand.error_info = *(c->err_info);
969                         if (copy_to_user
970                             (argp, &iocommand, sizeof(IOCTL_Command_struct))) {
971                                 kfree(buff);
972                                 cmd_free(host, c, 0);
973                                 return -EFAULT;
974                         }
975
976                         if (iocommand.Request.Type.Direction == XFER_READ) {
977                                 /* Copy the data out of the buffer we created */
978                                 if (copy_to_user
979                                     (iocommand.buf, buff, iocommand.buf_size)) {
980                                         kfree(buff);
981                                         cmd_free(host, c, 0);
982                                         return -EFAULT;
983                                 }
984                         }
985                         kfree(buff);
986                         cmd_free(host, c, 0);
987                         return 0;
988                 }
989         case CCISS_BIG_PASSTHRU:{
990                         BIG_IOCTL_Command_struct *ioc;
991                         CommandList_struct *c;
992                         unsigned char **buff = NULL;
993                         int *buff_size = NULL;
994                         u64bit temp64;
995                         unsigned long flags;
996                         BYTE sg_used = 0;
997                         int status = 0;
998                         int i;
999                         DECLARE_COMPLETION_ONSTACK(wait);
1000                         __u32 left;
1001                         __u32 sz;
1002                         BYTE __user *data_ptr;
1003
1004                         if (!arg)
1005                                 return -EINVAL;
1006                         if (!capable(CAP_SYS_RAWIO))
1007                                 return -EPERM;
1008                         ioc = (BIG_IOCTL_Command_struct *)
1009                             kmalloc(sizeof(*ioc), GFP_KERNEL);
1010                         if (!ioc) {
1011                                 status = -ENOMEM;
1012                                 goto cleanup1;
1013                         }
1014                         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1015                                 status = -EFAULT;
1016                                 goto cleanup1;
1017                         }
1018                         if ((ioc->buf_size < 1) &&
1019                             (ioc->Request.Type.Direction != XFER_NONE)) {
1020                                 status = -EINVAL;
1021                                 goto cleanup1;
1022                         }
1023                         /* Check kmalloc limits  using all SGs */
1024                         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1025                                 status = -EINVAL;
1026                                 goto cleanup1;
1027                         }
1028                         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1029                                 status = -EINVAL;
1030                                 goto cleanup1;
1031                         }
1032                         buff =
1033                             kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1034                         if (!buff) {
1035                                 status = -ENOMEM;
1036                                 goto cleanup1;
1037                         }
1038                         buff_size = (int *)kmalloc(MAXSGENTRIES * sizeof(int),
1039                                                    GFP_KERNEL);
1040                         if (!buff_size) {
1041                                 status = -ENOMEM;
1042                                 goto cleanup1;
1043                         }
1044                         left = ioc->buf_size;
1045                         data_ptr = ioc->buf;
1046                         while (left) {
1047                                 sz = (left >
1048                                       ioc->malloc_size) ? ioc->
1049                                     malloc_size : left;
1050                                 buff_size[sg_used] = sz;
1051                                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1052                                 if (buff[sg_used] == NULL) {
1053                                         status = -ENOMEM;
1054                                         goto cleanup1;
1055                                 }
1056                                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1057                                         if (copy_from_user
1058                                             (buff[sg_used], data_ptr, sz)) {
1059                                                 status = -ENOMEM;
1060                                                 goto cleanup1;
1061                                         }
1062                                 } else {
1063                                         memset(buff[sg_used], 0, sz);
1064                                 }
1065                                 left -= sz;
1066                                 data_ptr += sz;
1067                                 sg_used++;
1068                         }
1069                         if ((c = cmd_alloc(host, 0)) == NULL) {
1070                                 status = -ENOMEM;
1071                                 goto cleanup1;
1072                         }
1073                         c->cmd_type = CMD_IOCTL_PEND;
1074                         c->Header.ReplyQueue = 0;
1075
1076                         if (ioc->buf_size > 0) {
1077                                 c->Header.SGList = sg_used;
1078                                 c->Header.SGTotal = sg_used;
1079                         } else {
1080                                 c->Header.SGList = 0;
1081                                 c->Header.SGTotal = 0;
1082                         }
1083                         c->Header.LUN = ioc->LUN_info;
1084                         c->Header.Tag.lower = c->busaddr;
1085
1086                         c->Request = ioc->Request;
1087                         if (ioc->buf_size > 0) {
1088                                 int i;
1089                                 for (i = 0; i < sg_used; i++) {
1090                                         temp64.val =
1091                                             pci_map_single(host->pdev, buff[i],
1092                                                     buff_size[i],
1093                                                     PCI_DMA_BIDIRECTIONAL);
1094                                         c->SG[i].Addr.lower =
1095                                             temp64.val32.lower;
1096                                         c->SG[i].Addr.upper =
1097                                             temp64.val32.upper;
1098                                         c->SG[i].Len = buff_size[i];
1099                                         c->SG[i].Ext = 0;       /* we are not chaining */
1100                                 }
1101                         }
1102                         c->waiting = &wait;
1103                         /* Put the request on the tail of the request queue */
1104                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1105                         addQ(&host->reqQ, c);
1106                         host->Qdepth++;
1107                         start_io(host);
1108                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1109                         wait_for_completion(&wait);
1110                         /* unlock the buffers from DMA */
1111                         for (i = 0; i < sg_used; i++) {
1112                                 temp64.val32.lower = c->SG[i].Addr.lower;
1113                                 temp64.val32.upper = c->SG[i].Addr.upper;
1114                                 pci_unmap_single(host->pdev,
1115                                         (dma_addr_t) temp64.val, buff_size[i],
1116                                         PCI_DMA_BIDIRECTIONAL);
1117                         }
1118                         /* Copy the error information out */
1119                         ioc->error_info = *(c->err_info);
1120                         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1121                                 cmd_free(host, c, 0);
1122                                 status = -EFAULT;
1123                                 goto cleanup1;
1124                         }
1125                         if (ioc->Request.Type.Direction == XFER_READ) {
1126                                 /* Copy the data out of the buffer we created */
1127                                 BYTE __user *ptr = ioc->buf;
1128                                 for (i = 0; i < sg_used; i++) {
1129                                         if (copy_to_user
1130                                             (ptr, buff[i], buff_size[i])) {
1131                                                 cmd_free(host, c, 0);
1132                                                 status = -EFAULT;
1133                                                 goto cleanup1;
1134                                         }
1135                                         ptr += buff_size[i];
1136                                 }
1137                         }
1138                         cmd_free(host, c, 0);
1139                         status = 0;
1140                       cleanup1:
1141                         if (buff) {
1142                                 for (i = 0; i < sg_used; i++)
1143                                         kfree(buff[i]);
1144                                 kfree(buff);
1145                         }
1146                         kfree(buff_size);
1147                         kfree(ioc);
1148                         return status;
1149                 }
1150         default:
1151                 return -ENOTTY;
1152         }
1153 }
1154
1155 /*
1156  * revalidate_allvol is for online array config utilities.  After a
1157  * utility reconfigures the drives in the array, it can use this function
1158  * (through an ioctl) to make the driver zap any previous disk structs for
1159  * that controller and get new ones.
1160  *
1161  * Right now I'm using the getgeometry() function to do this, but this
1162  * function should probably be finer grained and allow you to revalidate one
1163  * particular logical volume (instead of all of them on a particular
1164  * controller).
1165  */
1166 static int revalidate_allvol(ctlr_info_t *host)
1167 {
1168         int ctlr = host->ctlr, i;
1169         unsigned long flags;
1170
1171         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1172         if (host->usage_count > 1) {
1173                 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1174                 printk(KERN_WARNING "cciss: Device busy for volume"
1175                        " revalidation (usage=%d)\n", host->usage_count);
1176                 return -EBUSY;
1177         }
1178         host->usage_count++;
1179         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1180
1181         for (i = 0; i < NWD; i++) {
1182                 struct gendisk *disk = host->gendisk[i];
1183                 if (disk) {
1184                         request_queue_t *q = disk->queue;
1185
1186                         if (disk->flags & GENHD_FL_UP)
1187                                 del_gendisk(disk);
1188                         if (q)
1189                                 blk_cleanup_queue(q);
1190                 }
1191         }
1192
1193         /*
1194          * Set the partition and block size structures for all volumes
1195          * on this controller to zero.  We will reread all of this data
1196          */
1197         memset(host->drv, 0, sizeof(drive_info_struct)
1198                * CISS_MAX_LUN);
1199         /*
1200          * Tell the array controller not to give us any interrupts while
1201          * we check the new geometry.  Then turn interrupts back on when
1202          * we're done.
1203          */
1204         host->access.set_intr_mask(host, CCISS_INTR_OFF);
1205         cciss_getgeometry(ctlr);
1206         host->access.set_intr_mask(host, CCISS_INTR_ON);
1207
1208         /* Loop through each real device */
1209         for (i = 0; i < NWD; i++) {
1210                 struct gendisk *disk = host->gendisk[i];
1211                 drive_info_struct *drv = &(host->drv[i]);
1212                 /* we must register the controller even if no disks exist */
1213                 /* this is for the online array utilities */
1214                 if (!drv->heads && i)
1215                         continue;
1216                 blk_queue_hardsect_size(drv->queue, drv->block_size);
1217                 set_capacity(disk, drv->nr_blocks);
1218                 add_disk(disk);
1219         }
1220         host->usage_count--;
1221         return 0;
1222 }
1223
1224 static inline void complete_buffers(struct bio *bio, int status)
1225 {
1226         while (bio) {
1227                 struct bio *xbh = bio->bi_next;
1228                 int nr_sectors = bio_sectors(bio);
1229
1230                 bio->bi_next = NULL;
1231                 bio_endio(bio, nr_sectors << 9, status ? 0 : -EIO);
1232                 bio = xbh;
1233         }
1234 }
1235
1236 static void cciss_check_queues(ctlr_info_t *h)
1237 {
1238         int start_queue = h->next_to_run;
1239         int i;
1240
1241         /* check to see if we have maxed out the number of commands that can
1242          * be placed on the queue.  If so then exit.  We do this check here
1243          * in case the interrupt we serviced was from an ioctl and did not
1244          * free any new commands.
1245          */
1246         if ((find_first_zero_bit(h->cmd_pool_bits, NR_CMDS)) == NR_CMDS)
1247                 return;
1248
1249         /* We have room on the queue for more commands.  Now we need to queue
1250          * them up.  We will also keep track of the next queue to run so
1251          * that every queue gets a chance to be started first.
1252          */
1253         for (i = 0; i < h->highest_lun + 1; i++) {
1254                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1255                 /* make sure the disk has been added and the drive is real
1256                  * because this can be called from the middle of init_one.
1257                  */
1258                 if (!(h->drv[curr_queue].queue) || !(h->drv[curr_queue].heads))
1259                         continue;
1260                 blk_start_queue(h->gendisk[curr_queue]->queue);
1261
1262                 /* check to see if we have maxed out the number of commands
1263                  * that can be placed on the queue.
1264                  */
1265                 if ((find_first_zero_bit(h->cmd_pool_bits, NR_CMDS)) == NR_CMDS) {
1266                         if (curr_queue == start_queue) {
1267                                 h->next_to_run =
1268                                     (start_queue + 1) % (h->highest_lun + 1);
1269                                 break;
1270                         } else {
1271                                 h->next_to_run = curr_queue;
1272                                 break;
1273                         }
1274                 } else {
1275                         curr_queue = (curr_queue + 1) % (h->highest_lun + 1);
1276                 }
1277         }
1278 }
1279
1280 static void cciss_softirq_done(struct request *rq)
1281 {
1282         CommandList_struct *cmd = rq->completion_data;
1283         ctlr_info_t *h = hba[cmd->ctlr];
1284         unsigned long flags;
1285         u64bit temp64;
1286         int i, ddir;
1287
1288         if (cmd->Request.Type.Direction == XFER_READ)
1289                 ddir = PCI_DMA_FROMDEVICE;
1290         else
1291                 ddir = PCI_DMA_TODEVICE;
1292
1293         /* command did not need to be retried */
1294         /* unmap the DMA mapping for all the scatter gather elements */
1295         for (i = 0; i < cmd->Header.SGList; i++) {
1296                 temp64.val32.lower = cmd->SG[i].Addr.lower;
1297                 temp64.val32.upper = cmd->SG[i].Addr.upper;
1298                 pci_unmap_page(h->pdev, temp64.val, cmd->SG[i].Len, ddir);
1299         }
1300
1301         complete_buffers(rq->bio, rq->errors);
1302
1303 #ifdef CCISS_DEBUG
1304         printk("Done with %p\n", rq);
1305 #endif                          /* CCISS_DEBUG */
1306
1307         add_disk_randomness(rq->rq_disk);
1308         spin_lock_irqsave(&h->lock, flags);
1309         end_that_request_last(rq, rq->errors);
1310         cmd_free(h, cmd, 1);
1311         cciss_check_queues(h);
1312         spin_unlock_irqrestore(&h->lock, flags);
1313 }
1314
1315 /* This function will check the usage_count of the drive to be updated/added.
1316  * If the usage_count is zero then the drive information will be updated and
1317  * the disk will be re-registered with the kernel.  If not then it will be
1318  * left alone for the next reboot.  The exception to this is disk 0 which
1319  * will always be left registered with the kernel since it is also the
1320  * controller node.  Any changes to disk 0 will show up on the next
1321  * reboot.
1322  */
1323 static void cciss_update_drive_info(int ctlr, int drv_index)
1324 {
1325         ctlr_info_t *h = hba[ctlr];
1326         struct gendisk *disk;
1327         InquiryData_struct *inq_buff = NULL;
1328         unsigned int block_size;
1329         sector_t total_size;
1330         unsigned long flags = 0;
1331         int ret = 0;
1332
1333         /* if the disk already exists then deregister it before proceeding */
1334         if (h->drv[drv_index].raid_level != -1) {
1335                 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1336                 h->drv[drv_index].busy_configuring = 1;
1337                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1338                 ret = deregister_disk(h->gendisk[drv_index],
1339                                       &h->drv[drv_index], 0);
1340                 h->drv[drv_index].busy_configuring = 0;
1341         }
1342
1343         /* If the disk is in use return */
1344         if (ret)
1345                 return;
1346
1347         /* Get information about the disk and modify the driver structure */
1348         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1349         if (inq_buff == NULL)
1350                 goto mem_msg;
1351
1352         cciss_read_capacity(ctlr, drv_index, 1,
1353                             &total_size, &block_size);
1354
1355         /* total size = last LBA + 1 */
1356         /* FFFFFFFF + 1 = 0, cannot have a logical volume of size 0 */
1357         /* so we assume this volume this must be >2TB in size */
1358         if (total_size == (__u32) 0) {
1359                 cciss_read_capacity_16(ctlr, drv_index, 1,
1360                 &total_size, &block_size);
1361                 h->cciss_read = CCISS_READ_16;
1362                 h->cciss_write = CCISS_WRITE_16;
1363         } else {
1364                 h->cciss_read = CCISS_READ_10;
1365                 h->cciss_write = CCISS_WRITE_10;
1366         }
1367         cciss_geometry_inquiry(ctlr, drv_index, 1, total_size, block_size,
1368                                inq_buff, &h->drv[drv_index]);
1369
1370         ++h->num_luns;
1371         disk = h->gendisk[drv_index];
1372         set_capacity(disk, h->drv[drv_index].nr_blocks);
1373
1374         /* if it's the controller it's already added */
1375         if (drv_index) {
1376                 disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1377
1378                 /* Set up queue information */
1379                 disk->queue->backing_dev_info.ra_pages = READ_AHEAD;
1380                 blk_queue_bounce_limit(disk->queue, hba[ctlr]->pdev->dma_mask);
1381
1382                 /* This is a hardware imposed limit. */
1383                 blk_queue_max_hw_segments(disk->queue, MAXSGENTRIES);
1384
1385                 /* This is a limit in the driver and could be eliminated. */
1386                 blk_queue_max_phys_segments(disk->queue, MAXSGENTRIES);
1387
1388                 blk_queue_max_sectors(disk->queue, 512);
1389
1390                 blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1391
1392                 disk->queue->queuedata = hba[ctlr];
1393
1394                 blk_queue_hardsect_size(disk->queue,
1395                                         hba[ctlr]->drv[drv_index].block_size);
1396
1397                 h->drv[drv_index].queue = disk->queue;
1398                 add_disk(disk);
1399         }
1400
1401       freeret:
1402         kfree(inq_buff);
1403         return;
1404       mem_msg:
1405         printk(KERN_ERR "cciss: out of memory\n");
1406         goto freeret;
1407 }
1408
1409 /* This function will find the first index of the controllers drive array
1410  * that has a -1 for the raid_level and will return that index.  This is
1411  * where new drives will be added.  If the index to be returned is greater
1412  * than the highest_lun index for the controller then highest_lun is set
1413  * to this new index.  If there are no available indexes then -1 is returned.
1414  */
1415 static int cciss_find_free_drive_index(int ctlr)
1416 {
1417         int i;
1418
1419         for (i = 0; i < CISS_MAX_LUN; i++) {
1420                 if (hba[ctlr]->drv[i].raid_level == -1) {
1421                         if (i > hba[ctlr]->highest_lun)
1422                                 hba[ctlr]->highest_lun = i;
1423                         return i;
1424                 }
1425         }
1426         return -1;
1427 }
1428
1429 /* This function will add and remove logical drives from the Logical
1430  * drive array of the controller and maintain persistency of ordering
1431  * so that mount points are preserved until the next reboot.  This allows
1432  * for the removal of logical drives in the middle of the drive array
1433  * without a re-ordering of those drives.
1434  * INPUT
1435  * h            = The controller to perform the operations on
1436  * del_disk     = The disk to remove if specified.  If the value given
1437  *                is NULL then no disk is removed.
1438  */
1439 static int rebuild_lun_table(ctlr_info_t *h, struct gendisk *del_disk)
1440 {
1441         int ctlr = h->ctlr;
1442         int num_luns;
1443         ReportLunData_struct *ld_buff = NULL;
1444         drive_info_struct *drv = NULL;
1445         int return_code;
1446         int listlength = 0;
1447         int i;
1448         int drv_found;
1449         int drv_index = 0;
1450         __u32 lunid = 0;
1451         unsigned long flags;
1452
1453         /* Set busy_configuring flag for this operation */
1454         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1455         if (h->num_luns >= CISS_MAX_LUN) {
1456                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1457                 return -EINVAL;
1458         }
1459
1460         if (h->busy_configuring) {
1461                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1462                 return -EBUSY;
1463         }
1464         h->busy_configuring = 1;
1465
1466         /* if del_disk is NULL then we are being called to add a new disk
1467          * and update the logical drive table.  If it is not NULL then
1468          * we will check if the disk is in use or not.
1469          */
1470         if (del_disk != NULL) {
1471                 drv = get_drv(del_disk);
1472                 drv->busy_configuring = 1;
1473                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1474                 return_code = deregister_disk(del_disk, drv, 1);
1475                 drv->busy_configuring = 0;
1476                 h->busy_configuring = 0;
1477                 return return_code;
1478         } else {
1479                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1480                 if (!capable(CAP_SYS_RAWIO))
1481                         return -EPERM;
1482
1483                 ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
1484                 if (ld_buff == NULL)
1485                         goto mem_msg;
1486
1487                 return_code = sendcmd_withirq(CISS_REPORT_LOG, ctlr, ld_buff,
1488                                               sizeof(ReportLunData_struct), 0,
1489                                               0, 0, TYPE_CMD);
1490
1491                 if (return_code == IO_OK) {
1492                         listlength |=
1493                             (0xff & (unsigned int)(ld_buff->LUNListLength[0]))
1494                             << 24;
1495                         listlength |=
1496                             (0xff & (unsigned int)(ld_buff->LUNListLength[1]))
1497                             << 16;
1498                         listlength |=
1499                             (0xff & (unsigned int)(ld_buff->LUNListLength[2]))
1500                             << 8;
1501                         listlength |=
1502                             0xff & (unsigned int)(ld_buff->LUNListLength[3]);
1503                 } else {        /* reading number of logical volumes failed */
1504                         printk(KERN_WARNING "cciss: report logical volume"
1505                                " command failed\n");
1506                         listlength = 0;
1507                         goto freeret;
1508                 }
1509
1510                 num_luns = listlength / 8;      /* 8 bytes per entry */
1511                 if (num_luns > CISS_MAX_LUN) {
1512                         num_luns = CISS_MAX_LUN;
1513                         printk(KERN_WARNING "cciss: more luns configured"
1514                                " on controller than can be handled by"
1515                                " this driver.\n");
1516                 }
1517
1518                 /* Compare controller drive array to drivers drive array.
1519                  * Check for updates in the drive information and any new drives
1520                  * on the controller.
1521                  */
1522                 for (i = 0; i < num_luns; i++) {
1523                         int j;
1524
1525                         drv_found = 0;
1526
1527                         lunid = (0xff &
1528                                  (unsigned int)(ld_buff->LUN[i][3])) << 24;
1529                         lunid |= (0xff &
1530                                   (unsigned int)(ld_buff->LUN[i][2])) << 16;
1531                         lunid |= (0xff &
1532                                   (unsigned int)(ld_buff->LUN[i][1])) << 8;
1533                         lunid |= 0xff & (unsigned int)(ld_buff->LUN[i][0]);
1534
1535                         /* Find if the LUN is already in the drive array
1536                          * of the controller.  If so then update its info
1537                          * if not is use.  If it does not exist then find
1538                          * the first free index and add it.
1539                          */
1540                         for (j = 0; j <= h->highest_lun; j++) {
1541                                 if (h->drv[j].LunID == lunid) {
1542                                         drv_index = j;
1543                                         drv_found = 1;
1544                                 }
1545                         }
1546
1547                         /* check if the drive was found already in the array */
1548                         if (!drv_found) {
1549                                 drv_index = cciss_find_free_drive_index(ctlr);
1550                                 if (drv_index == -1)
1551                                         goto freeret;
1552
1553                         }
1554                         h->drv[drv_index].LunID = lunid;
1555                         cciss_update_drive_info(ctlr, drv_index);
1556                 }               /* end for */
1557         }                       /* end else */
1558
1559       freeret:
1560         kfree(ld_buff);
1561         h->busy_configuring = 0;
1562         /* We return -1 here to tell the ACU that we have registered/updated
1563          * all of the drives that we can and to keep it from calling us
1564          * additional times.
1565          */
1566         return -1;
1567       mem_msg:
1568         printk(KERN_ERR "cciss: out of memory\n");
1569         goto freeret;
1570 }
1571
1572 /* This function will deregister the disk and it's queue from the
1573  * kernel.  It must be called with the controller lock held and the
1574  * drv structures busy_configuring flag set.  It's parameters are:
1575  *
1576  * disk = This is the disk to be deregistered
1577  * drv  = This is the drive_info_struct associated with the disk to be
1578  *        deregistered.  It contains information about the disk used
1579  *        by the driver.
1580  * clear_all = This flag determines whether or not the disk information
1581  *             is going to be completely cleared out and the highest_lun
1582  *             reset.  Sometimes we want to clear out information about
1583  *             the disk in preparation for re-adding it.  In this case
1584  *             the highest_lun should be left unchanged and the LunID
1585  *             should not be cleared.
1586 */
1587 static int deregister_disk(struct gendisk *disk, drive_info_struct *drv,
1588                            int clear_all)
1589 {
1590         ctlr_info_t *h = get_host(disk);
1591
1592         if (!capable(CAP_SYS_RAWIO))
1593                 return -EPERM;
1594
1595         /* make sure logical volume is NOT is use */
1596         if (clear_all || (h->gendisk[0] == disk)) {
1597                 if (drv->usage_count > 1)
1598                         return -EBUSY;
1599         } else if (drv->usage_count > 0)
1600                 return -EBUSY;
1601
1602         /* invalidate the devices and deregister the disk.  If it is disk
1603          * zero do not deregister it but just zero out it's values.  This
1604          * allows us to delete disk zero but keep the controller registered.
1605          */
1606         if (h->gendisk[0] != disk) {
1607                 if (disk) {
1608                         request_queue_t *q = disk->queue;
1609                         if (disk->flags & GENHD_FL_UP)
1610                                 del_gendisk(disk);
1611                         if (q) {
1612                                 blk_cleanup_queue(q);
1613                                 drv->queue = NULL;
1614                         }
1615                 }
1616         }
1617
1618         --h->num_luns;
1619         /* zero out the disk size info */
1620         drv->nr_blocks = 0;
1621         drv->block_size = 0;
1622         drv->heads = 0;
1623         drv->sectors = 0;
1624         drv->cylinders = 0;
1625         drv->raid_level = -1;   /* This can be used as a flag variable to
1626                                  * indicate that this element of the drive
1627                                  * array is free.
1628                                  */
1629
1630         if (clear_all) {
1631                 /* check to see if it was the last disk */
1632                 if (drv == h->drv + h->highest_lun) {
1633                         /* if so, find the new hightest lun */
1634                         int i, newhighest = -1;
1635                         for (i = 0; i < h->highest_lun; i++) {
1636                                 /* if the disk has size > 0, it is available */
1637                                 if (h->drv[i].heads)
1638                                         newhighest = i;
1639                         }
1640                         h->highest_lun = newhighest;
1641                 }
1642
1643                 drv->LunID = 0;
1644         }
1645         return 0;
1646 }
1647
1648 static int fill_cmd(CommandList_struct *c, __u8 cmd, int ctlr, void *buff, size_t size, unsigned int use_unit_num,      /* 0: address the controller,
1649                                                                                                                            1: address logical volume log_unit,
1650                                                                                                                            2: periph device address is scsi3addr */
1651                     unsigned int log_unit, __u8 page_code,
1652                     unsigned char *scsi3addr, int cmd_type)
1653 {
1654         ctlr_info_t *h = hba[ctlr];
1655         u64bit buff_dma_handle;
1656         int status = IO_OK;
1657
1658         c->cmd_type = CMD_IOCTL_PEND;
1659         c->Header.ReplyQueue = 0;
1660         if (buff != NULL) {
1661                 c->Header.SGList = 1;
1662                 c->Header.SGTotal = 1;
1663         } else {
1664                 c->Header.SGList = 0;
1665                 c->Header.SGTotal = 0;
1666         }
1667         c->Header.Tag.lower = c->busaddr;
1668
1669         c->Request.Type.Type = cmd_type;
1670         if (cmd_type == TYPE_CMD) {
1671                 switch (cmd) {
1672                 case CISS_INQUIRY:
1673                         /* If the logical unit number is 0 then, this is going
1674                            to controller so It's a physical command
1675                            mode = 0 target = 0.  So we have nothing to write.
1676                            otherwise, if use_unit_num == 1,
1677                            mode = 1(volume set addressing) target = LUNID
1678                            otherwise, if use_unit_num == 2,
1679                            mode = 0(periph dev addr) target = scsi3addr */
1680                         if (use_unit_num == 1) {
1681                                 c->Header.LUN.LogDev.VolId =
1682                                     h->drv[log_unit].LunID;
1683                                 c->Header.LUN.LogDev.Mode = 1;
1684                         } else if (use_unit_num == 2) {
1685                                 memcpy(c->Header.LUN.LunAddrBytes, scsi3addr,
1686                                        8);
1687                                 c->Header.LUN.LogDev.Mode = 0;
1688                         }
1689                         /* are we trying to read a vital product page */
1690                         if (page_code != 0) {
1691                                 c->Request.CDB[1] = 0x01;
1692                                 c->Request.CDB[2] = page_code;
1693                         }
1694                         c->Request.CDBLen = 6;
1695                         c->Request.Type.Attribute = ATTR_SIMPLE;
1696                         c->Request.Type.Direction = XFER_READ;
1697                         c->Request.Timeout = 0;
1698                         c->Request.CDB[0] = CISS_INQUIRY;
1699                         c->Request.CDB[4] = size & 0xFF;
1700                         break;
1701                 case CISS_REPORT_LOG:
1702                 case CISS_REPORT_PHYS:
1703                         /* Talking to controller so It's a physical command
1704                            mode = 00 target = 0.  Nothing to write.
1705                          */
1706                         c->Request.CDBLen = 12;
1707                         c->Request.Type.Attribute = ATTR_SIMPLE;
1708                         c->Request.Type.Direction = XFER_READ;
1709                         c->Request.Timeout = 0;
1710                         c->Request.CDB[0] = cmd;
1711                         c->Request.CDB[6] = (size >> 24) & 0xFF;        //MSB
1712                         c->Request.CDB[7] = (size >> 16) & 0xFF;
1713                         c->Request.CDB[8] = (size >> 8) & 0xFF;
1714                         c->Request.CDB[9] = size & 0xFF;
1715                         break;
1716
1717                 case CCISS_READ_CAPACITY:
1718                         c->Header.LUN.LogDev.VolId = h->drv[log_unit].LunID;
1719                         c->Header.LUN.LogDev.Mode = 1;
1720                         c->Request.CDBLen = 10;
1721                         c->Request.Type.Attribute = ATTR_SIMPLE;
1722                         c->Request.Type.Direction = XFER_READ;
1723                         c->Request.Timeout = 0;
1724                         c->Request.CDB[0] = cmd;
1725                         break;
1726                 case CCISS_READ_CAPACITY_16:
1727                         c->Header.LUN.LogDev.VolId = h->drv[log_unit].LunID;
1728                         c->Header.LUN.LogDev.Mode = 1;
1729                         c->Request.CDBLen = 16;
1730                         c->Request.Type.Attribute = ATTR_SIMPLE;
1731                         c->Request.Type.Direction = XFER_READ;
1732                         c->Request.Timeout = 0;
1733                         c->Request.CDB[0] = cmd;
1734                         c->Request.CDB[1] = 0x10;
1735                         c->Request.CDB[10] = (size >> 24) & 0xFF;
1736                         c->Request.CDB[11] = (size >> 16) & 0xFF;
1737                         c->Request.CDB[12] = (size >> 8) & 0xFF;
1738                         c->Request.CDB[13] = size & 0xFF;
1739                         c->Request.Timeout = 0;
1740                         c->Request.CDB[0] = cmd;
1741                         break;
1742                 case CCISS_CACHE_FLUSH:
1743                         c->Request.CDBLen = 12;
1744                         c->Request.Type.Attribute = ATTR_SIMPLE;
1745                         c->Request.Type.Direction = XFER_WRITE;
1746                         c->Request.Timeout = 0;
1747                         c->Request.CDB[0] = BMIC_WRITE;
1748                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
1749                         break;
1750                 default:
1751                         printk(KERN_WARNING
1752                                "cciss%d:  Unknown Command 0x%c\n", ctlr, cmd);
1753                         return IO_ERROR;
1754                 }
1755         } else if (cmd_type == TYPE_MSG) {
1756                 switch (cmd) {
1757                 case 0: /* ABORT message */
1758                         c->Request.CDBLen = 12;
1759                         c->Request.Type.Attribute = ATTR_SIMPLE;
1760                         c->Request.Type.Direction = XFER_WRITE;
1761                         c->Request.Timeout = 0;
1762                         c->Request.CDB[0] = cmd;        /* abort */
1763                         c->Request.CDB[1] = 0;  /* abort a command */
1764                         /* buff contains the tag of the command to abort */
1765                         memcpy(&c->Request.CDB[4], buff, 8);
1766                         break;
1767                 case 1: /* RESET message */
1768                         c->Request.CDBLen = 12;
1769                         c->Request.Type.Attribute = ATTR_SIMPLE;
1770                         c->Request.Type.Direction = XFER_WRITE;
1771                         c->Request.Timeout = 0;
1772                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
1773                         c->Request.CDB[0] = cmd;        /* reset */
1774                         c->Request.CDB[1] = 0x04;       /* reset a LUN */
1775                         break;
1776                 case 3: /* No-Op message */
1777                         c->Request.CDBLen = 1;
1778                         c->Request.Type.Attribute = ATTR_SIMPLE;
1779                         c->Request.Type.Direction = XFER_WRITE;
1780                         c->Request.Timeout = 0;
1781                         c->Request.CDB[0] = cmd;
1782                         break;
1783                 default:
1784                         printk(KERN_WARNING
1785                                "cciss%d: unknown message type %d\n", ctlr, cmd);
1786                         return IO_ERROR;
1787                 }
1788         } else {
1789                 printk(KERN_WARNING
1790                        "cciss%d: unknown command type %d\n", ctlr, cmd_type);
1791                 return IO_ERROR;
1792         }
1793         /* Fill in the scatter gather information */
1794         if (size > 0) {
1795                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
1796                                                              buff, size,
1797                                                              PCI_DMA_BIDIRECTIONAL);
1798                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
1799                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
1800                 c->SG[0].Len = size;
1801                 c->SG[0].Ext = 0;       /* we are not chaining */
1802         }
1803         return status;
1804 }
1805
1806 static int sendcmd_withirq(__u8 cmd,
1807                            int ctlr,
1808                            void *buff,
1809                            size_t size,
1810                            unsigned int use_unit_num,
1811                            unsigned int log_unit, __u8 page_code, int cmd_type)
1812 {
1813         ctlr_info_t *h = hba[ctlr];
1814         CommandList_struct *c;
1815         u64bit buff_dma_handle;
1816         unsigned long flags;
1817         int return_status;
1818         DECLARE_COMPLETION_ONSTACK(wait);
1819
1820         if ((c = cmd_alloc(h, 0)) == NULL)
1821                 return -ENOMEM;
1822         return_status = fill_cmd(c, cmd, ctlr, buff, size, use_unit_num,
1823                                  log_unit, page_code, NULL, cmd_type);
1824         if (return_status != IO_OK) {
1825                 cmd_free(h, c, 0);
1826                 return return_status;
1827         }
1828       resend_cmd2:
1829         c->waiting = &wait;
1830
1831         /* Put the request on the tail of the queue and send it */
1832         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1833         addQ(&h->reqQ, c);
1834         h->Qdepth++;
1835         start_io(h);
1836         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1837
1838         wait_for_completion(&wait);
1839
1840         if (c->err_info->CommandStatus != 0) {  /* an error has occurred */
1841                 switch (c->err_info->CommandStatus) {
1842                 case CMD_TARGET_STATUS:
1843                         printk(KERN_WARNING "cciss: cmd %p has "
1844                                " completed with errors\n", c);
1845                         if (c->err_info->ScsiStatus) {
1846                                 printk(KERN_WARNING "cciss: cmd %p "
1847                                        "has SCSI Status = %x\n",
1848                                        c, c->err_info->ScsiStatus);
1849                         }
1850
1851                         break;
1852                 case CMD_DATA_UNDERRUN:
1853                 case CMD_DATA_OVERRUN:
1854                         /* expected for inquire and report lun commands */
1855                         break;
1856                 case CMD_INVALID:
1857                         printk(KERN_WARNING "cciss: Cmd %p is "
1858                                "reported invalid\n", c);
1859                         return_status = IO_ERROR;
1860                         break;
1861                 case CMD_PROTOCOL_ERR:
1862                         printk(KERN_WARNING "cciss: cmd %p has "
1863                                "protocol error \n", c);
1864                         return_status = IO_ERROR;
1865                         break;
1866                 case CMD_HARDWARE_ERR:
1867                         printk(KERN_WARNING "cciss: cmd %p had "
1868                                " hardware error\n", c);
1869                         return_status = IO_ERROR;
1870                         break;
1871                 case CMD_CONNECTION_LOST:
1872                         printk(KERN_WARNING "cciss: cmd %p had "
1873                                "connection lost\n", c);
1874                         return_status = IO_ERROR;
1875                         break;
1876                 case CMD_ABORTED:
1877                         printk(KERN_WARNING "cciss: cmd %p was "
1878                                "aborted\n", c);
1879                         return_status = IO_ERROR;
1880                         break;
1881                 case CMD_ABORT_FAILED:
1882                         printk(KERN_WARNING "cciss: cmd %p reports "
1883                                "abort failed\n", c);
1884                         return_status = IO_ERROR;
1885                         break;
1886                 case CMD_UNSOLICITED_ABORT:
1887                         printk(KERN_WARNING
1888                                "cciss%d: unsolicited abort %p\n", ctlr, c);
1889                         if (c->retry_count < MAX_CMD_RETRIES) {
1890                                 printk(KERN_WARNING
1891                                        "cciss%d: retrying %p\n", ctlr, c);
1892                                 c->retry_count++;
1893                                 /* erase the old error information */
1894                                 memset(c->err_info, 0,
1895                                        sizeof(ErrorInfo_struct));
1896                                 return_status = IO_OK;
1897                                 INIT_COMPLETION(wait);
1898                                 goto resend_cmd2;
1899                         }
1900                         return_status = IO_ERROR;
1901                         break;
1902                 default:
1903                         printk(KERN_WARNING "cciss: cmd %p returned "
1904                                "unknown status %x\n", c,
1905                                c->err_info->CommandStatus);
1906                         return_status = IO_ERROR;
1907                 }
1908         }
1909         /* unlock the buffers from DMA */
1910         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
1911         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
1912         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
1913                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
1914         cmd_free(h, c, 0);
1915         return return_status;
1916 }
1917
1918 static void cciss_geometry_inquiry(int ctlr, int logvol,
1919                                    int withirq, sector_t total_size,
1920                                    unsigned int block_size,
1921                                    InquiryData_struct *inq_buff,
1922                                    drive_info_struct *drv)
1923 {
1924         int return_code;
1925         unsigned long t;
1926         unsigned long rem;
1927
1928         memset(inq_buff, 0, sizeof(InquiryData_struct));
1929         if (withirq)
1930                 return_code = sendcmd_withirq(CISS_INQUIRY, ctlr,
1931                                               inq_buff, sizeof(*inq_buff), 1,
1932                                               logvol, 0xC1, TYPE_CMD);
1933         else
1934                 return_code = sendcmd(CISS_INQUIRY, ctlr, inq_buff,
1935                                       sizeof(*inq_buff), 1, logvol, 0xC1, NULL,
1936                                       TYPE_CMD);
1937         if (return_code == IO_OK) {
1938                 if (inq_buff->data_byte[8] == 0xFF) {
1939                         printk(KERN_WARNING
1940                                "cciss: reading geometry failed, volume "
1941                                "does not support reading geometry\n");
1942                         drv->block_size = block_size;
1943                         drv->nr_blocks = total_size;
1944                         drv->heads = 255;
1945                         drv->sectors = 32;      // Sectors per track
1946                         t = drv->heads * drv->sectors;
1947                         drv->cylinders = total_size;
1948                         rem = do_div(drv->cylinders, t);
1949                 } else {
1950                         drv->block_size = block_size;
1951                         drv->nr_blocks = total_size;
1952                         drv->heads = inq_buff->data_byte[6];
1953                         drv->sectors = inq_buff->data_byte[7];
1954                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
1955                         drv->cylinders += inq_buff->data_byte[5];
1956                         drv->raid_level = inq_buff->data_byte[8];
1957                         t = drv->heads * drv->sectors;
1958                         if (t > 1) {
1959                                 drv->cylinders = total_size;
1960                                 rem = do_div(drv->cylinders, t);
1961                         }
1962                 }
1963         } else {                /* Get geometry failed */
1964                 printk(KERN_WARNING "cciss: reading geometry failed\n");
1965         }
1966         printk(KERN_INFO "      heads=%d, sectors=%d, cylinders=%d\n\n",
1967                drv->heads, drv->sectors, drv->cylinders);
1968 }
1969
1970 static void
1971 cciss_read_capacity(int ctlr, int logvol, int withirq, sector_t *total_size,
1972                     unsigned int *block_size)
1973 {
1974         ReadCapdata_struct *buf;
1975         int return_code;
1976         buf = kmalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
1977         if (buf == NULL) {
1978                 printk(KERN_WARNING "cciss: out of memory\n");
1979                 return;
1980         }
1981         memset(buf, 0, sizeof(ReadCapdata_struct));
1982         if (withirq)
1983                 return_code = sendcmd_withirq(CCISS_READ_CAPACITY,
1984                                 ctlr, buf, sizeof(ReadCapdata_struct),
1985                                         1, logvol, 0, TYPE_CMD);
1986         else
1987                 return_code = sendcmd(CCISS_READ_CAPACITY,
1988                                 ctlr, buf, sizeof(ReadCapdata_struct),
1989                                         1, logvol, 0, NULL, TYPE_CMD);
1990         if (return_code == IO_OK) {
1991                 *total_size = be32_to_cpu(*(__u32 *) buf->total_size)+1;
1992                 *block_size = be32_to_cpu(*(__u32 *) buf->block_size);
1993         } else {                /* read capacity command failed */
1994                 printk(KERN_WARNING "cciss: read capacity failed\n");
1995                 *total_size = 0;
1996                 *block_size = BLOCK_SIZE;
1997         }
1998         if (*total_size != (__u32) 0)
1999                 printk(KERN_INFO "      blocks= %lld block_size= %d\n",
2000                 *total_size, *block_size);
2001         kfree(buf);
2002         return;
2003 }
2004
2005 static void
2006 cciss_read_capacity_16(int ctlr, int logvol, int withirq, sector_t *total_size,                                 unsigned int *block_size)
2007 {
2008         ReadCapdata_struct_16 *buf;
2009         int return_code;
2010         buf = kmalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2011         if (buf == NULL) {
2012                 printk(KERN_WARNING "cciss: out of memory\n");
2013                 return;
2014         }
2015         memset(buf, 0, sizeof(ReadCapdata_struct_16));
2016         if (withirq) {
2017                 return_code = sendcmd_withirq(CCISS_READ_CAPACITY_16,
2018                         ctlr, buf, sizeof(ReadCapdata_struct_16),
2019                                 1, logvol, 0, TYPE_CMD);
2020         }
2021         else {
2022                 return_code = sendcmd(CCISS_READ_CAPACITY_16,
2023                         ctlr, buf, sizeof(ReadCapdata_struct_16),
2024                                 1, logvol, 0, NULL, TYPE_CMD);
2025         }
2026         if (return_code == IO_OK) {
2027                 *total_size = be64_to_cpu(*(__u64 *) buf->total_size)+1;
2028                 *block_size = be32_to_cpu(*(__u32 *) buf->block_size);
2029         } else {                /* read capacity command failed */
2030                 printk(KERN_WARNING "cciss: read capacity failed\n");
2031                 *total_size = 0;
2032                 *block_size = BLOCK_SIZE;
2033         }
2034         printk(KERN_INFO "      blocks= %lld block_size= %d\n",
2035                *total_size, *block_size);
2036         kfree(buf);
2037         return;
2038 }
2039
2040 static int cciss_revalidate(struct gendisk *disk)
2041 {
2042         ctlr_info_t *h = get_host(disk);
2043         drive_info_struct *drv = get_drv(disk);
2044         int logvol;
2045         int FOUND = 0;
2046         unsigned int block_size;
2047         sector_t total_size;
2048         InquiryData_struct *inq_buff = NULL;
2049
2050         for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
2051                 if (h->drv[logvol].LunID == drv->LunID) {
2052                         FOUND = 1;
2053                         break;
2054                 }
2055         }
2056
2057         if (!FOUND)
2058                 return 1;
2059
2060         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2061         if (inq_buff == NULL) {
2062                 printk(KERN_WARNING "cciss: out of memory\n");
2063                 return 1;
2064         }
2065         if (h->cciss_read == CCISS_READ_10) {
2066                 cciss_read_capacity(h->ctlr, logvol, 1,
2067                                         &total_size, &block_size);
2068         } else {
2069                 cciss_read_capacity_16(h->ctlr, logvol, 1,
2070                                         &total_size, &block_size);
2071         }
2072         cciss_geometry_inquiry(h->ctlr, logvol, 1, total_size, block_size,
2073                                inq_buff, drv);
2074
2075         blk_queue_hardsect_size(drv->queue, drv->block_size);
2076         set_capacity(disk, drv->nr_blocks);
2077
2078         kfree(inq_buff);
2079         return 0;
2080 }
2081
2082 /*
2083  *   Wait polling for a command to complete.
2084  *   The memory mapped FIFO is polled for the completion.
2085  *   Used only at init time, interrupts from the HBA are disabled.
2086  */
2087 static unsigned long pollcomplete(int ctlr)
2088 {
2089         unsigned long done;
2090         int i;
2091
2092         /* Wait (up to 20 seconds) for a command to complete */
2093
2094         for (i = 20 * HZ; i > 0; i--) {
2095                 done = hba[ctlr]->access.command_completed(hba[ctlr]);
2096                 if (done == FIFO_EMPTY)
2097                         schedule_timeout_uninterruptible(1);
2098                 else
2099                         return done;
2100         }
2101         /* Invalid address to tell caller we ran out of time */
2102         return 1;
2103 }
2104
2105 static int add_sendcmd_reject(__u8 cmd, int ctlr, unsigned long complete)
2106 {
2107         /* We get in here if sendcmd() is polling for completions
2108            and gets some command back that it wasn't expecting --
2109            something other than that which it just sent down.
2110            Ordinarily, that shouldn't happen, but it can happen when
2111            the scsi tape stuff gets into error handling mode, and
2112            starts using sendcmd() to try to abort commands and
2113            reset tape drives.  In that case, sendcmd may pick up
2114            completions of commands that were sent to logical drives
2115            through the block i/o system, or cciss ioctls completing, etc.
2116            In that case, we need to save those completions for later
2117            processing by the interrupt handler.
2118          */
2119
2120 #ifdef CONFIG_CISS_SCSI_TAPE
2121         struct sendcmd_reject_list *srl = &hba[ctlr]->scsi_rejects;
2122
2123         /* If it's not the scsi tape stuff doing error handling, (abort */
2124         /* or reset) then we don't expect anything weird. */
2125         if (cmd != CCISS_RESET_MSG && cmd != CCISS_ABORT_MSG) {
2126 #endif
2127                 printk(KERN_WARNING "cciss cciss%d: SendCmd "
2128                        "Invalid command list address returned! (%lx)\n",
2129                        ctlr, complete);
2130                 /* not much we can do. */
2131 #ifdef CONFIG_CISS_SCSI_TAPE
2132                 return 1;
2133         }
2134
2135         /* We've sent down an abort or reset, but something else
2136            has completed */
2137         if (srl->ncompletions >= (NR_CMDS + 2)) {
2138                 /* Uh oh.  No room to save it for later... */
2139                 printk(KERN_WARNING "cciss%d: Sendcmd: Invalid command addr, "
2140                        "reject list overflow, command lost!\n", ctlr);
2141                 return 1;
2142         }
2143         /* Save it for later */
2144         srl->complete[srl->ncompletions] = complete;
2145         srl->ncompletions++;
2146 #endif
2147         return 0;
2148 }
2149
2150 /*
2151  * Send a command to the controller, and wait for it to complete.
2152  * Only used at init time.
2153  */
2154 static int sendcmd(__u8 cmd, int ctlr, void *buff, size_t size, unsigned int use_unit_num,      /* 0: address the controller,
2155                                                                                                    1: address logical volume log_unit,
2156                                                                                                    2: periph device address is scsi3addr */
2157                    unsigned int log_unit,
2158                    __u8 page_code, unsigned char *scsi3addr, int cmd_type)
2159 {
2160         CommandList_struct *c;
2161         int i;
2162         unsigned long complete;
2163         ctlr_info_t *info_p = hba[ctlr];
2164         u64bit buff_dma_handle;
2165         int status, done = 0;
2166
2167         if ((c = cmd_alloc(info_p, 1)) == NULL) {
2168                 printk(KERN_WARNING "cciss: unable to get memory");
2169                 return IO_ERROR;
2170         }
2171         status = fill_cmd(c, cmd, ctlr, buff, size, use_unit_num,
2172                           log_unit, page_code, scsi3addr, cmd_type);
2173         if (status != IO_OK) {
2174                 cmd_free(info_p, c, 1);
2175                 return status;
2176         }
2177       resend_cmd1:
2178         /*
2179          * Disable interrupt
2180          */
2181 #ifdef CCISS_DEBUG
2182         printk(KERN_DEBUG "cciss: turning intr off\n");
2183 #endif                          /* CCISS_DEBUG */
2184         info_p->access.set_intr_mask(info_p, CCISS_INTR_OFF);
2185
2186         /* Make sure there is room in the command FIFO */
2187         /* Actually it should be completely empty at this time */
2188         /* unless we are in here doing error handling for the scsi */
2189         /* tape side of the driver. */
2190         for (i = 200000; i > 0; i--) {
2191                 /* if fifo isn't full go */
2192                 if (!(info_p->access.fifo_full(info_p))) {
2193
2194                         break;
2195                 }
2196                 udelay(10);
2197                 printk(KERN_WARNING "cciss cciss%d: SendCmd FIFO full,"
2198                        " waiting!\n", ctlr);
2199         }
2200         /*
2201          * Send the cmd
2202          */
2203         info_p->access.submit_command(info_p, c);
2204         done = 0;
2205         do {
2206                 complete = pollcomplete(ctlr);
2207
2208 #ifdef CCISS_DEBUG
2209                 printk(KERN_DEBUG "cciss: command completed\n");
2210 #endif                          /* CCISS_DEBUG */
2211
2212                 if (complete == 1) {
2213                         printk(KERN_WARNING
2214                                "cciss cciss%d: SendCmd Timeout out, "
2215                                "No command list address returned!\n", ctlr);
2216                         status = IO_ERROR;
2217                         done = 1;
2218                         break;
2219                 }
2220
2221                 /* This will need to change for direct lookup completions */
2222                 if ((complete & CISS_ERROR_BIT)
2223                     && (complete & ~CISS_ERROR_BIT) == c->busaddr) {
2224                         /* if data overrun or underun on Report command
2225                            ignore it
2226                          */
2227                         if (((c->Request.CDB[0] == CISS_REPORT_LOG) ||
2228                              (c->Request.CDB[0] == CISS_REPORT_PHYS) ||
2229                              (c->Request.CDB[0] == CISS_INQUIRY)) &&
2230                             ((c->err_info->CommandStatus ==
2231                               CMD_DATA_OVERRUN) ||
2232                              (c->err_info->CommandStatus == CMD_DATA_UNDERRUN)
2233                             )) {
2234                                 complete = c->busaddr;
2235                         } else {
2236                                 if (c->err_info->CommandStatus ==
2237                                     CMD_UNSOLICITED_ABORT) {
2238                                         printk(KERN_WARNING "cciss%d: "
2239                                                "unsolicited abort %p\n",
2240                                                ctlr, c);
2241                                         if (c->retry_count < MAX_CMD_RETRIES) {
2242                                                 printk(KERN_WARNING
2243                                                        "cciss%d: retrying %p\n",
2244                                                        ctlr, c);
2245                                                 c->retry_count++;
2246                                                 /* erase the old error */
2247                                                 /* information */
2248                                                 memset(c->err_info, 0,
2249                                                        sizeof
2250                                                        (ErrorInfo_struct));
2251                                                 goto resend_cmd1;
2252                                         } else {
2253                                                 printk(KERN_WARNING
2254                                                        "cciss%d: retried %p too "
2255                                                        "many times\n", ctlr, c);
2256                                                 status = IO_ERROR;
2257                                                 goto cleanup1;
2258                                         }
2259                                 } else if (c->err_info->CommandStatus ==
2260                                            CMD_UNABORTABLE) {
2261                                         printk(KERN_WARNING
2262                                                "cciss%d: command could not be aborted.\n",
2263                                                ctlr);
2264                                         status = IO_ERROR;
2265                                         goto cleanup1;
2266                                 }
2267                                 printk(KERN_WARNING "ciss ciss%d: sendcmd"
2268                                        " Error %x \n", ctlr,
2269                                        c->err_info->CommandStatus);
2270                                 printk(KERN_WARNING "ciss ciss%d: sendcmd"
2271                                        " offensive info\n"
2272                                        "  size %x\n   num %x   value %x\n",
2273                                        ctlr,
2274                                        c->err_info->MoreErrInfo.Invalid_Cmd.
2275                                        offense_size,
2276                                        c->err_info->MoreErrInfo.Invalid_Cmd.
2277                                        offense_num,
2278                                        c->err_info->MoreErrInfo.Invalid_Cmd.
2279                                        offense_value);
2280                                 status = IO_ERROR;
2281                                 goto cleanup1;
2282                         }
2283                 }
2284                 /* This will need changing for direct lookup completions */
2285                 if (complete != c->busaddr) {
2286                         if (add_sendcmd_reject(cmd, ctlr, complete) != 0) {
2287                                 BUG();  /* we are pretty much hosed if we get here. */
2288                         }
2289                         continue;
2290                 } else
2291                         done = 1;
2292         } while (!done);
2293
2294       cleanup1:
2295         /* unlock the data buffer from DMA */
2296         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2297         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2298         pci_unmap_single(info_p->pdev, (dma_addr_t) buff_dma_handle.val,
2299                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2300 #ifdef CONFIG_CISS_SCSI_TAPE
2301         /* if we saved some commands for later, process them now. */
2302         if (info_p->scsi_rejects.ncompletions > 0)
2303                 do_cciss_intr(0, info_p);
2304 #endif
2305         cmd_free(info_p, c, 1);
2306         return status;
2307 }
2308
2309 /*
2310  * Map (physical) PCI mem into (virtual) kernel space
2311  */
2312 static void __iomem *remap_pci_mem(ulong base, ulong size)
2313 {
2314         ulong page_base = ((ulong) base) & PAGE_MASK;
2315         ulong page_offs = ((ulong) base) - page_base;
2316         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2317
2318         return page_remapped ? (page_remapped + page_offs) : NULL;
2319 }
2320
2321 /*
2322  * Takes jobs of the Q and sends them to the hardware, then puts it on
2323  * the Q to wait for completion.
2324  */
2325 static void start_io(ctlr_info_t *h)
2326 {
2327         CommandList_struct *c;
2328
2329         while ((c = h->reqQ) != NULL) {
2330                 /* can't do anything if fifo is full */
2331                 if ((h->access.fifo_full(h))) {
2332                         printk(KERN_WARNING "cciss: fifo full\n");
2333                         break;
2334                 }
2335
2336                 /* Get the first entry from the Request Q */
2337                 removeQ(&(h->reqQ), c);
2338                 h->Qdepth--;
2339
2340                 /* Tell the controller execute command */
2341                 h->access.submit_command(h, c);
2342
2343                 /* Put job onto the completed Q */
2344                 addQ(&(h->cmpQ), c);
2345         }
2346 }
2347
2348 /* Assumes that CCISS_LOCK(h->ctlr) is held. */
2349 /* Zeros out the error record and then resends the command back */
2350 /* to the controller */
2351 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2352 {
2353         /* erase the old error information */
2354         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2355
2356         /* add it to software queue and then send it to the controller */
2357         addQ(&(h->reqQ), c);
2358         h->Qdepth++;
2359         if (h->Qdepth > h->maxQsinceinit)
2360                 h->maxQsinceinit = h->Qdepth;
2361
2362         start_io(h);
2363 }
2364
2365 /* checks the status of the job and calls complete buffers to mark all
2366  * buffers for the completed job. Note that this function does not need
2367  * to hold the hba/queue lock.
2368  */
2369 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
2370                                     int timeout)
2371 {
2372         int status = 1;
2373         int retry_cmd = 0;
2374
2375         if (timeout)
2376                 status = 0;
2377
2378         if (cmd->err_info->CommandStatus != 0) {        /* an error has occurred */
2379                 switch (cmd->err_info->CommandStatus) {
2380                         unsigned char sense_key;
2381                 case CMD_TARGET_STATUS:
2382                         status = 0;
2383
2384                         if (cmd->err_info->ScsiStatus == 0x02) {
2385                                 printk(KERN_WARNING "cciss: cmd %p "
2386                                        "has CHECK CONDITION "
2387                                        " byte 2 = 0x%x\n", cmd,
2388                                        cmd->err_info->SenseInfo[2]
2389                                     );
2390                                 /* check the sense key */
2391                                 sense_key = 0xf & cmd->err_info->SenseInfo[2];
2392                                 /* no status or recovered error */
2393                                 if ((sense_key == 0x0) || (sense_key == 0x1)) {
2394                                         status = 1;
2395                                 }
2396                         } else {
2397                                 printk(KERN_WARNING "cciss: cmd %p "
2398                                        "has SCSI Status 0x%x\n",
2399                                        cmd, cmd->err_info->ScsiStatus);
2400                         }
2401                         break;
2402                 case CMD_DATA_UNDERRUN:
2403                         printk(KERN_WARNING "cciss: cmd %p has"
2404                                " completed with data underrun "
2405                                "reported\n", cmd);
2406                         break;
2407                 case CMD_DATA_OVERRUN:
2408                         printk(KERN_WARNING "cciss: cmd %p has"
2409                                " completed with data overrun "
2410                                "reported\n", cmd);
2411                         break;
2412                 case CMD_INVALID:
2413                         printk(KERN_WARNING "cciss: cmd %p is "
2414                                "reported invalid\n", cmd);
2415                         status = 0;
2416                         break;
2417                 case CMD_PROTOCOL_ERR:
2418                         printk(KERN_WARNING "cciss: cmd %p has "
2419                                "protocol error \n", cmd);
2420                         status = 0;
2421                         break;
2422                 case CMD_HARDWARE_ERR:
2423                         printk(KERN_WARNING "cciss: cmd %p had "
2424                                " hardware error\n", cmd);
2425                         status = 0;
2426                         break;
2427                 case CMD_CONNECTION_LOST:
2428                         printk(KERN_WARNING "cciss: cmd %p had "
2429                                "connection lost\n", cmd);
2430                         status = 0;
2431                         break;
2432                 case CMD_ABORTED:
2433                         printk(KERN_WARNING "cciss: cmd %p was "
2434                                "aborted\n", cmd);
2435                         status = 0;
2436                         break;
2437                 case CMD_ABORT_FAILED:
2438                         printk(KERN_WARNING "cciss: cmd %p reports "
2439                                "abort failed\n", cmd);
2440                         status = 0;
2441                         break;
2442                 case CMD_UNSOLICITED_ABORT:
2443                         printk(KERN_WARNING "cciss%d: unsolicited "
2444                                "abort %p\n", h->ctlr, cmd);
2445                         if (cmd->retry_count < MAX_CMD_RETRIES) {
2446                                 retry_cmd = 1;
2447                                 printk(KERN_WARNING
2448                                        "cciss%d: retrying %p\n", h->ctlr, cmd);
2449                                 cmd->retry_count++;
2450                         } else
2451                                 printk(KERN_WARNING
2452                                        "cciss%d: %p retried too "
2453                                        "many times\n", h->ctlr, cmd);
2454                         status = 0;
2455                         break;
2456                 case CMD_TIMEOUT:
2457                         printk(KERN_WARNING "cciss: cmd %p timedout\n", cmd);
2458                         status = 0;
2459                         break;
2460                 default:
2461                         printk(KERN_WARNING "cciss: cmd %p returned "
2462                                "unknown status %x\n", cmd,
2463                                cmd->err_info->CommandStatus);
2464                         status = 0;
2465                 }
2466         }
2467         /* We need to return this command */
2468         if (retry_cmd) {
2469                 resend_cciss_cmd(h, cmd);
2470                 return;
2471         }
2472
2473         cmd->rq->completion_data = cmd;
2474         cmd->rq->errors = status;
2475         blk_add_trace_rq(cmd->rq->q, cmd->rq, BLK_TA_COMPLETE);
2476         blk_complete_request(cmd->rq);
2477 }
2478
2479 /*
2480  * Get a request and submit it to the controller.
2481  */
2482 static void do_cciss_request(request_queue_t *q)
2483 {
2484         ctlr_info_t *h = q->queuedata;
2485         CommandList_struct *c;
2486         sector_t start_blk;
2487         int seg;
2488         struct request *creq;
2489         u64bit temp64;
2490         struct scatterlist tmp_sg[MAXSGENTRIES];
2491         drive_info_struct *drv;
2492         int i, dir;
2493
2494         /* We call start_io here in case there is a command waiting on the
2495          * queue that has not been sent.
2496          */
2497         if (blk_queue_plugged(q))
2498                 goto startio;
2499
2500       queue:
2501         creq = elv_next_request(q);
2502         if (!creq)
2503                 goto startio;
2504
2505         BUG_ON(creq->nr_phys_segments > MAXSGENTRIES);
2506
2507         if ((c = cmd_alloc(h, 1)) == NULL)
2508                 goto full;
2509
2510         blkdev_dequeue_request(creq);
2511
2512         spin_unlock_irq(q->queue_lock);
2513
2514         c->cmd_type = CMD_RWREQ;
2515         c->rq = creq;
2516
2517         /* fill in the request */
2518         drv = creq->rq_disk->private_data;
2519         c->Header.ReplyQueue = 0;       // unused in simple mode
2520         /* got command from pool, so use the command block index instead */
2521         /* for direct lookups. */
2522         /* The first 2 bits are reserved for controller error reporting. */
2523         c->Header.Tag.lower = (c->cmdindex << 3);
2524         c->Header.Tag.lower |= 0x04;    /* flag for direct lookup. */
2525         c->Header.LUN.LogDev.VolId = drv->LunID;
2526         c->Header.LUN.LogDev.Mode = 1;
2527         c->Request.CDBLen = 10; // 12 byte commands not in FW yet;
2528         c->Request.Type.Type = TYPE_CMD;        // It is a command.
2529         c->Request.Type.Attribute = ATTR_SIMPLE;
2530         c->Request.Type.Direction =
2531             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
2532         c->Request.Timeout = 0; // Don't time out
2533         c->Request.CDB[0] =
2534             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
2535         start_blk = creq->sector;
2536 #ifdef CCISS_DEBUG
2537         printk(KERN_DEBUG "ciss: sector =%d nr_sectors=%d\n", (int)creq->sector,
2538                (int)creq->nr_sectors);
2539 #endif                          /* CCISS_DEBUG */
2540
2541         seg = blk_rq_map_sg(q, creq, tmp_sg);
2542
2543         /* get the DMA records for the setup */
2544         if (c->Request.Type.Direction == XFER_READ)
2545                 dir = PCI_DMA_FROMDEVICE;
2546         else
2547                 dir = PCI_DMA_TODEVICE;
2548
2549         for (i = 0; i < seg; i++) {
2550                 c->SG[i].Len = tmp_sg[i].length;
2551                 temp64.val = (__u64) pci_map_page(h->pdev, tmp_sg[i].page,
2552                                                   tmp_sg[i].offset,
2553                                                   tmp_sg[i].length, dir);
2554                 c->SG[i].Addr.lower = temp64.val32.lower;
2555                 c->SG[i].Addr.upper = temp64.val32.upper;
2556                 c->SG[i].Ext = 0;       // we are not chaining
2557         }
2558         /* track how many SG entries we are using */
2559         if (seg > h->maxSG)
2560                 h->maxSG = seg;
2561
2562 #ifdef CCISS_DEBUG
2563         printk(KERN_DEBUG "cciss: Submitting %d sectors in %d segments\n",
2564                creq->nr_sectors, seg);
2565 #endif                          /* CCISS_DEBUG */
2566
2567         c->Header.SGList = c->Header.SGTotal = seg;
2568         if(h->cciss_read == CCISS_READ_10) {
2569                 c->Request.CDB[1] = 0;
2570                 c->Request.CDB[2] = (start_blk >> 24) & 0xff;   //MSB
2571                 c->Request.CDB[3] = (start_blk >> 16) & 0xff;
2572                 c->Request.CDB[4] = (start_blk >> 8) & 0xff;
2573                 c->Request.CDB[5] = start_blk & 0xff;
2574                 c->Request.CDB[6] = 0;  // (sect >> 24) & 0xff; MSB
2575                 c->Request.CDB[7] = (creq->nr_sectors >> 8) & 0xff;
2576                 c->Request.CDB[8] = creq->nr_sectors & 0xff;
2577                 c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
2578         } else {
2579                 c->Request.CDBLen = 16;
2580                 c->Request.CDB[1]= 0;
2581                 c->Request.CDB[2]= (start_blk >> 56) & 0xff;    //MSB
2582                 c->Request.CDB[3]= (start_blk >> 48) & 0xff;
2583                 c->Request.CDB[4]= (start_blk >> 40) & 0xff;
2584                 c->Request.CDB[5]= (start_blk >> 32) & 0xff;
2585                 c->Request.CDB[6]= (start_blk >> 24) & 0xff;
2586                 c->Request.CDB[7]= (start_blk >> 16) & 0xff;
2587                 c->Request.CDB[8]= (start_blk >>  8) & 0xff;
2588                 c->Request.CDB[9]= start_blk & 0xff;
2589                 c->Request.CDB[10]= (creq->nr_sectors >>  24) & 0xff;
2590                 c->Request.CDB[11]= (creq->nr_sectors >>  16) & 0xff;
2591                 c->Request.CDB[12]= (creq->nr_sectors >>  8) & 0xff;
2592                 c->Request.CDB[13]= creq->nr_sectors & 0xff;
2593                 c->Request.CDB[14] = c->Request.CDB[15] = 0;
2594         }
2595
2596         spin_lock_irq(q->queue_lock);
2597
2598         addQ(&(h->reqQ), c);
2599         h->Qdepth++;
2600         if (h->Qdepth > h->maxQsinceinit)
2601                 h->maxQsinceinit = h->Qdepth;
2602
2603         goto queue;
2604 full:
2605         blk_stop_queue(q);
2606 startio:
2607         /* We will already have the driver lock here so not need
2608          * to lock it.
2609          */
2610         start_io(h);
2611 }
2612
2613 static inline unsigned long get_next_completion(ctlr_info_t *h)
2614 {
2615 #ifdef CONFIG_CISS_SCSI_TAPE
2616         /* Any rejects from sendcmd() lying around? Process them first */
2617         if (h->scsi_rejects.ncompletions == 0)
2618                 return h->access.command_completed(h);
2619         else {
2620                 struct sendcmd_reject_list *srl;
2621                 int n;
2622                 srl = &h->scsi_rejects;
2623                 n = --srl->ncompletions;
2624                 /* printk("cciss%d: processing saved reject\n", h->ctlr); */
2625                 printk("p");
2626                 return srl->complete[n];
2627         }
2628 #else
2629         return h->access.command_completed(h);
2630 #endif
2631 }
2632
2633 static inline int interrupt_pending(ctlr_info_t *h)
2634 {
2635 #ifdef CONFIG_CISS_SCSI_TAPE
2636         return (h->access.intr_pending(h)
2637                 || (h->scsi_rejects.ncompletions > 0));
2638 #else
2639         return h->access.intr_pending(h);
2640 #endif
2641 }
2642
2643 static inline long interrupt_not_for_us(ctlr_info_t *h)
2644 {
2645 #ifdef CONFIG_CISS_SCSI_TAPE
2646         return (((h->access.intr_pending(h) == 0) ||
2647                  (h->interrupts_enabled == 0))
2648                 && (h->scsi_rejects.ncompletions == 0));
2649 #else
2650         return (((h->access.intr_pending(h) == 0) ||
2651                  (h->interrupts_enabled == 0)));
2652 #endif
2653 }
2654
2655 static irqreturn_t do_cciss_intr(int irq, void *dev_id)
2656 {
2657         ctlr_info_t *h = dev_id;
2658         CommandList_struct *c;
2659         unsigned long flags;
2660         __u32 a, a1, a2;
2661
2662         if (interrupt_not_for_us(h))
2663                 return IRQ_NONE;
2664         /*
2665          * If there are completed commands in the completion queue,
2666          * we had better do something about it.
2667          */
2668         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2669         while (interrupt_pending(h)) {
2670                 while ((a = get_next_completion(h)) != FIFO_EMPTY) {
2671                         a1 = a;
2672                         if ((a & 0x04)) {
2673                                 a2 = (a >> 3);
2674                                 if (a2 >= NR_CMDS) {
2675                                         printk(KERN_WARNING
2676                                                "cciss: controller cciss%d failed, stopping.\n",
2677                                                h->ctlr);
2678                                         fail_all_cmds(h->ctlr);
2679                                         return IRQ_HANDLED;
2680                                 }
2681
2682                                 c = h->cmd_pool + a2;
2683                                 a = c->busaddr;
2684
2685                         } else {
2686                                 a &= ~3;
2687                                 if ((c = h->cmpQ) == NULL) {
2688                                         printk(KERN_WARNING
2689                                                "cciss: Completion of %08x ignored\n",
2690                                                a1);
2691                                         continue;
2692                                 }
2693                                 while (c->busaddr != a) {
2694                                         c = c->next;
2695                                         if (c == h->cmpQ)
2696                                                 break;
2697                                 }
2698                         }
2699                         /*
2700                          * If we've found the command, take it off the
2701                          * completion Q and free it
2702                          */
2703                         if (c->busaddr == a) {
2704                                 removeQ(&h->cmpQ, c);
2705                                 if (c->cmd_type == CMD_RWREQ) {
2706                                         complete_command(h, c, 0);
2707                                 } else if (c->cmd_type == CMD_IOCTL_PEND) {
2708                                         complete(c->waiting);
2709                                 }
2710 #                               ifdef CONFIG_CISS_SCSI_TAPE
2711                                 else if (c->cmd_type == CMD_SCSI)
2712                                         complete_scsi_command(c, 0, a1);
2713 #                               endif
2714                                 continue;
2715                         }
2716                 }
2717         }
2718
2719         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2720         return IRQ_HANDLED;
2721 }
2722
2723 /*
2724  *  We cannot read the structure directly, for portability we must use
2725  *   the io functions.
2726  *   This is for debug only.
2727  */
2728 #ifdef CCISS_DEBUG
2729 static void print_cfg_table(CfgTable_struct *tb)
2730 {
2731         int i;
2732         char temp_name[17];
2733
2734         printk("Controller Configuration information\n");
2735         printk("------------------------------------\n");
2736         for (i = 0; i < 4; i++)
2737                 temp_name[i] = readb(&(tb->Signature[i]));
2738         temp_name[4] = '\0';
2739         printk("   Signature = %s\n", temp_name);
2740         printk("   Spec Number = %d\n", readl(&(tb->SpecValence)));
2741         printk("   Transport methods supported = 0x%x\n",
2742                readl(&(tb->TransportSupport)));
2743         printk("   Transport methods active = 0x%x\n",
2744                readl(&(tb->TransportActive)));
2745         printk("   Requested transport Method = 0x%x\n",
2746                readl(&(tb->HostWrite.TransportRequest)));
2747         printk("   Coalesce Interrupt Delay = 0x%x\n",
2748                readl(&(tb->HostWrite.CoalIntDelay)));
2749         printk("   Coalesce Interrupt Count = 0x%x\n",
2750                readl(&(tb->HostWrite.CoalIntCount)));
2751         printk("   Max outstanding commands = 0x%d\n",
2752                readl(&(tb->CmdsOutMax)));
2753         printk("   Bus Types = 0x%x\n", readl(&(tb->BusTypes)));
2754         for (i = 0; i < 16; i++)
2755                 temp_name[i] = readb(&(tb->ServerName[i]));
2756         temp_name[16] = '\0';
2757         printk("   Server Name = %s\n", temp_name);
2758         printk("   Heartbeat Counter = 0x%x\n\n\n", readl(&(tb->HeartBeat)));
2759 }
2760 #endif                          /* CCISS_DEBUG */
2761
2762 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
2763 {
2764         int i, offset, mem_type, bar_type;
2765         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
2766                 return 0;
2767         offset = 0;
2768         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
2769                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
2770                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
2771                         offset += 4;
2772                 else {
2773                         mem_type = pci_resource_flags(pdev, i) &
2774                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
2775                         switch (mem_type) {
2776                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
2777                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
2778                                 offset += 4;    /* 32 bit */
2779                                 break;
2780                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
2781                                 offset += 8;
2782                                 break;
2783                         default:        /* reserved in PCI 2.2 */
2784                                 printk(KERN_WARNING
2785                                        "Base address is invalid\n");
2786                                 return -1;
2787                                 break;
2788                         }
2789                 }
2790                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
2791                         return i + 1;
2792         }
2793         return -1;
2794 }
2795
2796 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
2797  * controllers that are capable. If not, we use IO-APIC mode.
2798  */
2799
2800 static void __devinit cciss_interrupt_mode(ctlr_info_t *c,
2801                                            struct pci_dev *pdev, __u32 board_id)
2802 {
2803 #ifdef CONFIG_PCI_MSI
2804         int err;
2805         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
2806         {0, 2}, {0, 3}
2807         };
2808
2809         /* Some boards advertise MSI but don't really support it */
2810         if ((board_id == 0x40700E11) ||
2811             (board_id == 0x40800E11) ||
2812             (board_id == 0x40820E11) || (board_id == 0x40830E11))
2813                 goto default_int_mode;
2814
2815         if (pci_find_capability(pdev, PCI_CAP_ID_MSIX)) {
2816                 err = pci_enable_msix(pdev, cciss_msix_entries, 4);
2817                 if (!err) {
2818                         c->intr[0] = cciss_msix_entries[0].vector;
2819                         c->intr[1] = cciss_msix_entries[1].vector;
2820                         c->intr[2] = cciss_msix_entries[2].vector;
2821                         c->intr[3] = cciss_msix_entries[3].vector;
2822                         c->msix_vector = 1;
2823                         return;
2824                 }
2825                 if (err > 0) {
2826                         printk(KERN_WARNING "cciss: only %d MSI-X vectors "
2827                                "available\n", err);
2828                 } else {
2829                         printk(KERN_WARNING "cciss: MSI-X init failed %d\n",
2830                                err);
2831                 }
2832         }
2833         if (pci_find_capability(pdev, PCI_CAP_ID_MSI)) {
2834                 if (!pci_enable_msi(pdev)) {
2835                         c->intr[SIMPLE_MODE_INT] = pdev->irq;
2836                         c->msi_vector = 1;
2837                         return;
2838                 } else {
2839                         printk(KERN_WARNING "cciss: MSI init failed\n");
2840                         c->intr[SIMPLE_MODE_INT] = pdev->irq;
2841                         return;
2842                 }
2843         }
2844       default_int_mode:
2845 #endif                          /* CONFIG_PCI_MSI */
2846         /* if we get here we're going to use the default interrupt mode */
2847         c->intr[SIMPLE_MODE_INT] = pdev->irq;
2848         return;
2849 }
2850
2851 static int cciss_pci_init(ctlr_info_t *c, struct pci_dev *pdev)
2852 {
2853         ushort subsystem_vendor_id, subsystem_device_id, command;
2854         __u32 board_id, scratchpad = 0;
2855         __u64 cfg_offset;
2856         __u32 cfg_base_addr;
2857         __u64 cfg_base_addr_index;
2858         int i, err;
2859
2860         /* check to see if controller has been disabled */
2861         /* BEFORE trying to enable it */
2862         (void)pci_read_config_word(pdev, PCI_COMMAND, &command);
2863         if (!(command & 0x02)) {
2864                 printk(KERN_WARNING
2865                        "cciss: controller appears to be disabled\n");
2866                 return -ENODEV;
2867         }
2868
2869         err = pci_enable_device(pdev);
2870         if (err) {
2871                 printk(KERN_ERR "cciss: Unable to Enable PCI device\n");
2872                 return err;
2873         }
2874
2875         err = pci_request_regions(pdev, "cciss");
2876         if (err) {
2877                 printk(KERN_ERR "cciss: Cannot obtain PCI resources, "
2878                        "aborting\n");
2879                 goto err_out_disable_pdev;
2880         }
2881
2882         subsystem_vendor_id = pdev->subsystem_vendor;
2883         subsystem_device_id = pdev->subsystem_device;
2884         board_id = (((__u32) (subsystem_device_id << 16) & 0xffff0000) |
2885                     subsystem_vendor_id);
2886
2887 #ifdef CCISS_DEBUG
2888         printk("command = %x\n", command);
2889         printk("irq = %x\n", pdev->irq);
2890         printk("board_id = %x\n", board_id);
2891 #endif                          /* CCISS_DEBUG */
2892
2893 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
2894  * else we use the IO-APIC interrupt assigned to us by system ROM.
2895  */
2896         cciss_interrupt_mode(c, pdev, board_id);
2897
2898         /*
2899          * Memory base addr is first addr , the second points to the config
2900          *   table
2901          */
2902
2903         c->paddr = pci_resource_start(pdev, 0); /* addressing mode bits already removed */
2904 #ifdef CCISS_DEBUG
2905         printk("address 0 = %x\n", c->paddr);
2906 #endif                          /* CCISS_DEBUG */
2907         c->vaddr = remap_pci_mem(c->paddr, 200);
2908
2909         /* Wait for the board to become ready.  (PCI hotplug needs this.)
2910          * We poll for up to 120 secs, once per 100ms. */
2911         for (i = 0; i < 1200; i++) {
2912                 scratchpad = readl(c->vaddr + SA5_SCRATCHPAD_OFFSET);
2913                 if (scratchpad == CCISS_FIRMWARE_READY)
2914                         break;
2915                 set_current_state(TASK_INTERRUPTIBLE);
2916                 schedule_timeout(HZ / 10);      /* wait 100ms */
2917         }
2918         if (scratchpad != CCISS_FIRMWARE_READY) {
2919                 printk(KERN_WARNING "cciss: Board not ready.  Timed out.\n");
2920                 err = -ENODEV;
2921                 goto err_out_free_res;
2922         }
2923
2924         /* get the address index number */
2925         cfg_base_addr = readl(c->vaddr + SA5_CTCFG_OFFSET);
2926         cfg_base_addr &= (__u32) 0x0000ffff;
2927 #ifdef CCISS_DEBUG
2928         printk("cfg base address = %x\n", cfg_base_addr);
2929 #endif                          /* CCISS_DEBUG */
2930         cfg_base_addr_index = find_PCI_BAR_index(pdev, cfg_base_addr);
2931 #ifdef CCISS_DEBUG
2932         printk("cfg base address index = %x\n", cfg_base_addr_index);
2933 #endif                          /* CCISS_DEBUG */
2934         if (cfg_base_addr_index == -1) {
2935                 printk(KERN_WARNING "cciss: Cannot find cfg_base_addr_index\n");
2936                 err = -ENODEV;
2937                 goto err_out_free_res;
2938         }
2939
2940         cfg_offset = readl(c->vaddr + SA5_CTMEM_OFFSET);
2941 #ifdef CCISS_DEBUG
2942         printk("cfg offset = %x\n", cfg_offset);
2943 #endif                          /* CCISS_DEBUG */
2944         c->cfgtable = remap_pci_mem(pci_resource_start(pdev,
2945                                                        cfg_base_addr_index) +
2946                                     cfg_offset, sizeof(CfgTable_struct));
2947         c->board_id = board_id;
2948
2949 #ifdef CCISS_DEBUG
2950         print_cfg_table(c->cfgtable);
2951 #endif                          /* CCISS_DEBUG */
2952
2953         for (i = 0; i < ARRAY_SIZE(products); i++) {
2954                 if (board_id == products[i].board_id) {
2955                         c->product_name = products[i].product_name;
2956                         c->access = *(products[i].access);
2957                         break;
2958                 }
2959         }
2960         if (i == ARRAY_SIZE(products)) {
2961                 printk(KERN_WARNING "cciss: Sorry, I don't know how"
2962                        " to access the Smart Array controller %08lx\n",
2963                        (unsigned long)board_id);
2964                 err = -ENODEV;
2965                 goto err_out_free_res;
2966         }
2967         if ((readb(&c->cfgtable->Signature[0]) != 'C') ||
2968             (readb(&c->cfgtable->Signature[1]) != 'I') ||
2969             (readb(&c->cfgtable->Signature[2]) != 'S') ||
2970             (readb(&c->cfgtable->Signature[3]) != 'S')) {
2971                 printk("Does not appear to be a valid CISS config table\n");
2972                 err = -ENODEV;
2973                 goto err_out_free_res;
2974         }
2975 #ifdef CONFIG_X86
2976         {
2977                 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
2978                 __u32 prefetch;
2979                 prefetch = readl(&(c->cfgtable->SCSI_Prefetch));
2980                 prefetch |= 0x100;
2981                 writel(prefetch, &(c->cfgtable->SCSI_Prefetch));
2982         }
2983 #endif
2984
2985 #ifdef CCISS_DEBUG
2986         printk("Trying to put board into Simple mode\n");
2987 #endif                          /* CCISS_DEBUG */
2988         c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
2989         /* Update the field, and then ring the doorbell */
2990         writel(CFGTBL_Trans_Simple, &(c->cfgtable->HostWrite.TransportRequest));
2991         writel(CFGTBL_ChangeReq, c->vaddr + SA5_DOORBELL);
2992
2993         /* under certain very rare conditions, this can take awhile.
2994          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
2995          * as we enter this code.) */
2996         for (i = 0; i < MAX_CONFIG_WAIT; i++) {
2997                 if (!(readl(c->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
2998                         break;
2999                 /* delay and try again */
3000                 set_current_state(TASK_INTERRUPTIBLE);
3001                 schedule_timeout(10);
3002         }
3003
3004 #ifdef CCISS_DEBUG
3005         printk(KERN_DEBUG "I counter got to %d %x\n", i,
3006                readl(c->vaddr + SA5_DOORBELL));
3007 #endif                          /* CCISS_DEBUG */
3008 #ifdef CCISS_DEBUG
3009         print_cfg_table(c->cfgtable);
3010 #endif                          /* CCISS_DEBUG */
3011
3012         if (!(readl(&(c->cfgtable->TransportActive)) & CFGTBL_Trans_Simple)) {
3013                 printk(KERN_WARNING "cciss: unable to get board into"
3014                        " simple mode\n");
3015                 err = -ENODEV;
3016                 goto err_out_free_res;
3017         }
3018         return 0;
3019
3020       err_out_free_res:
3021         pci_release_regions(pdev);
3022
3023       err_out_disable_pdev:
3024         pci_disable_device(pdev);
3025         return err;
3026 }
3027
3028 /*
3029  * Gets information about the local volumes attached to the controller.
3030  */
3031 static void cciss_getgeometry(int cntl_num)
3032 {
3033         ReportLunData_struct *ld_buff;
3034         InquiryData_struct *inq_buff;
3035         int return_code;
3036         int i;
3037         int listlength = 0;
3038         __u32 lunid = 0;
3039         int block_size;
3040         sector_t total_size;
3041
3042         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
3043         if (ld_buff == NULL) {
3044                 printk(KERN_ERR "cciss: out of memory\n");
3045                 return;
3046         }
3047         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
3048         if (inq_buff == NULL) {
3049                 printk(KERN_ERR "cciss: out of memory\n");
3050                 kfree(ld_buff);
3051                 return;
3052         }
3053         /* Get the firmware version */
3054         return_code = sendcmd(CISS_INQUIRY, cntl_num, inq_buff,
3055                               sizeof(InquiryData_struct), 0, 0, 0, NULL,
3056                               TYPE_CMD);
3057         if (return_code == IO_OK) {
3058                 hba[cntl_num]->firm_ver[0] = inq_buff->data_byte[32];
3059                 hba[cntl_num]->firm_ver[1] = inq_buff->data_byte[33];
3060                 hba[cntl_num]->firm_ver[2] = inq_buff->data_byte[34];
3061                 hba[cntl_num]->firm_ver[3] = inq_buff->data_byte[35];
3062         } else {                /* send command failed */
3063
3064                 printk(KERN_WARNING "cciss: unable to determine firmware"
3065                        " version of controller\n");
3066         }
3067         /* Get the number of logical volumes */
3068         return_code = sendcmd(CISS_REPORT_LOG, cntl_num, ld_buff,
3069                               sizeof(ReportLunData_struct), 0, 0, 0, NULL,
3070                               TYPE_CMD);
3071
3072         if (return_code == IO_OK) {
3073 #ifdef CCISS_DEBUG
3074                 printk("LUN Data\n--------------------------\n");
3075 #endif                          /* CCISS_DEBUG */
3076
3077                 listlength |=
3078                     (0xff & (unsigned int)(ld_buff->LUNListLength[0])) << 24;
3079                 listlength |=
3080                     (0xff & (unsigned int)(ld_buff->LUNListLength[1])) << 16;
3081                 listlength |=
3082                     (0xff & (unsigned int)(ld_buff->LUNListLength[2])) << 8;
3083                 listlength |= 0xff & (unsigned int)(ld_buff->LUNListLength[3]);
3084         } else {                /* reading number of logical volumes failed */
3085
3086                 printk(KERN_WARNING "cciss: report logical volume"
3087                        " command failed\n");
3088                 listlength = 0;
3089         }
3090         hba[cntl_num]->num_luns = listlength / 8;       // 8 bytes pre entry
3091         if (hba[cntl_num]->num_luns > CISS_MAX_LUN) {
3092                 printk(KERN_ERR
3093                        "ciss:  only %d number of logical volumes supported\n",
3094                        CISS_MAX_LUN);
3095                 hba[cntl_num]->num_luns = CISS_MAX_LUN;
3096         }
3097 #ifdef CCISS_DEBUG
3098         printk(KERN_DEBUG "Length = %x %x %x %x = %d\n",
3099                ld_buff->LUNListLength[0], ld_buff->LUNListLength[1],
3100                ld_buff->LUNListLength[2], ld_buff->LUNListLength[3],
3101                hba[cntl_num]->num_luns);
3102 #endif                          /* CCISS_DEBUG */
3103
3104         hba[cntl_num]->highest_lun = hba[cntl_num]->num_luns - 1;
3105         for (i = 0; i < CISS_MAX_LUN; i++) {
3106                 if (i < hba[cntl_num]->num_luns) {
3107                         lunid = (0xff & (unsigned int)(ld_buff->LUN[i][3]))
3108                             << 24;
3109                         lunid |= (0xff & (unsigned int)(ld_buff->LUN[i][2]))
3110                             << 16;
3111                         lunid |= (0xff & (unsigned int)(ld_buff->LUN[i][1]))
3112                             << 8;
3113                         lunid |= 0xff & (unsigned int)(ld_buff->LUN[i][0]);
3114
3115                         hba[cntl_num]->drv[i].LunID = lunid;
3116
3117 #ifdef CCISS_DEBUG
3118                         printk(KERN_DEBUG "LUN[%d]:  %x %x %x %x = %x\n", i,
3119                                ld_buff->LUN[i][0], ld_buff->LUN[i][1],
3120                                ld_buff->LUN[i][2], ld_buff->LUN[i][3],
3121                                hba[cntl_num]->drv[i].LunID);
3122 #endif                          /* CCISS_DEBUG */
3123
3124                 /* testing to see if 16-byte CDBs are already being used */
3125                 if(hba[cntl_num]->cciss_read == CCISS_READ_16) {
3126                         cciss_read_capacity_16(cntl_num, i, 0,
3127                                             &total_size, &block_size);
3128                         goto geo_inq;
3129                 }
3130                 cciss_read_capacity(cntl_num, i, 0, &total_size, &block_size);
3131
3132                 /* total_size = last LBA + 1 */
3133                 if(total_size == (__u32) 0) {
3134                         cciss_read_capacity_16(cntl_num, i, 0,
3135                         &total_size, &block_size);
3136                         hba[cntl_num]->cciss_read = CCISS_READ_16;
3137                         hba[cntl_num]->cciss_write = CCISS_WRITE_16;
3138                 } else {
3139                         hba[cntl_num]->cciss_read = CCISS_READ_10;
3140                         hba[cntl_num]->cciss_write = CCISS_WRITE_10;
3141                 }
3142 geo_inq:
3143                         cciss_geometry_inquiry(cntl_num, i, 0, total_size,
3144                                                block_size, inq_buff,
3145                                                &hba[cntl_num]->drv[i]);
3146                 } else {
3147                         /* initialize raid_level to indicate a free space */
3148                         hba[cntl_num]->drv[i].raid_level = -1;
3149                 }
3150         }
3151         kfree(ld_buff);
3152         kfree(inq_buff);
3153 }
3154
3155 /* Function to find the first free pointer into our hba[] array */
3156 /* Returns -1 if no free entries are left.  */
3157 static int alloc_cciss_hba(void)
3158 {
3159         struct gendisk *disk[NWD];
3160         int i, n;
3161         for (n = 0; n < NWD; n++) {
3162                 disk[n] = alloc_disk(1 << NWD_SHIFT);
3163                 if (!disk[n])
3164                         goto out;
3165         }
3166
3167         for (i = 0; i < MAX_CTLR; i++) {
3168                 if (!hba[i]) {
3169                         ctlr_info_t *p;
3170                         p = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
3171                         if (!p)
3172                                 goto Enomem;
3173                         for (n = 0; n < NWD; n++)
3174                                 p->gendisk[n] = disk[n];
3175                         hba[i] = p;
3176                         return i;
3177                 }
3178         }
3179         printk(KERN_WARNING "cciss: This driver supports a maximum"
3180                " of %d controllers.\n", MAX_CTLR);
3181         goto out;
3182       Enomem:
3183         printk(KERN_ERR "cciss: out of memory.\n");
3184       out:
3185         while (n--)
3186                 put_disk(disk[n]);
3187         return -1;
3188 }
3189
3190 static void free_hba(int i)
3191 {
3192         ctlr_info_t *p = hba[i];
3193         int n;
3194
3195         hba[i] = NULL;
3196         for (n = 0; n < NWD; n++)
3197                 put_disk(p->gendisk[n]);
3198         kfree(p);
3199 }
3200
3201 /*
3202  *  This is it.  Find all the controllers and register them.  I really hate
3203  *  stealing all these major device numbers.
3204  *  returns the number of block devices registered.
3205  */
3206 static int __devinit cciss_init_one(struct pci_dev *pdev,
3207                                     const struct pci_device_id *ent)
3208 {
3209         request_queue_t *q;
3210         int i;
3211         int j;
3212         int rc;
3213         int dac;
3214
3215         i = alloc_cciss_hba();
3216         if (i < 0)
3217                 return -1;
3218
3219         hba[i]->busy_initializing = 1;
3220
3221         if (cciss_pci_init(hba[i], pdev) != 0)
3222                 goto clean1;
3223
3224         sprintf(hba[i]->devname, "cciss%d", i);
3225         hba[i]->ctlr = i;
3226         hba[i]->pdev = pdev;
3227
3228         /* configure PCI DMA stuff */
3229         if (!pci_set_dma_mask(pdev, DMA_64BIT_MASK))
3230                 dac = 1;
3231         else if (!pci_set_dma_mask(pdev, DMA_32BIT_MASK))
3232                 dac = 0;
3233         else {
3234                 printk(KERN_ERR "cciss: no suitable DMA available\n");
3235                 goto clean1;
3236         }
3237
3238         /*
3239          * register with the major number, or get a dynamic major number
3240          * by passing 0 as argument.  This is done for greater than
3241          * 8 controller support.
3242          */
3243         if (i < MAX_CTLR_ORIG)
3244                 hba[i]->major = COMPAQ_CISS_MAJOR + i;
3245         rc = register_blkdev(hba[i]->major, hba[i]->devname);
3246         if (rc == -EBUSY || rc == -EINVAL) {
3247                 printk(KERN_ERR
3248                        "cciss:  Unable to get major number %d for %s "
3249                        "on hba %d\n", hba[i]->major, hba[i]->devname, i);
3250                 goto clean1;
3251         } else {
3252                 if (i >= MAX_CTLR_ORIG)
3253                         hba[i]->major = rc;
3254         }
3255
3256         /* make sure the board interrupts are off */
3257         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_OFF);
3258         if (request_irq(hba[i]->intr[SIMPLE_MODE_INT], do_cciss_intr,
3259                         IRQF_DISABLED | IRQF_SHARED, hba[i]->devname, hba[i])) {
3260                 printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
3261                        hba[i]->intr[SIMPLE_MODE_INT], hba[i]->devname);
3262                 goto clean2;
3263         }
3264
3265         printk(KERN_INFO "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
3266                hba[i]->devname, pdev->device, pci_name(pdev),
3267                hba[i]->intr[SIMPLE_MODE_INT], dac ? "" : " not");
3268
3269         hba[i]->cmd_pool_bits =
3270             kmalloc(((NR_CMDS + BITS_PER_LONG -
3271                       1) / BITS_PER_LONG) * sizeof(unsigned long), GFP_KERNEL);
3272         hba[i]->cmd_pool = (CommandList_struct *)
3273             pci_alloc_consistent(hba[i]->pdev,
3274                     NR_CMDS * sizeof(CommandList_struct),
3275                     &(hba[i]->cmd_pool_dhandle));
3276         hba[i]->errinfo_pool = (ErrorInfo_struct *)
3277             pci_alloc_consistent(hba[i]->pdev,
3278                     NR_CMDS * sizeof(ErrorInfo_struct),
3279                     &(hba[i]->errinfo_pool_dhandle));
3280         if ((hba[i]->cmd_pool_bits == NULL)
3281             || (hba[i]->cmd_pool == NULL)
3282             || (hba[i]->errinfo_pool == NULL)) {
3283                 printk(KERN_ERR "cciss: out of memory");
3284                 goto clean4;
3285         }
3286 #ifdef CONFIG_CISS_SCSI_TAPE
3287         hba[i]->scsi_rejects.complete =
3288             kmalloc(sizeof(hba[i]->scsi_rejects.complete[0]) *
3289                     (NR_CMDS + 5), GFP_KERNEL);
3290         if (hba[i]->scsi_rejects.complete == NULL) {
3291                 printk(KERN_ERR "cciss: out of memory");
3292                 goto clean4;
3293         }
3294 #endif
3295         spin_lock_init(&hba[i]->lock);
3296
3297         /* Initialize the pdev driver private data.
3298            have it point to hba[i].  */
3299         pci_set_drvdata(pdev, hba[i]);
3300         /* command and error info recs zeroed out before
3301            they are used */
3302         memset(hba[i]->cmd_pool_bits, 0,
3303                ((NR_CMDS + BITS_PER_LONG -
3304                  1) / BITS_PER_LONG) * sizeof(unsigned long));
3305
3306 #ifdef CCISS_DEBUG
3307         printk(KERN_DEBUG "Scanning for drives on controller cciss%d\n", i);
3308 #endif                          /* CCISS_DEBUG */
3309
3310         cciss_getgeometry(i);
3311
3312         cciss_scsi_setup(i);
3313
3314         /* Turn the interrupts on so we can service requests */
3315         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_ON);
3316
3317         cciss_procinit(i);
3318         hba[i]->busy_initializing = 0;
3319
3320         for (j = 0; j < NWD; j++) {     /* mfm */
3321                 drive_info_struct *drv = &(hba[i]->drv[j]);
3322                 struct gendisk *disk = hba[i]->gendisk[j];
3323
3324                 q = blk_init_queue(do_cciss_request, &hba[i]->lock);
3325                 if (!q) {
3326                         printk(KERN_ERR
3327                                "cciss:  unable to allocate queue for disk %d\n",
3328                                j);
3329                         break;
3330                 }
3331                 drv->queue = q;
3332
3333                 q->backing_dev_info.ra_pages = READ_AHEAD;
3334                 blk_queue_bounce_limit(q, hba[i]->pdev->dma_mask);
3335
3336                 /* This is a hardware imposed limit. */
3337                 blk_queue_max_hw_segments(q, MAXSGENTRIES);
3338
3339                 /* This is a limit in the driver and could be eliminated. */
3340                 blk_queue_max_phys_segments(q, MAXSGENTRIES);
3341
3342                 blk_queue_max_sectors(q, 512);
3343
3344                 blk_queue_softirq_done(q, cciss_softirq_done);
3345
3346                 q->queuedata = hba[i];
3347                 sprintf(disk->disk_name, "cciss/c%dd%d", i, j);
3348                 disk->major = hba[i]->major;
3349                 disk->first_minor = j << NWD_SHIFT;
3350                 disk->fops = &cciss_fops;
3351                 disk->queue = q;
3352                 disk->private_data = drv;
3353                 disk->driverfs_dev = &pdev->dev;
3354                 /* we must register the controller even if no disks exist */
3355                 /* this is for the online array utilities */
3356                 if (!drv->heads && j)
3357                         continue;
3358                 blk_queue_hardsect_size(q, drv->block_size);
3359                 set_capacity(disk, drv->nr_blocks);
3360                 add_disk(disk);
3361         }
3362
3363         return 1;
3364
3365       clean4:
3366 #ifdef CONFIG_CISS_SCSI_TAPE
3367         kfree(hba[i]->scsi_rejects.complete);
3368 #endif
3369         kfree(hba[i]->cmd_pool_bits);
3370         if (hba[i]->cmd_pool)
3371                 pci_free_consistent(hba[i]->pdev,
3372                                     NR_CMDS * sizeof(CommandList_struct),
3373                                     hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
3374         if (hba[i]->errinfo_pool)
3375                 pci_free_consistent(hba[i]->pdev,
3376                                     NR_CMDS * sizeof(ErrorInfo_struct),
3377                                     hba[i]->errinfo_pool,
3378                                     hba[i]->errinfo_pool_dhandle);
3379         free_irq(hba[i]->intr[SIMPLE_MODE_INT], hba[i]);
3380       clean2:
3381         unregister_blkdev(hba[i]->major, hba[i]->devname);
3382       clean1:
3383         hba[i]->busy_initializing = 0;
3384         free_hba(i);
3385         return -1;
3386 }
3387
3388 static void __devexit cciss_remove_one(struct pci_dev *pdev)
3389 {
3390         ctlr_info_t *tmp_ptr;
3391         int i, j;
3392         char flush_buf[4];
3393         int return_code;
3394
3395         if (pci_get_drvdata(pdev) == NULL) {
3396                 printk(KERN_ERR "cciss: Unable to remove device \n");
3397                 return;
3398         }
3399         tmp_ptr = pci_get_drvdata(pdev);
3400         i = tmp_ptr->ctlr;
3401         if (hba[i] == NULL) {
3402                 printk(KERN_ERR "cciss: device appears to "
3403                        "already be removed \n");
3404                 return;
3405         }
3406         /* Turn board interrupts off  and send the flush cache command */
3407         /* sendcmd will turn off interrupt, and send the flush...
3408          * To write all data in the battery backed cache to disks */
3409         memset(flush_buf, 0, 4);
3410         return_code = sendcmd(CCISS_CACHE_FLUSH, i, flush_buf, 4, 0, 0, 0, NULL,
3411                               TYPE_CMD);
3412         if (return_code != IO_OK) {
3413                 printk(KERN_WARNING "Error Flushing cache on controller %d\n",
3414                        i);
3415         }
3416         free_irq(hba[i]->intr[2], hba[i]);
3417
3418 #ifdef CONFIG_PCI_MSI
3419         if (hba[i]->msix_vector)
3420                 pci_disable_msix(hba[i]->pdev);
3421         else if (hba[i]->msi_vector)
3422                 pci_disable_msi(hba[i]->pdev);
3423 #endif                          /* CONFIG_PCI_MSI */
3424
3425         iounmap(hba[i]->vaddr);
3426         cciss_unregister_scsi(i);       /* unhook from SCSI subsystem */
3427         unregister_blkdev(hba[i]->major, hba[i]->devname);
3428         remove_proc_entry(hba[i]->devname, proc_cciss);
3429
3430         /* remove it from the disk list */
3431         for (j = 0; j < NWD; j++) {
3432                 struct gendisk *disk = hba[i]->gendisk[j];
3433                 if (disk) {
3434                         request_queue_t *q = disk->queue;
3435
3436                         if (disk->flags & GENHD_FL_UP)
3437                                 del_gendisk(disk);
3438                         if (q)
3439                                 blk_cleanup_queue(q);
3440                 }
3441         }
3442
3443         pci_free_consistent(hba[i]->pdev, NR_CMDS * sizeof(CommandList_struct),
3444                             hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
3445         pci_free_consistent(hba[i]->pdev, NR_CMDS * sizeof(ErrorInfo_struct),
3446                             hba[i]->errinfo_pool, hba[i]->errinfo_pool_dhandle);
3447         kfree(hba[i]->cmd_pool_bits);
3448 #ifdef CONFIG_CISS_SCSI_TAPE
3449         kfree(hba[i]->scsi_rejects.complete);
3450 #endif
3451         pci_release_regions(pdev);
3452         pci_disable_device(pdev);
3453         pci_set_drvdata(pdev, NULL);
3454         free_hba(i);
3455 }
3456
3457 static struct pci_driver cciss_pci_driver = {
3458         .name = "cciss",
3459         .probe = cciss_init_one,
3460         .remove = __devexit_p(cciss_remove_one),
3461         .id_table = cciss_pci_device_id,        /* id_table */
3462 };
3463
3464 /*
3465  *  This is it.  Register the PCI driver information for the cards we control
3466  *  the OS will call our registered routines when it finds one of our cards.
3467  */
3468 static int __init cciss_init(void)
3469 {
3470         printk(KERN_INFO DRIVER_NAME "\n");
3471
3472         /* Register for our PCI devices */
3473         return pci_register_driver(&cciss_pci_driver);
3474 }
3475
3476 static void __exit cciss_cleanup(void)
3477 {
3478         int i;
3479
3480         pci_unregister_driver(&cciss_pci_driver);
3481         /* double check that all controller entrys have been removed */
3482         for (i = 0; i < MAX_CTLR; i++) {
3483                 if (hba[i] != NULL) {
3484                         printk(KERN_WARNING "cciss: had to remove"
3485                                " controller %d\n", i);
3486                         cciss_remove_one(hba[i]->pdev);
3487                 }
3488         }
3489         remove_proc_entry("cciss", proc_root_driver);
3490 }
3491
3492 static void fail_all_cmds(unsigned long ctlr)
3493 {
3494         /* If we get here, the board is apparently dead. */
3495         ctlr_info_t *h = hba[ctlr];
3496         CommandList_struct *c;
3497         unsigned long flags;
3498
3499         printk(KERN_WARNING "cciss%d: controller not responding.\n", h->ctlr);
3500         h->alive = 0;           /* the controller apparently died... */
3501
3502         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
3503
3504         pci_disable_device(h->pdev);    /* Make sure it is really dead. */
3505
3506         /* move everything off the request queue onto the completed queue */
3507         while ((c = h->reqQ) != NULL) {
3508                 removeQ(&(h->reqQ), c);
3509                 h->Qdepth--;
3510                 addQ(&(h->cmpQ), c);
3511         }
3512
3513         /* Now, fail everything on the completed queue with a HW error */
3514         while ((c = h->cmpQ) != NULL) {
3515                 removeQ(&h->cmpQ, c);
3516                 c->err_info->CommandStatus = CMD_HARDWARE_ERR;
3517                 if (c->cmd_type == CMD_RWREQ) {
3518                         complete_command(h, c, 0);
3519                 } else if (c->cmd_type == CMD_IOCTL_PEND)
3520                         complete(c->waiting);
3521 #ifdef CONFIG_CISS_SCSI_TAPE
3522                 else if (c->cmd_type == CMD_SCSI)
3523                         complete_scsi_command(c, 0, 0);
3524 #endif
3525         }
3526         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
3527         return;
3528 }
3529
3530 module_init(cciss_init);
3531 module_exit(cciss_cleanup);