[PATCH] cciss: version update, new hw
[linux-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/config.h>       /* CONFIG_PROC_FS */
24 #include <linux/module.h>
25 #include <linux/interrupt.h>
26 #include <linux/types.h>
27 #include <linux/pci.h>
28 #include <linux/kernel.h>
29 #include <linux/slab.h>
30 #include <linux/delay.h>
31 #include <linux/major.h>
32 #include <linux/fs.h>
33 #include <linux/bio.h>
34 #include <linux/blkpg.h>
35 #include <linux/timer.h>
36 #include <linux/proc_fs.h>
37 #include <linux/init.h>
38 #include <linux/hdreg.h>
39 #include <linux/spinlock.h>
40 #include <linux/compat.h>
41 #include <linux/blktrace_api.h>
42 #include <asm/uaccess.h>
43 #include <asm/io.h>
44
45 #include <linux/dma-mapping.h>
46 #include <linux/blkdev.h>
47 #include <linux/genhd.h>
48 #include <linux/completion.h>
49
50 #define CCISS_DRIVER_VERSION(maj,min,submin) ((maj<<16)|(min<<8)|(submin))
51 #define DRIVER_NAME "HP CISS Driver (v 3.6.10)"
52 #define DRIVER_VERSION CCISS_DRIVER_VERSION(3,6,10)
53
54 /* Embedded module documentation macros - see modules.h */
55 MODULE_AUTHOR("Hewlett-Packard Company");
56 MODULE_DESCRIPTION("Driver for HP Controller SA5xxx SA6xxx version 3.6.10");
57 MODULE_SUPPORTED_DEVICE("HP SA5i SA5i+ SA532 SA5300 SA5312 SA641 SA642 SA6400"
58                         " SA6i P600 P800 P400 P400i E200 E200i E500");
59 MODULE_LICENSE("GPL");
60
61 #include "cciss_cmd.h"
62 #include "cciss.h"
63 #include <linux/cciss_ioctl.h>
64
65 /* define the PCI info for the cards we can control */
66 static const struct pci_device_id cciss_pci_device_id[] = {
67         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISS,  0x0E11, 0x4070},
68         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4080},
69         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4082},
70         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4083},
71         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x4091},
72         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409A},
73         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409B},
74         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409C},
75         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409D},
76         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSA,     0x103C, 0x3225},
77         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3223},
78         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3234},
79         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3235},
80         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3211},
81         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3212},
82         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3213},
83         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3214},
84         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3215},
85         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3233},
86         {0,}
87 };
88
89 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
90
91 /*  board_id = Subsystem Device ID & Vendor ID
92  *  product = Marketing Name for the board
93  *  access = Address of the struct of function pointers
94  */
95 static struct board_type products[] = {
96         {0x40700E11, "Smart Array 5300", &SA5_access},
97         {0x40800E11, "Smart Array 5i", &SA5B_access},
98         {0x40820E11, "Smart Array 532", &SA5B_access},
99         {0x40830E11, "Smart Array 5312", &SA5B_access},
100         {0x409A0E11, "Smart Array 641", &SA5_access},
101         {0x409B0E11, "Smart Array 642", &SA5_access},
102         {0x409C0E11, "Smart Array 6400", &SA5_access},
103         {0x409D0E11, "Smart Array 6400 EM", &SA5_access},
104         {0x40910E11, "Smart Array 6i", &SA5_access},
105         {0x3225103C, "Smart Array P600", &SA5_access},
106         {0x3223103C, "Smart Array P800", &SA5_access},
107         {0x3234103C, "Smart Array P400", &SA5_access},
108         {0x3235103C, "Smart Array P400i", &SA5_access},
109         {0x3211103C, "Smart Array E200i", &SA5_access},
110         {0x3212103C, "Smart Array E200", &SA5_access},
111         {0x3213103C, "Smart Array E200i", &SA5_access},
112         {0x3214103C, "Smart Array E200i", &SA5_access},
113         {0x3215103C, "Smart Array E200i", &SA5_access},
114         {0x3233103C, "Smart Array E500", &SA5_access},
115 };
116
117 /* How long to wait (in milliseconds) for board to go into simple mode */
118 #define MAX_CONFIG_WAIT 30000
119 #define MAX_IOCTL_CONFIG_WAIT 1000
120
121 /*define how many times we will try a command because of bus resets */
122 #define MAX_CMD_RETRIES 3
123
124 #define READ_AHEAD       1024
125 #define NR_CMDS          384    /* #commands that can be outstanding */
126 #define MAX_CTLR        32
127
128 /* Originally cciss driver only supports 8 major numbers */
129 #define MAX_CTLR_ORIG   8
130
131 static ctlr_info_t *hba[MAX_CTLR];
132
133 static void do_cciss_request(request_queue_t *q);
134 static irqreturn_t do_cciss_intr(int irq, void *dev_id, struct pt_regs *regs);
135 static int cciss_open(struct inode *inode, struct file *filep);
136 static int cciss_release(struct inode *inode, struct file *filep);
137 static int cciss_ioctl(struct inode *inode, struct file *filep,
138                        unsigned int cmd, unsigned long arg);
139 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
140
141 static int revalidate_allvol(ctlr_info_t *host);
142 static int cciss_revalidate(struct gendisk *disk);
143 static int rebuild_lun_table(ctlr_info_t *h, struct gendisk *del_disk);
144 static int deregister_disk(struct gendisk *disk, drive_info_struct *drv,
145                            int clear_all);
146
147 static void cciss_read_capacity(int ctlr, int logvol, ReadCapdata_struct *buf,
148                                 int withirq, unsigned int *total_size,
149                                 unsigned int *block_size);
150 static void cciss_geometry_inquiry(int ctlr, int logvol, int withirq,
151                                    unsigned int total_size,
152                                    unsigned int block_size,
153                                    InquiryData_struct *inq_buff,
154                                    drive_info_struct *drv);
155 static void cciss_getgeometry(int cntl_num);
156 static void __devinit cciss_interrupt_mode(ctlr_info_t *, struct pci_dev *,
157                                            __u32);
158 static void start_io(ctlr_info_t *h);
159 static int sendcmd(__u8 cmd, int ctlr, void *buff, size_t size,
160                    unsigned int use_unit_num, unsigned int log_unit,
161                    __u8 page_code, unsigned char *scsi3addr, int cmd_type);
162 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
163                            unsigned int use_unit_num, unsigned int log_unit,
164                            __u8 page_code, int cmd_type);
165
166 static void fail_all_cmds(unsigned long ctlr);
167
168 #ifdef CONFIG_PROC_FS
169 static int cciss_proc_get_info(char *buffer, char **start, off_t offset,
170                                int length, int *eof, void *data);
171 static void cciss_procinit(int i);
172 #else
173 static void cciss_procinit(int i)
174 {
175 }
176 #endif                          /* CONFIG_PROC_FS */
177
178 #ifdef CONFIG_COMPAT
179 static long cciss_compat_ioctl(struct file *f, unsigned cmd, unsigned long arg);
180 #endif
181
182 static struct block_device_operations cciss_fops = {
183         .owner = THIS_MODULE,
184         .open = cciss_open,
185         .release = cciss_release,
186         .ioctl = cciss_ioctl,
187         .getgeo = cciss_getgeo,
188 #ifdef CONFIG_COMPAT
189         .compat_ioctl = cciss_compat_ioctl,
190 #endif
191         .revalidate_disk = cciss_revalidate,
192 };
193
194 /*
195  * Enqueuing and dequeuing functions for cmdlists.
196  */
197 static inline void addQ(CommandList_struct **Qptr, CommandList_struct *c)
198 {
199         if (*Qptr == NULL) {
200                 *Qptr = c;
201                 c->next = c->prev = c;
202         } else {
203                 c->prev = (*Qptr)->prev;
204                 c->next = (*Qptr);
205                 (*Qptr)->prev->next = c;
206                 (*Qptr)->prev = c;
207         }
208 }
209
210 static inline CommandList_struct *removeQ(CommandList_struct **Qptr,
211                                           CommandList_struct *c)
212 {
213         if (c && c->next != c) {
214                 if (*Qptr == c)
215                         *Qptr = c->next;
216                 c->prev->next = c->next;
217                 c->next->prev = c->prev;
218         } else {
219                 *Qptr = NULL;
220         }
221         return c;
222 }
223
224 #include "cciss_scsi.c"         /* For SCSI tape support */
225
226 #ifdef CONFIG_PROC_FS
227
228 /*
229  * Report information about this controller.
230  */
231 #define ENG_GIG 1000000000
232 #define ENG_GIG_FACTOR (ENG_GIG/512)
233 #define RAID_UNKNOWN 6
234 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
235         "UNKNOWN"
236 };
237
238 static struct proc_dir_entry *proc_cciss;
239
240 static int cciss_proc_get_info(char *buffer, char **start, off_t offset,
241                                int length, int *eof, void *data)
242 {
243         off_t pos = 0;
244         off_t len = 0;
245         int size, i, ctlr;
246         ctlr_info_t *h = (ctlr_info_t *) data;
247         drive_info_struct *drv;
248         unsigned long flags;
249         sector_t vol_sz, vol_sz_frac;
250
251         ctlr = h->ctlr;
252
253         /* prevent displaying bogus info during configuration
254          * or deconfiguration of a logical volume
255          */
256         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
257         if (h->busy_configuring) {
258                 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
259                 return -EBUSY;
260         }
261         h->busy_configuring = 1;
262         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
263
264         size = sprintf(buffer, "%s: HP %s Controller\n"
265                        "Board ID: 0x%08lx\n"
266                        "Firmware Version: %c%c%c%c\n"
267                        "IRQ: %d\n"
268                        "Logical drives: %d\n"
269                        "Current Q depth: %d\n"
270                        "Current # commands on controller: %d\n"
271                        "Max Q depth since init: %d\n"
272                        "Max # commands on controller since init: %d\n"
273                        "Max SG entries since init: %d\n\n",
274                        h->devname,
275                        h->product_name,
276                        (unsigned long)h->board_id,
277                        h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
278                        h->firm_ver[3], (unsigned int)h->intr[SIMPLE_MODE_INT],
279                        h->num_luns, h->Qdepth, h->commands_outstanding,
280                        h->maxQsinceinit, h->max_outstanding, h->maxSG);
281
282         pos += size;
283         len += size;
284         cciss_proc_tape_report(ctlr, buffer, &pos, &len);
285         for (i = 0; i <= h->highest_lun; i++) {
286
287                 drv = &h->drv[i];
288                 if (drv->heads == 0)
289                         continue;
290
291                 vol_sz = drv->nr_blocks;
292                 vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
293                 vol_sz_frac *= 100;
294                 sector_div(vol_sz_frac, ENG_GIG_FACTOR);
295
296                 if (drv->raid_level > 5)
297                         drv->raid_level = RAID_UNKNOWN;
298                 size = sprintf(buffer + len, "cciss/c%dd%d:"
299                                "\t%4u.%02uGB\tRAID %s\n",
300                                ctlr, i, (int)vol_sz, (int)vol_sz_frac,
301                                raid_label[drv->raid_level]);
302                 pos += size;
303                 len += size;
304         }
305
306         *eof = 1;
307         *start = buffer + offset;
308         len -= offset;
309         if (len > length)
310                 len = length;
311         h->busy_configuring = 0;
312         return len;
313 }
314
315 static int
316 cciss_proc_write(struct file *file, const char __user *buffer,
317                  unsigned long count, void *data)
318 {
319         unsigned char cmd[80];
320         int len;
321 #ifdef CONFIG_CISS_SCSI_TAPE
322         ctlr_info_t *h = (ctlr_info_t *) data;
323         int rc;
324 #endif
325
326         if (count > sizeof(cmd) - 1)
327                 return -EINVAL;
328         if (copy_from_user(cmd, buffer, count))
329                 return -EFAULT;
330         cmd[count] = '\0';
331         len = strlen(cmd);      // above 3 lines ensure safety
332         if (len && cmd[len - 1] == '\n')
333                 cmd[--len] = '\0';
334 #       ifdef CONFIG_CISS_SCSI_TAPE
335         if (strcmp("engage scsi", cmd) == 0) {
336                 rc = cciss_engage_scsi(h->ctlr);
337                 if (rc != 0)
338                         return -rc;
339                 return count;
340         }
341         /* might be nice to have "disengage" too, but it's not
342            safely possible. (only 1 module use count, lock issues.) */
343 #       endif
344         return -EINVAL;
345 }
346
347 /*
348  * Get us a file in /proc/cciss that says something about each controller.
349  * Create /proc/cciss if it doesn't exist yet.
350  */
351 static void __devinit cciss_procinit(int i)
352 {
353         struct proc_dir_entry *pde;
354
355         if (proc_cciss == NULL) {
356                 proc_cciss = proc_mkdir("cciss", proc_root_driver);
357                 if (!proc_cciss)
358                         return;
359         }
360
361         pde = create_proc_read_entry(hba[i]->devname,
362                                      S_IWUSR | S_IRUSR | S_IRGRP | S_IROTH,
363                                      proc_cciss, cciss_proc_get_info, hba[i]);
364         pde->write_proc = cciss_proc_write;
365 }
366 #endif                          /* CONFIG_PROC_FS */
367
368 /*
369  * For operations that cannot sleep, a command block is allocated at init,
370  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
371  * which ones are free or in use.  For operations that can wait for kmalloc
372  * to possible sleep, this routine can be called with get_from_pool set to 0.
373  * cmd_free() MUST be called with a got_from_pool set to 0 if cmd_alloc was.
374  */
375 static CommandList_struct *cmd_alloc(ctlr_info_t *h, int get_from_pool)
376 {
377         CommandList_struct *c;
378         int i;
379         u64bit temp64;
380         dma_addr_t cmd_dma_handle, err_dma_handle;
381
382         if (!get_from_pool) {
383                 c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
384                         sizeof(CommandList_struct), &cmd_dma_handle);
385                 if (c == NULL)
386                         return NULL;
387                 memset(c, 0, sizeof(CommandList_struct));
388
389                 c->cmdindex = -1;
390
391                 c->err_info = (ErrorInfo_struct *)
392                     pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
393                             &err_dma_handle);
394
395                 if (c->err_info == NULL) {
396                         pci_free_consistent(h->pdev,
397                                 sizeof(CommandList_struct), c, cmd_dma_handle);
398                         return NULL;
399                 }
400                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
401         } else {                /* get it out of the controllers pool */
402
403                 do {
404                         i = find_first_zero_bit(h->cmd_pool_bits, NR_CMDS);
405                         if (i == NR_CMDS)
406                                 return NULL;
407                 } while (test_and_set_bit
408                          (i & (BITS_PER_LONG - 1),
409                           h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
410 #ifdef CCISS_DEBUG
411                 printk(KERN_DEBUG "cciss: using command buffer %d\n", i);
412 #endif
413                 c = h->cmd_pool + i;
414                 memset(c, 0, sizeof(CommandList_struct));
415                 cmd_dma_handle = h->cmd_pool_dhandle
416                     + i * sizeof(CommandList_struct);
417                 c->err_info = h->errinfo_pool + i;
418                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
419                 err_dma_handle = h->errinfo_pool_dhandle
420                     + i * sizeof(ErrorInfo_struct);
421                 h->nr_allocs++;
422
423                 c->cmdindex = i;
424         }
425
426         c->busaddr = (__u32) cmd_dma_handle;
427         temp64.val = (__u64) err_dma_handle;
428         c->ErrDesc.Addr.lower = temp64.val32.lower;
429         c->ErrDesc.Addr.upper = temp64.val32.upper;
430         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
431
432         c->ctlr = h->ctlr;
433         return c;
434 }
435
436 /*
437  * Frees a command block that was previously allocated with cmd_alloc().
438  */
439 static void cmd_free(ctlr_info_t *h, CommandList_struct *c, int got_from_pool)
440 {
441         int i;
442         u64bit temp64;
443
444         if (!got_from_pool) {
445                 temp64.val32.lower = c->ErrDesc.Addr.lower;
446                 temp64.val32.upper = c->ErrDesc.Addr.upper;
447                 pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
448                                     c->err_info, (dma_addr_t) temp64.val);
449                 pci_free_consistent(h->pdev, sizeof(CommandList_struct),
450                                     c, (dma_addr_t) c->busaddr);
451         } else {
452                 i = c - h->cmd_pool;
453                 clear_bit(i & (BITS_PER_LONG - 1),
454                           h->cmd_pool_bits + (i / BITS_PER_LONG));
455                 h->nr_frees++;
456         }
457 }
458
459 static inline ctlr_info_t *get_host(struct gendisk *disk)
460 {
461         return disk->queue->queuedata;
462 }
463
464 static inline drive_info_struct *get_drv(struct gendisk *disk)
465 {
466         return disk->private_data;
467 }
468
469 /*
470  * Open.  Make sure the device is really there.
471  */
472 static int cciss_open(struct inode *inode, struct file *filep)
473 {
474         ctlr_info_t *host = get_host(inode->i_bdev->bd_disk);
475         drive_info_struct *drv = get_drv(inode->i_bdev->bd_disk);
476
477 #ifdef CCISS_DEBUG
478         printk(KERN_DEBUG "cciss_open %s\n", inode->i_bdev->bd_disk->disk_name);
479 #endif                          /* CCISS_DEBUG */
480
481         if (host->busy_initializing || drv->busy_configuring)
482                 return -EBUSY;
483         /*
484          * Root is allowed to open raw volume zero even if it's not configured
485          * so array config can still work. Root is also allowed to open any
486          * volume that has a LUN ID, so it can issue IOCTL to reread the
487          * disk information.  I don't think I really like this
488          * but I'm already using way to many device nodes to claim another one
489          * for "raw controller".
490          */
491         if (drv->nr_blocks == 0) {
492                 if (iminor(inode) != 0) {       /* not node 0? */
493                         /* if not node 0 make sure it is a partition = 0 */
494                         if (iminor(inode) & 0x0f) {
495                                 return -ENXIO;
496                                 /* if it is, make sure we have a LUN ID */
497                         } else if (drv->LunID == 0) {
498                                 return -ENXIO;
499                         }
500                 }
501                 if (!capable(CAP_SYS_ADMIN))
502                         return -EPERM;
503         }
504         drv->usage_count++;
505         host->usage_count++;
506         return 0;
507 }
508
509 /*
510  * Close.  Sync first.
511  */
512 static int cciss_release(struct inode *inode, struct file *filep)
513 {
514         ctlr_info_t *host = get_host(inode->i_bdev->bd_disk);
515         drive_info_struct *drv = get_drv(inode->i_bdev->bd_disk);
516
517 #ifdef CCISS_DEBUG
518         printk(KERN_DEBUG "cciss_release %s\n",
519                inode->i_bdev->bd_disk->disk_name);
520 #endif                          /* CCISS_DEBUG */
521
522         drv->usage_count--;
523         host->usage_count--;
524         return 0;
525 }
526
527 #ifdef CONFIG_COMPAT
528
529 static int do_ioctl(struct file *f, unsigned cmd, unsigned long arg)
530 {
531         int ret;
532         lock_kernel();
533         ret = cciss_ioctl(f->f_dentry->d_inode, f, cmd, arg);
534         unlock_kernel();
535         return ret;
536 }
537
538 static int cciss_ioctl32_passthru(struct file *f, unsigned cmd,
539                                   unsigned long arg);
540 static int cciss_ioctl32_big_passthru(struct file *f, unsigned cmd,
541                                       unsigned long arg);
542
543 static long cciss_compat_ioctl(struct file *f, unsigned cmd, unsigned long arg)
544 {
545         switch (cmd) {
546         case CCISS_GETPCIINFO:
547         case CCISS_GETINTINFO:
548         case CCISS_SETINTINFO:
549         case CCISS_GETNODENAME:
550         case CCISS_SETNODENAME:
551         case CCISS_GETHEARTBEAT:
552         case CCISS_GETBUSTYPES:
553         case CCISS_GETFIRMVER:
554         case CCISS_GETDRIVVER:
555         case CCISS_REVALIDVOLS:
556         case CCISS_DEREGDISK:
557         case CCISS_REGNEWDISK:
558         case CCISS_REGNEWD:
559         case CCISS_RESCANDISK:
560         case CCISS_GETLUNINFO:
561                 return do_ioctl(f, cmd, arg);
562
563         case CCISS_PASSTHRU32:
564                 return cciss_ioctl32_passthru(f, cmd, arg);
565         case CCISS_BIG_PASSTHRU32:
566                 return cciss_ioctl32_big_passthru(f, cmd, arg);
567
568         default:
569                 return -ENOIOCTLCMD;
570         }
571 }
572
573 static int cciss_ioctl32_passthru(struct file *f, unsigned cmd,
574                                   unsigned long arg)
575 {
576         IOCTL32_Command_struct __user *arg32 =
577             (IOCTL32_Command_struct __user *) arg;
578         IOCTL_Command_struct arg64;
579         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
580         int err;
581         u32 cp;
582
583         err = 0;
584         err |=
585             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
586                            sizeof(arg64.LUN_info));
587         err |=
588             copy_from_user(&arg64.Request, &arg32->Request,
589                            sizeof(arg64.Request));
590         err |=
591             copy_from_user(&arg64.error_info, &arg32->error_info,
592                            sizeof(arg64.error_info));
593         err |= get_user(arg64.buf_size, &arg32->buf_size);
594         err |= get_user(cp, &arg32->buf);
595         arg64.buf = compat_ptr(cp);
596         err |= copy_to_user(p, &arg64, sizeof(arg64));
597
598         if (err)
599                 return -EFAULT;
600
601         err = do_ioctl(f, CCISS_PASSTHRU, (unsigned long)p);
602         if (err)
603                 return err;
604         err |=
605             copy_in_user(&arg32->error_info, &p->error_info,
606                          sizeof(arg32->error_info));
607         if (err)
608                 return -EFAULT;
609         return err;
610 }
611
612 static int cciss_ioctl32_big_passthru(struct file *file, unsigned cmd,
613                                       unsigned long arg)
614 {
615         BIG_IOCTL32_Command_struct __user *arg32 =
616             (BIG_IOCTL32_Command_struct __user *) arg;
617         BIG_IOCTL_Command_struct arg64;
618         BIG_IOCTL_Command_struct __user *p =
619             compat_alloc_user_space(sizeof(arg64));
620         int err;
621         u32 cp;
622
623         err = 0;
624         err |=
625             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
626                            sizeof(arg64.LUN_info));
627         err |=
628             copy_from_user(&arg64.Request, &arg32->Request,
629                            sizeof(arg64.Request));
630         err |=
631             copy_from_user(&arg64.error_info, &arg32->error_info,
632                            sizeof(arg64.error_info));
633         err |= get_user(arg64.buf_size, &arg32->buf_size);
634         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
635         err |= get_user(cp, &arg32->buf);
636         arg64.buf = compat_ptr(cp);
637         err |= copy_to_user(p, &arg64, sizeof(arg64));
638
639         if (err)
640                 return -EFAULT;
641
642         err = do_ioctl(file, CCISS_BIG_PASSTHRU, (unsigned long)p);
643         if (err)
644                 return err;
645         err |=
646             copy_in_user(&arg32->error_info, &p->error_info,
647                          sizeof(arg32->error_info));
648         if (err)
649                 return -EFAULT;
650         return err;
651 }
652 #endif
653
654 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
655 {
656         drive_info_struct *drv = get_drv(bdev->bd_disk);
657
658         if (!drv->cylinders)
659                 return -ENXIO;
660
661         geo->heads = drv->heads;
662         geo->sectors = drv->sectors;
663         geo->cylinders = drv->cylinders;
664         return 0;
665 }
666
667 /*
668  * ioctl
669  */
670 static int cciss_ioctl(struct inode *inode, struct file *filep,
671                        unsigned int cmd, unsigned long arg)
672 {
673         struct block_device *bdev = inode->i_bdev;
674         struct gendisk *disk = bdev->bd_disk;
675         ctlr_info_t *host = get_host(disk);
676         drive_info_struct *drv = get_drv(disk);
677         int ctlr = host->ctlr;
678         void __user *argp = (void __user *)arg;
679
680 #ifdef CCISS_DEBUG
681         printk(KERN_DEBUG "cciss_ioctl: Called with cmd=%x %lx\n", cmd, arg);
682 #endif                          /* CCISS_DEBUG */
683
684         switch (cmd) {
685         case CCISS_GETPCIINFO:
686                 {
687                         cciss_pci_info_struct pciinfo;
688
689                         if (!arg)
690                                 return -EINVAL;
691                         pciinfo.domain = pci_domain_nr(host->pdev->bus);
692                         pciinfo.bus = host->pdev->bus->number;
693                         pciinfo.dev_fn = host->pdev->devfn;
694                         pciinfo.board_id = host->board_id;
695                         if (copy_to_user
696                             (argp, &pciinfo, sizeof(cciss_pci_info_struct)))
697                                 return -EFAULT;
698                         return 0;
699                 }
700         case CCISS_GETINTINFO:
701                 {
702                         cciss_coalint_struct intinfo;
703                         if (!arg)
704                                 return -EINVAL;
705                         intinfo.delay =
706                             readl(&host->cfgtable->HostWrite.CoalIntDelay);
707                         intinfo.count =
708                             readl(&host->cfgtable->HostWrite.CoalIntCount);
709                         if (copy_to_user
710                             (argp, &intinfo, sizeof(cciss_coalint_struct)))
711                                 return -EFAULT;
712                         return 0;
713                 }
714         case CCISS_SETINTINFO:
715                 {
716                         cciss_coalint_struct intinfo;
717                         unsigned long flags;
718                         int i;
719
720                         if (!arg)
721                                 return -EINVAL;
722                         if (!capable(CAP_SYS_ADMIN))
723                                 return -EPERM;
724                         if (copy_from_user
725                             (&intinfo, argp, sizeof(cciss_coalint_struct)))
726                                 return -EFAULT;
727                         if ((intinfo.delay == 0) && (intinfo.count == 0))
728                         {
729 //                      printk("cciss_ioctl: delay and count cannot be 0\n");
730                                 return -EINVAL;
731                         }
732                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
733                         /* Update the field, and then ring the doorbell */
734                         writel(intinfo.delay,
735                                &(host->cfgtable->HostWrite.CoalIntDelay));
736                         writel(intinfo.count,
737                                &(host->cfgtable->HostWrite.CoalIntCount));
738                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
739
740                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
741                                 if (!(readl(host->vaddr + SA5_DOORBELL)
742                                       & CFGTBL_ChangeReq))
743                                         break;
744                                 /* delay and try again */
745                                 udelay(1000);
746                         }
747                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
748                         if (i >= MAX_IOCTL_CONFIG_WAIT)
749                                 return -EAGAIN;
750                         return 0;
751                 }
752         case CCISS_GETNODENAME:
753                 {
754                         NodeName_type NodeName;
755                         int i;
756
757                         if (!arg)
758                                 return -EINVAL;
759                         for (i = 0; i < 16; i++)
760                                 NodeName[i] =
761                                     readb(&host->cfgtable->ServerName[i]);
762                         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
763                                 return -EFAULT;
764                         return 0;
765                 }
766         case CCISS_SETNODENAME:
767                 {
768                         NodeName_type NodeName;
769                         unsigned long flags;
770                         int i;
771
772                         if (!arg)
773                                 return -EINVAL;
774                         if (!capable(CAP_SYS_ADMIN))
775                                 return -EPERM;
776
777                         if (copy_from_user
778                             (NodeName, argp, sizeof(NodeName_type)))
779                                 return -EFAULT;
780
781                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
782
783                         /* Update the field, and then ring the doorbell */
784                         for (i = 0; i < 16; i++)
785                                 writeb(NodeName[i],
786                                        &host->cfgtable->ServerName[i]);
787
788                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
789
790                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
791                                 if (!(readl(host->vaddr + SA5_DOORBELL)
792                                       & CFGTBL_ChangeReq))
793                                         break;
794                                 /* delay and try again */
795                                 udelay(1000);
796                         }
797                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
798                         if (i >= MAX_IOCTL_CONFIG_WAIT)
799                                 return -EAGAIN;
800                         return 0;
801                 }
802
803         case CCISS_GETHEARTBEAT:
804                 {
805                         Heartbeat_type heartbeat;
806
807                         if (!arg)
808                                 return -EINVAL;
809                         heartbeat = readl(&host->cfgtable->HeartBeat);
810                         if (copy_to_user
811                             (argp, &heartbeat, sizeof(Heartbeat_type)))
812                                 return -EFAULT;
813                         return 0;
814                 }
815         case CCISS_GETBUSTYPES:
816                 {
817                         BusTypes_type BusTypes;
818
819                         if (!arg)
820                                 return -EINVAL;
821                         BusTypes = readl(&host->cfgtable->BusTypes);
822                         if (copy_to_user
823                             (argp, &BusTypes, sizeof(BusTypes_type)))
824                                 return -EFAULT;
825                         return 0;
826                 }
827         case CCISS_GETFIRMVER:
828                 {
829                         FirmwareVer_type firmware;
830
831                         if (!arg)
832                                 return -EINVAL;
833                         memcpy(firmware, host->firm_ver, 4);
834
835                         if (copy_to_user
836                             (argp, firmware, sizeof(FirmwareVer_type)))
837                                 return -EFAULT;
838                         return 0;
839                 }
840         case CCISS_GETDRIVVER:
841                 {
842                         DriverVer_type DriverVer = DRIVER_VERSION;
843
844                         if (!arg)
845                                 return -EINVAL;
846
847                         if (copy_to_user
848                             (argp, &DriverVer, sizeof(DriverVer_type)))
849                                 return -EFAULT;
850                         return 0;
851                 }
852
853         case CCISS_REVALIDVOLS:
854                 if (bdev != bdev->bd_contains || drv != host->drv)
855                         return -ENXIO;
856                 return revalidate_allvol(host);
857
858         case CCISS_GETLUNINFO:{
859                         LogvolInfo_struct luninfo;
860
861                         luninfo.LunID = drv->LunID;
862                         luninfo.num_opens = drv->usage_count;
863                         luninfo.num_parts = 0;
864                         if (copy_to_user(argp, &luninfo,
865                                          sizeof(LogvolInfo_struct)))
866                                 return -EFAULT;
867                         return 0;
868                 }
869         case CCISS_DEREGDISK:
870                 return rebuild_lun_table(host, disk);
871
872         case CCISS_REGNEWD:
873                 return rebuild_lun_table(host, NULL);
874
875         case CCISS_PASSTHRU:
876                 {
877                         IOCTL_Command_struct iocommand;
878                         CommandList_struct *c;
879                         char *buff = NULL;
880                         u64bit temp64;
881                         unsigned long flags;
882                         DECLARE_COMPLETION(wait);
883
884                         if (!arg)
885                                 return -EINVAL;
886
887                         if (!capable(CAP_SYS_RAWIO))
888                                 return -EPERM;
889
890                         if (copy_from_user
891                             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
892                                 return -EFAULT;
893                         if ((iocommand.buf_size < 1) &&
894                             (iocommand.Request.Type.Direction != XFER_NONE)) {
895                                 return -EINVAL;
896                         }
897 #if 0                           /* 'buf_size' member is 16-bits, and always smaller than kmalloc limit */
898                         /* Check kmalloc limits */
899                         if (iocommand.buf_size > 128000)
900                                 return -EINVAL;
901 #endif
902                         if (iocommand.buf_size > 0) {
903                                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
904                                 if (buff == NULL)
905                                         return -EFAULT;
906                         }
907                         if (iocommand.Request.Type.Direction == XFER_WRITE) {
908                                 /* Copy the data into the buffer we created */
909                                 if (copy_from_user
910                                     (buff, iocommand.buf, iocommand.buf_size)) {
911                                         kfree(buff);
912                                         return -EFAULT;
913                                 }
914                         } else {
915                                 memset(buff, 0, iocommand.buf_size);
916                         }
917                         if ((c = cmd_alloc(host, 0)) == NULL) {
918                                 kfree(buff);
919                                 return -ENOMEM;
920                         }
921                         // Fill in the command type
922                         c->cmd_type = CMD_IOCTL_PEND;
923                         // Fill in Command Header
924                         c->Header.ReplyQueue = 0;       // unused in simple mode
925                         if (iocommand.buf_size > 0)     // buffer to fill
926                         {
927                                 c->Header.SGList = 1;
928                                 c->Header.SGTotal = 1;
929                         } else  // no buffers to fill
930                         {
931                                 c->Header.SGList = 0;
932                                 c->Header.SGTotal = 0;
933                         }
934                         c->Header.LUN = iocommand.LUN_info;
935                         c->Header.Tag.lower = c->busaddr;       // use the kernel address the cmd block for tag
936
937                         // Fill in Request block
938                         c->Request = iocommand.Request;
939
940                         // Fill in the scatter gather information
941                         if (iocommand.buf_size > 0) {
942                                 temp64.val = pci_map_single(host->pdev, buff,
943                                         iocommand.buf_size,
944                                         PCI_DMA_BIDIRECTIONAL);
945                                 c->SG[0].Addr.lower = temp64.val32.lower;
946                                 c->SG[0].Addr.upper = temp64.val32.upper;
947                                 c->SG[0].Len = iocommand.buf_size;
948                                 c->SG[0].Ext = 0;       // we are not chaining
949                         }
950                         c->waiting = &wait;
951
952                         /* Put the request on the tail of the request queue */
953                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
954                         addQ(&host->reqQ, c);
955                         host->Qdepth++;
956                         start_io(host);
957                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
958
959                         wait_for_completion(&wait);
960
961                         /* unlock the buffers from DMA */
962                         temp64.val32.lower = c->SG[0].Addr.lower;
963                         temp64.val32.upper = c->SG[0].Addr.upper;
964                         pci_unmap_single(host->pdev, (dma_addr_t) temp64.val,
965                                          iocommand.buf_size,
966                                          PCI_DMA_BIDIRECTIONAL);
967
968                         /* Copy the error information out */
969                         iocommand.error_info = *(c->err_info);
970                         if (copy_to_user
971                             (argp, &iocommand, sizeof(IOCTL_Command_struct))) {
972                                 kfree(buff);
973                                 cmd_free(host, c, 0);
974                                 return -EFAULT;
975                         }
976
977                         if (iocommand.Request.Type.Direction == XFER_READ) {
978                                 /* Copy the data out of the buffer we created */
979                                 if (copy_to_user
980                                     (iocommand.buf, buff, iocommand.buf_size)) {
981                                         kfree(buff);
982                                         cmd_free(host, c, 0);
983                                         return -EFAULT;
984                                 }
985                         }
986                         kfree(buff);
987                         cmd_free(host, c, 0);
988                         return 0;
989                 }
990         case CCISS_BIG_PASSTHRU:{
991                         BIG_IOCTL_Command_struct *ioc;
992                         CommandList_struct *c;
993                         unsigned char **buff = NULL;
994                         int *buff_size = NULL;
995                         u64bit temp64;
996                         unsigned long flags;
997                         BYTE sg_used = 0;
998                         int status = 0;
999                         int i;
1000                         DECLARE_COMPLETION(wait);
1001                         __u32 left;
1002                         __u32 sz;
1003                         BYTE __user *data_ptr;
1004
1005                         if (!arg)
1006                                 return -EINVAL;
1007                         if (!capable(CAP_SYS_RAWIO))
1008                                 return -EPERM;
1009                         ioc = (BIG_IOCTL_Command_struct *)
1010                             kmalloc(sizeof(*ioc), GFP_KERNEL);
1011                         if (!ioc) {
1012                                 status = -ENOMEM;
1013                                 goto cleanup1;
1014                         }
1015                         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1016                                 status = -EFAULT;
1017                                 goto cleanup1;
1018                         }
1019                         if ((ioc->buf_size < 1) &&
1020                             (ioc->Request.Type.Direction != XFER_NONE)) {
1021                                 status = -EINVAL;
1022                                 goto cleanup1;
1023                         }
1024                         /* Check kmalloc limits  using all SGs */
1025                         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1026                                 status = -EINVAL;
1027                                 goto cleanup1;
1028                         }
1029                         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1030                                 status = -EINVAL;
1031                                 goto cleanup1;
1032                         }
1033                         buff =
1034                             kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1035                         if (!buff) {
1036                                 status = -ENOMEM;
1037                                 goto cleanup1;
1038                         }
1039                         buff_size = (int *)kmalloc(MAXSGENTRIES * sizeof(int),
1040                                                    GFP_KERNEL);
1041                         if (!buff_size) {
1042                                 status = -ENOMEM;
1043                                 goto cleanup1;
1044                         }
1045                         left = ioc->buf_size;
1046                         data_ptr = ioc->buf;
1047                         while (left) {
1048                                 sz = (left >
1049                                       ioc->malloc_size) ? ioc->
1050                                     malloc_size : left;
1051                                 buff_size[sg_used] = sz;
1052                                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1053                                 if (buff[sg_used] == NULL) {
1054                                         status = -ENOMEM;
1055                                         goto cleanup1;
1056                                 }
1057                                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1058                                         if (copy_from_user
1059                                             (buff[sg_used], data_ptr, sz)) {
1060                                                 status = -ENOMEM;
1061                                                 goto cleanup1;
1062                                         }
1063                                 } else {
1064                                         memset(buff[sg_used], 0, sz);
1065                                 }
1066                                 left -= sz;
1067                                 data_ptr += sz;
1068                                 sg_used++;
1069                         }
1070                         if ((c = cmd_alloc(host, 0)) == NULL) {
1071                                 status = -ENOMEM;
1072                                 goto cleanup1;
1073                         }
1074                         c->cmd_type = CMD_IOCTL_PEND;
1075                         c->Header.ReplyQueue = 0;
1076
1077                         if (ioc->buf_size > 0) {
1078                                 c->Header.SGList = sg_used;
1079                                 c->Header.SGTotal = sg_used;
1080                         } else {
1081                                 c->Header.SGList = 0;
1082                                 c->Header.SGTotal = 0;
1083                         }
1084                         c->Header.LUN = ioc->LUN_info;
1085                         c->Header.Tag.lower = c->busaddr;
1086
1087                         c->Request = ioc->Request;
1088                         if (ioc->buf_size > 0) {
1089                                 int i;
1090                                 for (i = 0; i < sg_used; i++) {
1091                                         temp64.val =
1092                                             pci_map_single(host->pdev, buff[i],
1093                                                     buff_size[i],
1094                                                     PCI_DMA_BIDIRECTIONAL);
1095                                         c->SG[i].Addr.lower =
1096                                             temp64.val32.lower;
1097                                         c->SG[i].Addr.upper =
1098                                             temp64.val32.upper;
1099                                         c->SG[i].Len = buff_size[i];
1100                                         c->SG[i].Ext = 0;       /* we are not chaining */
1101                                 }
1102                         }
1103                         c->waiting = &wait;
1104                         /* Put the request on the tail of the request queue */
1105                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1106                         addQ(&host->reqQ, c);
1107                         host->Qdepth++;
1108                         start_io(host);
1109                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1110                         wait_for_completion(&wait);
1111                         /* unlock the buffers from DMA */
1112                         for (i = 0; i < sg_used; i++) {
1113                                 temp64.val32.lower = c->SG[i].Addr.lower;
1114                                 temp64.val32.upper = c->SG[i].Addr.upper;
1115                                 pci_unmap_single(host->pdev,
1116                                         (dma_addr_t) temp64.val, buff_size[i],
1117                                         PCI_DMA_BIDIRECTIONAL);
1118                         }
1119                         /* Copy the error information out */
1120                         ioc->error_info = *(c->err_info);
1121                         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1122                                 cmd_free(host, c, 0);
1123                                 status = -EFAULT;
1124                                 goto cleanup1;
1125                         }
1126                         if (ioc->Request.Type.Direction == XFER_READ) {
1127                                 /* Copy the data out of the buffer we created */
1128                                 BYTE __user *ptr = ioc->buf;
1129                                 for (i = 0; i < sg_used; i++) {
1130                                         if (copy_to_user
1131                                             (ptr, buff[i], buff_size[i])) {
1132                                                 cmd_free(host, c, 0);
1133                                                 status = -EFAULT;
1134                                                 goto cleanup1;
1135                                         }
1136                                         ptr += buff_size[i];
1137                                 }
1138                         }
1139                         cmd_free(host, c, 0);
1140                         status = 0;
1141                       cleanup1:
1142                         if (buff) {
1143                                 for (i = 0; i < sg_used; i++)
1144                                         kfree(buff[i]);
1145                                 kfree(buff);
1146                         }
1147                         kfree(buff_size);
1148                         kfree(ioc);
1149                         return status;
1150                 }
1151         default:
1152                 return -ENOTTY;
1153         }
1154 }
1155
1156 /*
1157  * revalidate_allvol is for online array config utilities.  After a
1158  * utility reconfigures the drives in the array, it can use this function
1159  * (through an ioctl) to make the driver zap any previous disk structs for
1160  * that controller and get new ones.
1161  *
1162  * Right now I'm using the getgeometry() function to do this, but this
1163  * function should probably be finer grained and allow you to revalidate one
1164  * particular logical volume (instead of all of them on a particular
1165  * controller).
1166  */
1167 static int revalidate_allvol(ctlr_info_t *host)
1168 {
1169         int ctlr = host->ctlr, i;
1170         unsigned long flags;
1171
1172         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1173         if (host->usage_count > 1) {
1174                 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1175                 printk(KERN_WARNING "cciss: Device busy for volume"
1176                        " revalidation (usage=%d)\n", host->usage_count);
1177                 return -EBUSY;
1178         }
1179         host->usage_count++;
1180         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1181
1182         for (i = 0; i < NWD; i++) {
1183                 struct gendisk *disk = host->gendisk[i];
1184                 if (disk) {
1185                         request_queue_t *q = disk->queue;
1186
1187                         if (disk->flags & GENHD_FL_UP)
1188                                 del_gendisk(disk);
1189                         if (q)
1190                                 blk_cleanup_queue(q);
1191                 }
1192         }
1193
1194         /*
1195          * Set the partition and block size structures for all volumes
1196          * on this controller to zero.  We will reread all of this data
1197          */
1198         memset(host->drv, 0, sizeof(drive_info_struct)
1199                * CISS_MAX_LUN);
1200         /*
1201          * Tell the array controller not to give us any interrupts while
1202          * we check the new geometry.  Then turn interrupts back on when
1203          * we're done.
1204          */
1205         host->access.set_intr_mask(host, CCISS_INTR_OFF);
1206         cciss_getgeometry(ctlr);
1207         host->access.set_intr_mask(host, CCISS_INTR_ON);
1208
1209         /* Loop through each real device */
1210         for (i = 0; i < NWD; i++) {
1211                 struct gendisk *disk = host->gendisk[i];
1212                 drive_info_struct *drv = &(host->drv[i]);
1213                 /* we must register the controller even if no disks exist */
1214                 /* this is for the online array utilities */
1215                 if (!drv->heads && i)
1216                         continue;
1217                 blk_queue_hardsect_size(drv->queue, drv->block_size);
1218                 set_capacity(disk, drv->nr_blocks);
1219                 add_disk(disk);
1220         }
1221         host->usage_count--;
1222         return 0;
1223 }
1224
1225 static inline void complete_buffers(struct bio *bio, int status)
1226 {
1227         while (bio) {
1228                 struct bio *xbh = bio->bi_next;
1229                 int nr_sectors = bio_sectors(bio);
1230
1231                 bio->bi_next = NULL;
1232                 blk_finished_io(len);
1233                 bio_endio(bio, nr_sectors << 9, status ? 0 : -EIO);
1234                 bio = xbh;
1235         }
1236 }
1237
1238 static void cciss_check_queues(ctlr_info_t *h)
1239 {
1240         int start_queue = h->next_to_run;
1241         int i;
1242
1243         /* check to see if we have maxed out the number of commands that can
1244          * be placed on the queue.  If so then exit.  We do this check here
1245          * in case the interrupt we serviced was from an ioctl and did not
1246          * free any new commands.
1247          */
1248         if ((find_first_zero_bit(h->cmd_pool_bits, NR_CMDS)) == NR_CMDS)
1249                 return;
1250
1251         /* We have room on the queue for more commands.  Now we need to queue
1252          * them up.  We will also keep track of the next queue to run so
1253          * that every queue gets a chance to be started first.
1254          */
1255         for (i = 0; i < h->highest_lun + 1; i++) {
1256                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1257                 /* make sure the disk has been added and the drive is real
1258                  * because this can be called from the middle of init_one.
1259                  */
1260                 if (!(h->drv[curr_queue].queue) || !(h->drv[curr_queue].heads))
1261                         continue;
1262                 blk_start_queue(h->gendisk[curr_queue]->queue);
1263
1264                 /* check to see if we have maxed out the number of commands
1265                  * that can be placed on the queue.
1266                  */
1267                 if ((find_first_zero_bit(h->cmd_pool_bits, NR_CMDS)) == NR_CMDS) {
1268                         if (curr_queue == start_queue) {
1269                                 h->next_to_run =
1270                                     (start_queue + 1) % (h->highest_lun + 1);
1271                                 break;
1272                         } else {
1273                                 h->next_to_run = curr_queue;
1274                                 break;
1275                         }
1276                 } else {
1277                         curr_queue = (curr_queue + 1) % (h->highest_lun + 1);
1278                 }
1279         }
1280 }
1281
1282 static void cciss_softirq_done(struct request *rq)
1283 {
1284         CommandList_struct *cmd = rq->completion_data;
1285         ctlr_info_t *h = hba[cmd->ctlr];
1286         unsigned long flags;
1287         u64bit temp64;
1288         int i, ddir;
1289
1290         if (cmd->Request.Type.Direction == XFER_READ)
1291                 ddir = PCI_DMA_FROMDEVICE;
1292         else
1293                 ddir = PCI_DMA_TODEVICE;
1294
1295         /* command did not need to be retried */
1296         /* unmap the DMA mapping for all the scatter gather elements */
1297         for (i = 0; i < cmd->Header.SGList; i++) {
1298                 temp64.val32.lower = cmd->SG[i].Addr.lower;
1299                 temp64.val32.upper = cmd->SG[i].Addr.upper;
1300                 pci_unmap_page(h->pdev, temp64.val, cmd->SG[i].Len, ddir);
1301         }
1302
1303         complete_buffers(rq->bio, rq->errors);
1304
1305 #ifdef CCISS_DEBUG
1306         printk("Done with %p\n", rq);
1307 #endif                          /* CCISS_DEBUG */
1308
1309         add_disk_randomness(rq->rq_disk);
1310         spin_lock_irqsave(&h->lock, flags);
1311         end_that_request_last(rq, rq->errors);
1312         cmd_free(h, cmd, 1);
1313         cciss_check_queues(h);
1314         spin_unlock_irqrestore(&h->lock, flags);
1315 }
1316
1317 /* This function will check the usage_count of the drive to be updated/added.
1318  * If the usage_count is zero then the drive information will be updated and
1319  * the disk will be re-registered with the kernel.  If not then it will be
1320  * left alone for the next reboot.  The exception to this is disk 0 which
1321  * will always be left registered with the kernel since it is also the
1322  * controller node.  Any changes to disk 0 will show up on the next
1323  * reboot.
1324  */
1325 static void cciss_update_drive_info(int ctlr, int drv_index)
1326 {
1327         ctlr_info_t *h = hba[ctlr];
1328         struct gendisk *disk;
1329         ReadCapdata_struct *size_buff = NULL;
1330         InquiryData_struct *inq_buff = NULL;
1331         unsigned int block_size;
1332         unsigned int total_size;
1333         unsigned long flags = 0;
1334         int ret = 0;
1335
1336         /* if the disk already exists then deregister it before proceeding */
1337         if (h->drv[drv_index].raid_level != -1) {
1338                 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1339                 h->drv[drv_index].busy_configuring = 1;
1340                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1341                 ret = deregister_disk(h->gendisk[drv_index],
1342                                       &h->drv[drv_index], 0);
1343                 h->drv[drv_index].busy_configuring = 0;
1344         }
1345
1346         /* If the disk is in use return */
1347         if (ret)
1348                 return;
1349
1350         /* Get information about the disk and modify the driver structure */
1351         size_buff = kmalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
1352         if (size_buff == NULL)
1353                 goto mem_msg;
1354         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1355         if (inq_buff == NULL)
1356                 goto mem_msg;
1357
1358         cciss_read_capacity(ctlr, drv_index, size_buff, 1,
1359                             &total_size, &block_size);
1360         cciss_geometry_inquiry(ctlr, drv_index, 1, total_size, block_size,
1361                                inq_buff, &h->drv[drv_index]);
1362
1363         ++h->num_luns;
1364         disk = h->gendisk[drv_index];
1365         set_capacity(disk, h->drv[drv_index].nr_blocks);
1366
1367         /* if it's the controller it's already added */
1368         if (drv_index) {
1369                 disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1370
1371                 /* Set up queue information */
1372                 disk->queue->backing_dev_info.ra_pages = READ_AHEAD;
1373                 blk_queue_bounce_limit(disk->queue, hba[ctlr]->pdev->dma_mask);
1374
1375                 /* This is a hardware imposed limit. */
1376                 blk_queue_max_hw_segments(disk->queue, MAXSGENTRIES);
1377
1378                 /* This is a limit in the driver and could be eliminated. */
1379                 blk_queue_max_phys_segments(disk->queue, MAXSGENTRIES);
1380
1381                 blk_queue_max_sectors(disk->queue, 512);
1382
1383                 blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1384
1385                 disk->queue->queuedata = hba[ctlr];
1386
1387                 blk_queue_hardsect_size(disk->queue,
1388                                         hba[ctlr]->drv[drv_index].block_size);
1389
1390                 h->drv[drv_index].queue = disk->queue;
1391                 add_disk(disk);
1392         }
1393
1394       freeret:
1395         kfree(size_buff);
1396         kfree(inq_buff);
1397         return;
1398       mem_msg:
1399         printk(KERN_ERR "cciss: out of memory\n");
1400         goto freeret;
1401 }
1402
1403 /* This function will find the first index of the controllers drive array
1404  * that has a -1 for the raid_level and will return that index.  This is
1405  * where new drives will be added.  If the index to be returned is greater
1406  * than the highest_lun index for the controller then highest_lun is set
1407  * to this new index.  If there are no available indexes then -1 is returned.
1408  */
1409 static int cciss_find_free_drive_index(int ctlr)
1410 {
1411         int i;
1412
1413         for (i = 0; i < CISS_MAX_LUN; i++) {
1414                 if (hba[ctlr]->drv[i].raid_level == -1) {
1415                         if (i > hba[ctlr]->highest_lun)
1416                                 hba[ctlr]->highest_lun = i;
1417                         return i;
1418                 }
1419         }
1420         return -1;
1421 }
1422
1423 /* This function will add and remove logical drives from the Logical
1424  * drive array of the controller and maintain persistency of ordering
1425  * so that mount points are preserved until the next reboot.  This allows
1426  * for the removal of logical drives in the middle of the drive array
1427  * without a re-ordering of those drives.
1428  * INPUT
1429  * h            = The controller to perform the operations on
1430  * del_disk     = The disk to remove if specified.  If the value given
1431  *                is NULL then no disk is removed.
1432  */
1433 static int rebuild_lun_table(ctlr_info_t *h, struct gendisk *del_disk)
1434 {
1435         int ctlr = h->ctlr;
1436         int num_luns;
1437         ReportLunData_struct *ld_buff = NULL;
1438         drive_info_struct *drv = NULL;
1439         int return_code;
1440         int listlength = 0;
1441         int i;
1442         int drv_found;
1443         int drv_index = 0;
1444         __u32 lunid = 0;
1445         unsigned long flags;
1446
1447         /* Set busy_configuring flag for this operation */
1448         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1449         if (h->num_luns >= CISS_MAX_LUN) {
1450                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1451                 return -EINVAL;
1452         }
1453
1454         if (h->busy_configuring) {
1455                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1456                 return -EBUSY;
1457         }
1458         h->busy_configuring = 1;
1459
1460         /* if del_disk is NULL then we are being called to add a new disk
1461          * and update the logical drive table.  If it is not NULL then
1462          * we will check if the disk is in use or not.
1463          */
1464         if (del_disk != NULL) {
1465                 drv = get_drv(del_disk);
1466                 drv->busy_configuring = 1;
1467                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1468                 return_code = deregister_disk(del_disk, drv, 1);
1469                 drv->busy_configuring = 0;
1470                 h->busy_configuring = 0;
1471                 return return_code;
1472         } else {
1473                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1474                 if (!capable(CAP_SYS_RAWIO))
1475                         return -EPERM;
1476
1477                 ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
1478                 if (ld_buff == NULL)
1479                         goto mem_msg;
1480
1481                 return_code = sendcmd_withirq(CISS_REPORT_LOG, ctlr, ld_buff,
1482                                               sizeof(ReportLunData_struct), 0,
1483                                               0, 0, TYPE_CMD);
1484
1485                 if (return_code == IO_OK) {
1486                         listlength |=
1487                             (0xff & (unsigned int)(ld_buff->LUNListLength[0]))
1488                             << 24;
1489                         listlength |=
1490                             (0xff & (unsigned int)(ld_buff->LUNListLength[1]))
1491                             << 16;
1492                         listlength |=
1493                             (0xff & (unsigned int)(ld_buff->LUNListLength[2]))
1494                             << 8;
1495                         listlength |=
1496                             0xff & (unsigned int)(ld_buff->LUNListLength[3]);
1497                 } else {        /* reading number of logical volumes failed */
1498                         printk(KERN_WARNING "cciss: report logical volume"
1499                                " command failed\n");
1500                         listlength = 0;
1501                         goto freeret;
1502                 }
1503
1504                 num_luns = listlength / 8;      /* 8 bytes per entry */
1505                 if (num_luns > CISS_MAX_LUN) {
1506                         num_luns = CISS_MAX_LUN;
1507                         printk(KERN_WARNING "cciss: more luns configured"
1508                                " on controller than can be handled by"
1509                                " this driver.\n");
1510                 }
1511
1512                 /* Compare controller drive array to drivers drive array.
1513                  * Check for updates in the drive information and any new drives
1514                  * on the controller.
1515                  */
1516                 for (i = 0; i < num_luns; i++) {
1517                         int j;
1518
1519                         drv_found = 0;
1520
1521                         lunid = (0xff &
1522                                  (unsigned int)(ld_buff->LUN[i][3])) << 24;
1523                         lunid |= (0xff &
1524                                   (unsigned int)(ld_buff->LUN[i][2])) << 16;
1525                         lunid |= (0xff &
1526                                   (unsigned int)(ld_buff->LUN[i][1])) << 8;
1527                         lunid |= 0xff & (unsigned int)(ld_buff->LUN[i][0]);
1528
1529                         /* Find if the LUN is already in the drive array
1530                          * of the controller.  If so then update its info
1531                          * if not is use.  If it does not exist then find
1532                          * the first free index and add it.
1533                          */
1534                         for (j = 0; j <= h->highest_lun; j++) {
1535                                 if (h->drv[j].LunID == lunid) {
1536                                         drv_index = j;
1537                                         drv_found = 1;
1538                                 }
1539                         }
1540
1541                         /* check if the drive was found already in the array */
1542                         if (!drv_found) {
1543                                 drv_index = cciss_find_free_drive_index(ctlr);
1544                                 if (drv_index == -1)
1545                                         goto freeret;
1546
1547                         }
1548                         h->drv[drv_index].LunID = lunid;
1549                         cciss_update_drive_info(ctlr, drv_index);
1550                 }               /* end for */
1551         }                       /* end else */
1552
1553       freeret:
1554         kfree(ld_buff);
1555         h->busy_configuring = 0;
1556         /* We return -1 here to tell the ACU that we have registered/updated
1557          * all of the drives that we can and to keep it from calling us
1558          * additional times.
1559          */
1560         return -1;
1561       mem_msg:
1562         printk(KERN_ERR "cciss: out of memory\n");
1563         goto freeret;
1564 }
1565
1566 /* This function will deregister the disk and it's queue from the
1567  * kernel.  It must be called with the controller lock held and the
1568  * drv structures busy_configuring flag set.  It's parameters are:
1569  *
1570  * disk = This is the disk to be deregistered
1571  * drv  = This is the drive_info_struct associated with the disk to be
1572  *        deregistered.  It contains information about the disk used
1573  *        by the driver.
1574  * clear_all = This flag determines whether or not the disk information
1575  *             is going to be completely cleared out and the highest_lun
1576  *             reset.  Sometimes we want to clear out information about
1577  *             the disk in preparation for re-adding it.  In this case
1578  *             the highest_lun should be left unchanged and the LunID
1579  *             should not be cleared.
1580 */
1581 static int deregister_disk(struct gendisk *disk, drive_info_struct *drv,
1582                            int clear_all)
1583 {
1584         ctlr_info_t *h = get_host(disk);
1585
1586         if (!capable(CAP_SYS_RAWIO))
1587                 return -EPERM;
1588
1589         /* make sure logical volume is NOT is use */
1590         if (clear_all || (h->gendisk[0] == disk)) {
1591                 if (drv->usage_count > 1)
1592                         return -EBUSY;
1593         } else if (drv->usage_count > 0)
1594                 return -EBUSY;
1595
1596         /* invalidate the devices and deregister the disk.  If it is disk
1597          * zero do not deregister it but just zero out it's values.  This
1598          * allows us to delete disk zero but keep the controller registered.
1599          */
1600         if (h->gendisk[0] != disk) {
1601                 if (disk) {
1602                         request_queue_t *q = disk->queue;
1603                         if (disk->flags & GENHD_FL_UP)
1604                                 del_gendisk(disk);
1605                         if (q) {
1606                                 blk_cleanup_queue(q);
1607                                 drv->queue = NULL;
1608                         }
1609                 }
1610         }
1611
1612         --h->num_luns;
1613         /* zero out the disk size info */
1614         drv->nr_blocks = 0;
1615         drv->block_size = 0;
1616         drv->heads = 0;
1617         drv->sectors = 0;
1618         drv->cylinders = 0;
1619         drv->raid_level = -1;   /* This can be used as a flag variable to
1620                                  * indicate that this element of the drive
1621                                  * array is free.
1622                                  */
1623
1624         if (clear_all) {
1625                 /* check to see if it was the last disk */
1626                 if (drv == h->drv + h->highest_lun) {
1627                         /* if so, find the new hightest lun */
1628                         int i, newhighest = -1;
1629                         for (i = 0; i < h->highest_lun; i++) {
1630                                 /* if the disk has size > 0, it is available */
1631                                 if (h->drv[i].heads)
1632                                         newhighest = i;
1633                         }
1634                         h->highest_lun = newhighest;
1635                 }
1636
1637                 drv->LunID = 0;
1638         }
1639         return 0;
1640 }
1641
1642 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,
1643                                                                                                                            1: address logical volume log_unit,
1644                                                                                                                            2: periph device address is scsi3addr */
1645                     unsigned int log_unit, __u8 page_code,
1646                     unsigned char *scsi3addr, int cmd_type)
1647 {
1648         ctlr_info_t *h = hba[ctlr];
1649         u64bit buff_dma_handle;
1650         int status = IO_OK;
1651
1652         c->cmd_type = CMD_IOCTL_PEND;
1653         c->Header.ReplyQueue = 0;
1654         if (buff != NULL) {
1655                 c->Header.SGList = 1;
1656                 c->Header.SGTotal = 1;
1657         } else {
1658                 c->Header.SGList = 0;
1659                 c->Header.SGTotal = 0;
1660         }
1661         c->Header.Tag.lower = c->busaddr;
1662
1663         c->Request.Type.Type = cmd_type;
1664         if (cmd_type == TYPE_CMD) {
1665                 switch (cmd) {
1666                 case CISS_INQUIRY:
1667                         /* If the logical unit number is 0 then, this is going
1668                            to controller so It's a physical command
1669                            mode = 0 target = 0.  So we have nothing to write.
1670                            otherwise, if use_unit_num == 1,
1671                            mode = 1(volume set addressing) target = LUNID
1672                            otherwise, if use_unit_num == 2,
1673                            mode = 0(periph dev addr) target = scsi3addr */
1674                         if (use_unit_num == 1) {
1675                                 c->Header.LUN.LogDev.VolId =
1676                                     h->drv[log_unit].LunID;
1677                                 c->Header.LUN.LogDev.Mode = 1;
1678                         } else if (use_unit_num == 2) {
1679                                 memcpy(c->Header.LUN.LunAddrBytes, scsi3addr,
1680                                        8);
1681                                 c->Header.LUN.LogDev.Mode = 0;
1682                         }
1683                         /* are we trying to read a vital product page */
1684                         if (page_code != 0) {
1685                                 c->Request.CDB[1] = 0x01;
1686                                 c->Request.CDB[2] = page_code;
1687                         }
1688                         c->Request.CDBLen = 6;
1689                         c->Request.Type.Attribute = ATTR_SIMPLE;
1690                         c->Request.Type.Direction = XFER_READ;
1691                         c->Request.Timeout = 0;
1692                         c->Request.CDB[0] = CISS_INQUIRY;
1693                         c->Request.CDB[4] = size & 0xFF;
1694                         break;
1695                 case CISS_REPORT_LOG:
1696                 case CISS_REPORT_PHYS:
1697                         /* Talking to controller so It's a physical command
1698                            mode = 00 target = 0.  Nothing to write.
1699                          */
1700                         c->Request.CDBLen = 12;
1701                         c->Request.Type.Attribute = ATTR_SIMPLE;
1702                         c->Request.Type.Direction = XFER_READ;
1703                         c->Request.Timeout = 0;
1704                         c->Request.CDB[0] = cmd;
1705                         c->Request.CDB[6] = (size >> 24) & 0xFF;        //MSB
1706                         c->Request.CDB[7] = (size >> 16) & 0xFF;
1707                         c->Request.CDB[8] = (size >> 8) & 0xFF;
1708                         c->Request.CDB[9] = size & 0xFF;
1709                         break;
1710
1711                 case CCISS_READ_CAPACITY:
1712                         c->Header.LUN.LogDev.VolId = h->drv[log_unit].LunID;
1713                         c->Header.LUN.LogDev.Mode = 1;
1714                         c->Request.CDBLen = 10;
1715                         c->Request.Type.Attribute = ATTR_SIMPLE;
1716                         c->Request.Type.Direction = XFER_READ;
1717                         c->Request.Timeout = 0;
1718                         c->Request.CDB[0] = cmd;
1719                         break;
1720                 case CCISS_CACHE_FLUSH:
1721                         c->Request.CDBLen = 12;
1722                         c->Request.Type.Attribute = ATTR_SIMPLE;
1723                         c->Request.Type.Direction = XFER_WRITE;
1724                         c->Request.Timeout = 0;
1725                         c->Request.CDB[0] = BMIC_WRITE;
1726                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
1727                         break;
1728                 default:
1729                         printk(KERN_WARNING
1730                                "cciss%d:  Unknown Command 0x%c\n", ctlr, cmd);
1731                         return IO_ERROR;
1732                 }
1733         } else if (cmd_type == TYPE_MSG) {
1734                 switch (cmd) {
1735                 case 0: /* ABORT message */
1736                         c->Request.CDBLen = 12;
1737                         c->Request.Type.Attribute = ATTR_SIMPLE;
1738                         c->Request.Type.Direction = XFER_WRITE;
1739                         c->Request.Timeout = 0;
1740                         c->Request.CDB[0] = cmd;        /* abort */
1741                         c->Request.CDB[1] = 0;  /* abort a command */
1742                         /* buff contains the tag of the command to abort */
1743                         memcpy(&c->Request.CDB[4], buff, 8);
1744                         break;
1745                 case 1: /* RESET message */
1746                         c->Request.CDBLen = 12;
1747                         c->Request.Type.Attribute = ATTR_SIMPLE;
1748                         c->Request.Type.Direction = XFER_WRITE;
1749                         c->Request.Timeout = 0;
1750                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
1751                         c->Request.CDB[0] = cmd;        /* reset */
1752                         c->Request.CDB[1] = 0x04;       /* reset a LUN */
1753                 case 3: /* No-Op message */
1754                         c->Request.CDBLen = 1;
1755                         c->Request.Type.Attribute = ATTR_SIMPLE;
1756                         c->Request.Type.Direction = XFER_WRITE;
1757                         c->Request.Timeout = 0;
1758                         c->Request.CDB[0] = cmd;
1759                         break;
1760                 default:
1761                         printk(KERN_WARNING
1762                                "cciss%d: unknown message type %d\n", ctlr, cmd);
1763                         return IO_ERROR;
1764                 }
1765         } else {
1766                 printk(KERN_WARNING
1767                        "cciss%d: unknown command type %d\n", ctlr, cmd_type);
1768                 return IO_ERROR;
1769         }
1770         /* Fill in the scatter gather information */
1771         if (size > 0) {
1772                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
1773                                                              buff, size,
1774                                                              PCI_DMA_BIDIRECTIONAL);
1775                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
1776                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
1777                 c->SG[0].Len = size;
1778                 c->SG[0].Ext = 0;       /* we are not chaining */
1779         }
1780         return status;
1781 }
1782
1783 static int sendcmd_withirq(__u8 cmd,
1784                            int ctlr,
1785                            void *buff,
1786                            size_t size,
1787                            unsigned int use_unit_num,
1788                            unsigned int log_unit, __u8 page_code, int cmd_type)
1789 {
1790         ctlr_info_t *h = hba[ctlr];
1791         CommandList_struct *c;
1792         u64bit buff_dma_handle;
1793         unsigned long flags;
1794         int return_status;
1795         DECLARE_COMPLETION(wait);
1796
1797         if ((c = cmd_alloc(h, 0)) == NULL)
1798                 return -ENOMEM;
1799         return_status = fill_cmd(c, cmd, ctlr, buff, size, use_unit_num,
1800                                  log_unit, page_code, NULL, cmd_type);
1801         if (return_status != IO_OK) {
1802                 cmd_free(h, c, 0);
1803                 return return_status;
1804         }
1805       resend_cmd2:
1806         c->waiting = &wait;
1807
1808         /* Put the request on the tail of the queue and send it */
1809         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1810         addQ(&h->reqQ, c);
1811         h->Qdepth++;
1812         start_io(h);
1813         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1814
1815         wait_for_completion(&wait);
1816
1817         if (c->err_info->CommandStatus != 0) {  /* an error has occurred */
1818                 switch (c->err_info->CommandStatus) {
1819                 case CMD_TARGET_STATUS:
1820                         printk(KERN_WARNING "cciss: cmd %p has "
1821                                " completed with errors\n", c);
1822                         if (c->err_info->ScsiStatus) {
1823                                 printk(KERN_WARNING "cciss: cmd %p "
1824                                        "has SCSI Status = %x\n",
1825                                        c, c->err_info->ScsiStatus);
1826                         }
1827
1828                         break;
1829                 case CMD_DATA_UNDERRUN:
1830                 case CMD_DATA_OVERRUN:
1831                         /* expected for inquire and report lun commands */
1832                         break;
1833                 case CMD_INVALID:
1834                         printk(KERN_WARNING "cciss: Cmd %p is "
1835                                "reported invalid\n", c);
1836                         return_status = IO_ERROR;
1837                         break;
1838                 case CMD_PROTOCOL_ERR:
1839                         printk(KERN_WARNING "cciss: cmd %p has "
1840                                "protocol error \n", c);
1841                         return_status = IO_ERROR;
1842                         break;
1843                 case CMD_HARDWARE_ERR:
1844                         printk(KERN_WARNING "cciss: cmd %p had "
1845                                " hardware error\n", c);
1846                         return_status = IO_ERROR;
1847                         break;
1848                 case CMD_CONNECTION_LOST:
1849                         printk(KERN_WARNING "cciss: cmd %p had "
1850                                "connection lost\n", c);
1851                         return_status = IO_ERROR;
1852                         break;
1853                 case CMD_ABORTED:
1854                         printk(KERN_WARNING "cciss: cmd %p was "
1855                                "aborted\n", c);
1856                         return_status = IO_ERROR;
1857                         break;
1858                 case CMD_ABORT_FAILED:
1859                         printk(KERN_WARNING "cciss: cmd %p reports "
1860                                "abort failed\n", c);
1861                         return_status = IO_ERROR;
1862                         break;
1863                 case CMD_UNSOLICITED_ABORT:
1864                         printk(KERN_WARNING
1865                                "cciss%d: unsolicited abort %p\n", ctlr, c);
1866                         if (c->retry_count < MAX_CMD_RETRIES) {
1867                                 printk(KERN_WARNING
1868                                        "cciss%d: retrying %p\n", ctlr, c);
1869                                 c->retry_count++;
1870                                 /* erase the old error information */
1871                                 memset(c->err_info, 0,
1872                                        sizeof(ErrorInfo_struct));
1873                                 return_status = IO_OK;
1874                                 INIT_COMPLETION(wait);
1875                                 goto resend_cmd2;
1876                         }
1877                         return_status = IO_ERROR;
1878                         break;
1879                 default:
1880                         printk(KERN_WARNING "cciss: cmd %p returned "
1881                                "unknown status %x\n", c,
1882                                c->err_info->CommandStatus);
1883                         return_status = IO_ERROR;
1884                 }
1885         }
1886         /* unlock the buffers from DMA */
1887         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
1888         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
1889         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
1890                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
1891         cmd_free(h, c, 0);
1892         return return_status;
1893 }
1894
1895 static void cciss_geometry_inquiry(int ctlr, int logvol,
1896                                    int withirq, unsigned int total_size,
1897                                    unsigned int block_size,
1898                                    InquiryData_struct *inq_buff,
1899                                    drive_info_struct *drv)
1900 {
1901         int return_code;
1902         memset(inq_buff, 0, sizeof(InquiryData_struct));
1903         if (withirq)
1904                 return_code = sendcmd_withirq(CISS_INQUIRY, ctlr,
1905                                               inq_buff, sizeof(*inq_buff), 1,
1906                                               logvol, 0xC1, TYPE_CMD);
1907         else
1908                 return_code = sendcmd(CISS_INQUIRY, ctlr, inq_buff,
1909                                       sizeof(*inq_buff), 1, logvol, 0xC1, NULL,
1910                                       TYPE_CMD);
1911         if (return_code == IO_OK) {
1912                 if (inq_buff->data_byte[8] == 0xFF) {
1913                         printk(KERN_WARNING
1914                                "cciss: reading geometry failed, volume "
1915                                "does not support reading geometry\n");
1916                         drv->block_size = block_size;
1917                         drv->nr_blocks = total_size;
1918                         drv->heads = 255;
1919                         drv->sectors = 32;      // Sectors per track
1920                         drv->cylinders = total_size / 255 / 32;
1921                 } else {
1922                         unsigned int t;
1923
1924                         drv->block_size = block_size;
1925                         drv->nr_blocks = total_size;
1926                         drv->heads = inq_buff->data_byte[6];
1927                         drv->sectors = inq_buff->data_byte[7];
1928                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
1929                         drv->cylinders += inq_buff->data_byte[5];
1930                         drv->raid_level = inq_buff->data_byte[8];
1931                         t = drv->heads * drv->sectors;
1932                         if (t > 1) {
1933                                 drv->cylinders = total_size / t;
1934                         }
1935                 }
1936         } else {                /* Get geometry failed */
1937                 printk(KERN_WARNING "cciss: reading geometry failed\n");
1938         }
1939         printk(KERN_INFO "      heads= %d, sectors= %d, cylinders= %d\n\n",
1940                drv->heads, drv->sectors, drv->cylinders);
1941 }
1942
1943 static void
1944 cciss_read_capacity(int ctlr, int logvol, ReadCapdata_struct *buf,
1945                     int withirq, unsigned int *total_size,
1946                     unsigned int *block_size)
1947 {
1948         int return_code;
1949         memset(buf, 0, sizeof(*buf));
1950         if (withirq)
1951                 return_code = sendcmd_withirq(CCISS_READ_CAPACITY,
1952                                               ctlr, buf, sizeof(*buf), 1,
1953                                               logvol, 0, TYPE_CMD);
1954         else
1955                 return_code = sendcmd(CCISS_READ_CAPACITY,
1956                                       ctlr, buf, sizeof(*buf), 1, logvol, 0,
1957                                       NULL, TYPE_CMD);
1958         if (return_code == IO_OK) {
1959                 *total_size =
1960                     be32_to_cpu(*((__be32 *) & buf->total_size[0])) + 1;
1961                 *block_size = be32_to_cpu(*((__be32 *) & buf->block_size[0]));
1962         } else {                /* read capacity command failed */
1963                 printk(KERN_WARNING "cciss: read capacity failed\n");
1964                 *total_size = 0;
1965                 *block_size = BLOCK_SIZE;
1966         }
1967         printk(KERN_INFO "      blocks= %u block_size= %d\n",
1968                *total_size, *block_size);
1969         return;
1970 }
1971
1972 static int cciss_revalidate(struct gendisk *disk)
1973 {
1974         ctlr_info_t *h = get_host(disk);
1975         drive_info_struct *drv = get_drv(disk);
1976         int logvol;
1977         int FOUND = 0;
1978         unsigned int block_size;
1979         unsigned int total_size;
1980         ReadCapdata_struct *size_buff = NULL;
1981         InquiryData_struct *inq_buff = NULL;
1982
1983         for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
1984                 if (h->drv[logvol].LunID == drv->LunID) {
1985                         FOUND = 1;
1986                         break;
1987                 }
1988         }
1989
1990         if (!FOUND)
1991                 return 1;
1992
1993         size_buff = kmalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
1994         if (size_buff == NULL) {
1995                 printk(KERN_WARNING "cciss: out of memory\n");
1996                 return 1;
1997         }
1998         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1999         if (inq_buff == NULL) {
2000                 printk(KERN_WARNING "cciss: out of memory\n");
2001                 kfree(size_buff);
2002                 return 1;
2003         }
2004
2005         cciss_read_capacity(h->ctlr, logvol, size_buff, 1, &total_size,
2006                             &block_size);
2007         cciss_geometry_inquiry(h->ctlr, logvol, 1, total_size, block_size,
2008                                inq_buff, drv);
2009
2010         blk_queue_hardsect_size(drv->queue, drv->block_size);
2011         set_capacity(disk, drv->nr_blocks);
2012
2013         kfree(size_buff);
2014         kfree(inq_buff);
2015         return 0;
2016 }
2017
2018 /*
2019  *   Wait polling for a command to complete.
2020  *   The memory mapped FIFO is polled for the completion.
2021  *   Used only at init time, interrupts from the HBA are disabled.
2022  */
2023 static unsigned long pollcomplete(int ctlr)
2024 {
2025         unsigned long done;
2026         int i;
2027
2028         /* Wait (up to 20 seconds) for a command to complete */
2029
2030         for (i = 20 * HZ; i > 0; i--) {
2031                 done = hba[ctlr]->access.command_completed(hba[ctlr]);
2032                 if (done == FIFO_EMPTY)
2033                         schedule_timeout_uninterruptible(1);
2034                 else
2035                         return done;
2036         }
2037         /* Invalid address to tell caller we ran out of time */
2038         return 1;
2039 }
2040
2041 static int add_sendcmd_reject(__u8 cmd, int ctlr, unsigned long complete)
2042 {
2043         /* We get in here if sendcmd() is polling for completions
2044            and gets some command back that it wasn't expecting --
2045            something other than that which it just sent down.
2046            Ordinarily, that shouldn't happen, but it can happen when
2047            the scsi tape stuff gets into error handling mode, and
2048            starts using sendcmd() to try to abort commands and
2049            reset tape drives.  In that case, sendcmd may pick up
2050            completions of commands that were sent to logical drives
2051            through the block i/o system, or cciss ioctls completing, etc.
2052            In that case, we need to save those completions for later
2053            processing by the interrupt handler.
2054          */
2055
2056 #ifdef CONFIG_CISS_SCSI_TAPE
2057         struct sendcmd_reject_list *srl = &hba[ctlr]->scsi_rejects;
2058
2059         /* If it's not the scsi tape stuff doing error handling, (abort */
2060         /* or reset) then we don't expect anything weird. */
2061         if (cmd != CCISS_RESET_MSG && cmd != CCISS_ABORT_MSG) {
2062 #endif
2063                 printk(KERN_WARNING "cciss cciss%d: SendCmd "
2064                        "Invalid command list address returned! (%lx)\n",
2065                        ctlr, complete);
2066                 /* not much we can do. */
2067 #ifdef CONFIG_CISS_SCSI_TAPE
2068                 return 1;
2069         }
2070
2071         /* We've sent down an abort or reset, but something else
2072            has completed */
2073         if (srl->ncompletions >= (NR_CMDS + 2)) {
2074                 /* Uh oh.  No room to save it for later... */
2075                 printk(KERN_WARNING "cciss%d: Sendcmd: Invalid command addr, "
2076                        "reject list overflow, command lost!\n", ctlr);
2077                 return 1;
2078         }
2079         /* Save it for later */
2080         srl->complete[srl->ncompletions] = complete;
2081         srl->ncompletions++;
2082 #endif
2083         return 0;
2084 }
2085
2086 /*
2087  * Send a command to the controller, and wait for it to complete.
2088  * Only used at init time.
2089  */
2090 static int sendcmd(__u8 cmd, int ctlr, void *buff, size_t size, unsigned int use_unit_num,      /* 0: address the controller,
2091                                                                                                    1: address logical volume log_unit,
2092                                                                                                    2: periph device address is scsi3addr */
2093                    unsigned int log_unit,
2094                    __u8 page_code, unsigned char *scsi3addr, int cmd_type)
2095 {
2096         CommandList_struct *c;
2097         int i;
2098         unsigned long complete;
2099         ctlr_info_t *info_p = hba[ctlr];
2100         u64bit buff_dma_handle;
2101         int status, done = 0;
2102
2103         if ((c = cmd_alloc(info_p, 1)) == NULL) {
2104                 printk(KERN_WARNING "cciss: unable to get memory");
2105                 return IO_ERROR;
2106         }
2107         status = fill_cmd(c, cmd, ctlr, buff, size, use_unit_num,
2108                           log_unit, page_code, scsi3addr, cmd_type);
2109         if (status != IO_OK) {
2110                 cmd_free(info_p, c, 1);
2111                 return status;
2112         }
2113       resend_cmd1:
2114         /*
2115          * Disable interrupt
2116          */
2117 #ifdef CCISS_DEBUG
2118         printk(KERN_DEBUG "cciss: turning intr off\n");
2119 #endif                          /* CCISS_DEBUG */
2120         info_p->access.set_intr_mask(info_p, CCISS_INTR_OFF);
2121
2122         /* Make sure there is room in the command FIFO */
2123         /* Actually it should be completely empty at this time */
2124         /* unless we are in here doing error handling for the scsi */
2125         /* tape side of the driver. */
2126         for (i = 200000; i > 0; i--) {
2127                 /* if fifo isn't full go */
2128                 if (!(info_p->access.fifo_full(info_p))) {
2129
2130                         break;
2131                 }
2132                 udelay(10);
2133                 printk(KERN_WARNING "cciss cciss%d: SendCmd FIFO full,"
2134                        " waiting!\n", ctlr);
2135         }
2136         /*
2137          * Send the cmd
2138          */
2139         info_p->access.submit_command(info_p, c);
2140         done = 0;
2141         do {
2142                 complete = pollcomplete(ctlr);
2143
2144 #ifdef CCISS_DEBUG
2145                 printk(KERN_DEBUG "cciss: command completed\n");
2146 #endif                          /* CCISS_DEBUG */
2147
2148                 if (complete == 1) {
2149                         printk(KERN_WARNING
2150                                "cciss cciss%d: SendCmd Timeout out, "
2151                                "No command list address returned!\n", ctlr);
2152                         status = IO_ERROR;
2153                         done = 1;
2154                         break;
2155                 }
2156
2157                 /* This will need to change for direct lookup completions */
2158                 if ((complete & CISS_ERROR_BIT)
2159                     && (complete & ~CISS_ERROR_BIT) == c->busaddr) {
2160                         /* if data overrun or underun on Report command
2161                            ignore it
2162                          */
2163                         if (((c->Request.CDB[0] == CISS_REPORT_LOG) ||
2164                              (c->Request.CDB[0] == CISS_REPORT_PHYS) ||
2165                              (c->Request.CDB[0] == CISS_INQUIRY)) &&
2166                             ((c->err_info->CommandStatus ==
2167                               CMD_DATA_OVERRUN) ||
2168                              (c->err_info->CommandStatus == CMD_DATA_UNDERRUN)
2169                             )) {
2170                                 complete = c->busaddr;
2171                         } else {
2172                                 if (c->err_info->CommandStatus ==
2173                                     CMD_UNSOLICITED_ABORT) {
2174                                         printk(KERN_WARNING "cciss%d: "
2175                                                "unsolicited abort %p\n",
2176                                                ctlr, c);
2177                                         if (c->retry_count < MAX_CMD_RETRIES) {
2178                                                 printk(KERN_WARNING
2179                                                        "cciss%d: retrying %p\n",
2180                                                        ctlr, c);
2181                                                 c->retry_count++;
2182                                                 /* erase the old error */
2183                                                 /* information */
2184                                                 memset(c->err_info, 0,
2185                                                        sizeof
2186                                                        (ErrorInfo_struct));
2187                                                 goto resend_cmd1;
2188                                         } else {
2189                                                 printk(KERN_WARNING
2190                                                        "cciss%d: retried %p too "
2191                                                        "many times\n", ctlr, c);
2192                                                 status = IO_ERROR;
2193                                                 goto cleanup1;
2194                                         }
2195                                 } else if (c->err_info->CommandStatus ==
2196                                            CMD_UNABORTABLE) {
2197                                         printk(KERN_WARNING
2198                                                "cciss%d: command could not be aborted.\n",
2199                                                ctlr);
2200                                         status = IO_ERROR;
2201                                         goto cleanup1;
2202                                 }
2203                                 printk(KERN_WARNING "ciss ciss%d: sendcmd"
2204                                        " Error %x \n", ctlr,
2205                                        c->err_info->CommandStatus);
2206                                 printk(KERN_WARNING "ciss ciss%d: sendcmd"
2207                                        " offensive info\n"
2208                                        "  size %x\n   num %x   value %x\n",
2209                                        ctlr,
2210                                        c->err_info->MoreErrInfo.Invalid_Cmd.
2211                                        offense_size,
2212                                        c->err_info->MoreErrInfo.Invalid_Cmd.
2213                                        offense_num,
2214                                        c->err_info->MoreErrInfo.Invalid_Cmd.
2215                                        offense_value);
2216                                 status = IO_ERROR;
2217                                 goto cleanup1;
2218                         }
2219                 }
2220                 /* This will need changing for direct lookup completions */
2221                 if (complete != c->busaddr) {
2222                         if (add_sendcmd_reject(cmd, ctlr, complete) != 0) {
2223                                 BUG();  /* we are pretty much hosed if we get here. */
2224                         }
2225                         continue;
2226                 } else
2227                         done = 1;
2228         } while (!done);
2229
2230       cleanup1:
2231         /* unlock the data buffer from DMA */
2232         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2233         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2234         pci_unmap_single(info_p->pdev, (dma_addr_t) buff_dma_handle.val,
2235                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2236 #ifdef CONFIG_CISS_SCSI_TAPE
2237         /* if we saved some commands for later, process them now. */
2238         if (info_p->scsi_rejects.ncompletions > 0)
2239                 do_cciss_intr(0, info_p, NULL);
2240 #endif
2241         cmd_free(info_p, c, 1);
2242         return status;
2243 }
2244
2245 /*
2246  * Map (physical) PCI mem into (virtual) kernel space
2247  */
2248 static void __iomem *remap_pci_mem(ulong base, ulong size)
2249 {
2250         ulong page_base = ((ulong) base) & PAGE_MASK;
2251         ulong page_offs = ((ulong) base) - page_base;
2252         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2253
2254         return page_remapped ? (page_remapped + page_offs) : NULL;
2255 }
2256
2257 /*
2258  * Takes jobs of the Q and sends them to the hardware, then puts it on
2259  * the Q to wait for completion.
2260  */
2261 static void start_io(ctlr_info_t *h)
2262 {
2263         CommandList_struct *c;
2264
2265         while ((c = h->reqQ) != NULL) {
2266                 /* can't do anything if fifo is full */
2267                 if ((h->access.fifo_full(h))) {
2268                         printk(KERN_WARNING "cciss: fifo full\n");
2269                         break;
2270                 }
2271
2272                 /* Get the first entry from the Request Q */
2273                 removeQ(&(h->reqQ), c);
2274                 h->Qdepth--;
2275
2276                 /* Tell the controller execute command */
2277                 h->access.submit_command(h, c);
2278
2279                 /* Put job onto the completed Q */
2280                 addQ(&(h->cmpQ), c);
2281         }
2282 }
2283
2284 /* Assumes that CCISS_LOCK(h->ctlr) is held. */
2285 /* Zeros out the error record and then resends the command back */
2286 /* to the controller */
2287 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2288 {
2289         /* erase the old error information */
2290         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2291
2292         /* add it to software queue and then send it to the controller */
2293         addQ(&(h->reqQ), c);
2294         h->Qdepth++;
2295         if (h->Qdepth > h->maxQsinceinit)
2296                 h->maxQsinceinit = h->Qdepth;
2297
2298         start_io(h);
2299 }
2300
2301 /* checks the status of the job and calls complete buffers to mark all
2302  * buffers for the completed job. Note that this function does not need
2303  * to hold the hba/queue lock.
2304  */
2305 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
2306                                     int timeout)
2307 {
2308         int status = 1;
2309         int retry_cmd = 0;
2310
2311         if (timeout)
2312                 status = 0;
2313
2314         if (cmd->err_info->CommandStatus != 0) {        /* an error has occurred */
2315                 switch (cmd->err_info->CommandStatus) {
2316                         unsigned char sense_key;
2317                 case CMD_TARGET_STATUS:
2318                         status = 0;
2319
2320                         if (cmd->err_info->ScsiStatus == 0x02) {
2321                                 printk(KERN_WARNING "cciss: cmd %p "
2322                                        "has CHECK CONDITION "
2323                                        " byte 2 = 0x%x\n", cmd,
2324                                        cmd->err_info->SenseInfo[2]
2325                                     );
2326                                 /* check the sense key */
2327                                 sense_key = 0xf & cmd->err_info->SenseInfo[2];
2328                                 /* no status or recovered error */
2329                                 if ((sense_key == 0x0) || (sense_key == 0x1)) {
2330                                         status = 1;
2331                                 }
2332                         } else {
2333                                 printk(KERN_WARNING "cciss: cmd %p "
2334                                        "has SCSI Status 0x%x\n",
2335                                        cmd, cmd->err_info->ScsiStatus);
2336                         }
2337                         break;
2338                 case CMD_DATA_UNDERRUN:
2339                         printk(KERN_WARNING "cciss: cmd %p has"
2340                                " completed with data underrun "
2341                                "reported\n", cmd);
2342                         break;
2343                 case CMD_DATA_OVERRUN:
2344                         printk(KERN_WARNING "cciss: cmd %p has"
2345                                " completed with data overrun "
2346                                "reported\n", cmd);
2347                         break;
2348                 case CMD_INVALID:
2349                         printk(KERN_WARNING "cciss: cmd %p is "
2350                                "reported invalid\n", cmd);
2351                         status = 0;
2352                         break;
2353                 case CMD_PROTOCOL_ERR:
2354                         printk(KERN_WARNING "cciss: cmd %p has "
2355                                "protocol error \n", cmd);
2356                         status = 0;
2357                         break;
2358                 case CMD_HARDWARE_ERR:
2359                         printk(KERN_WARNING "cciss: cmd %p had "
2360                                " hardware error\n", cmd);
2361                         status = 0;
2362                         break;
2363                 case CMD_CONNECTION_LOST:
2364                         printk(KERN_WARNING "cciss: cmd %p had "
2365                                "connection lost\n", cmd);
2366                         status = 0;
2367                         break;
2368                 case CMD_ABORTED:
2369                         printk(KERN_WARNING "cciss: cmd %p was "
2370                                "aborted\n", cmd);
2371                         status = 0;
2372                         break;
2373                 case CMD_ABORT_FAILED:
2374                         printk(KERN_WARNING "cciss: cmd %p reports "
2375                                "abort failed\n", cmd);
2376                         status = 0;
2377                         break;
2378                 case CMD_UNSOLICITED_ABORT:
2379                         printk(KERN_WARNING "cciss%d: unsolicited "
2380                                "abort %p\n", h->ctlr, cmd);
2381                         if (cmd->retry_count < MAX_CMD_RETRIES) {
2382                                 retry_cmd = 1;
2383                                 printk(KERN_WARNING
2384                                        "cciss%d: retrying %p\n", h->ctlr, cmd);
2385                                 cmd->retry_count++;
2386                         } else
2387                                 printk(KERN_WARNING
2388                                        "cciss%d: %p retried too "
2389                                        "many times\n", h->ctlr, cmd);
2390                         status = 0;
2391                         break;
2392                 case CMD_TIMEOUT:
2393                         printk(KERN_WARNING "cciss: cmd %p timedout\n", cmd);
2394                         status = 0;
2395                         break;
2396                 default:
2397                         printk(KERN_WARNING "cciss: cmd %p returned "
2398                                "unknown status %x\n", cmd,
2399                                cmd->err_info->CommandStatus);
2400                         status = 0;
2401                 }
2402         }
2403         /* We need to return this command */
2404         if (retry_cmd) {
2405                 resend_cciss_cmd(h, cmd);
2406                 return;
2407         }
2408
2409         cmd->rq->completion_data = cmd;
2410         cmd->rq->errors = status;
2411         blk_add_trace_rq(cmd->rq->q, cmd->rq, BLK_TA_COMPLETE);
2412         blk_complete_request(cmd->rq);
2413 }
2414
2415 /*
2416  * Get a request and submit it to the controller.
2417  */
2418 static void do_cciss_request(request_queue_t *q)
2419 {
2420         ctlr_info_t *h = q->queuedata;
2421         CommandList_struct *c;
2422         int start_blk, seg;
2423         struct request *creq;
2424         u64bit temp64;
2425         struct scatterlist tmp_sg[MAXSGENTRIES];
2426         drive_info_struct *drv;
2427         int i, dir;
2428
2429         /* We call start_io here in case there is a command waiting on the
2430          * queue that has not been sent.
2431          */
2432         if (blk_queue_plugged(q))
2433                 goto startio;
2434
2435       queue:
2436         creq = elv_next_request(q);
2437         if (!creq)
2438                 goto startio;
2439
2440         BUG_ON(creq->nr_phys_segments > MAXSGENTRIES);
2441
2442         if ((c = cmd_alloc(h, 1)) == NULL)
2443                 goto full;
2444
2445         blkdev_dequeue_request(creq);
2446
2447         spin_unlock_irq(q->queue_lock);
2448
2449         c->cmd_type = CMD_RWREQ;
2450         c->rq = creq;
2451
2452         /* fill in the request */
2453         drv = creq->rq_disk->private_data;
2454         c->Header.ReplyQueue = 0;       // unused in simple mode
2455         /* got command from pool, so use the command block index instead */
2456         /* for direct lookups. */
2457         /* The first 2 bits are reserved for controller error reporting. */
2458         c->Header.Tag.lower = (c->cmdindex << 3);
2459         c->Header.Tag.lower |= 0x04;    /* flag for direct lookup. */
2460         c->Header.LUN.LogDev.VolId = drv->LunID;
2461         c->Header.LUN.LogDev.Mode = 1;
2462         c->Request.CDBLen = 10; // 12 byte commands not in FW yet;
2463         c->Request.Type.Type = TYPE_CMD;        // It is a command.
2464         c->Request.Type.Attribute = ATTR_SIMPLE;
2465         c->Request.Type.Direction =
2466             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
2467         c->Request.Timeout = 0; // Don't time out
2468         c->Request.CDB[0] =
2469             (rq_data_dir(creq) == READ) ? CCISS_READ : CCISS_WRITE;
2470         start_blk = creq->sector;
2471 #ifdef CCISS_DEBUG
2472         printk(KERN_DEBUG "ciss: sector =%d nr_sectors=%d\n", (int)creq->sector,
2473                (int)creq->nr_sectors);
2474 #endif                          /* CCISS_DEBUG */
2475
2476         seg = blk_rq_map_sg(q, creq, tmp_sg);
2477
2478         /* get the DMA records for the setup */
2479         if (c->Request.Type.Direction == XFER_READ)
2480                 dir = PCI_DMA_FROMDEVICE;
2481         else
2482                 dir = PCI_DMA_TODEVICE;
2483
2484         for (i = 0; i < seg; i++) {
2485                 c->SG[i].Len = tmp_sg[i].length;
2486                 temp64.val = (__u64) pci_map_page(h->pdev, tmp_sg[i].page,
2487                                                   tmp_sg[i].offset,
2488                                                   tmp_sg[i].length, dir);
2489                 c->SG[i].Addr.lower = temp64.val32.lower;
2490                 c->SG[i].Addr.upper = temp64.val32.upper;
2491                 c->SG[i].Ext = 0;       // we are not chaining
2492         }
2493         /* track how many SG entries we are using */
2494         if (seg > h->maxSG)
2495                 h->maxSG = seg;
2496
2497 #ifdef CCISS_DEBUG
2498         printk(KERN_DEBUG "cciss: Submitting %d sectors in %d segments\n",
2499                creq->nr_sectors, seg);
2500 #endif                          /* CCISS_DEBUG */
2501
2502         c->Header.SGList = c->Header.SGTotal = seg;
2503         c->Request.CDB[1] = 0;
2504         c->Request.CDB[2] = (start_blk >> 24) & 0xff;   //MSB
2505         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
2506         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
2507         c->Request.CDB[5] = start_blk & 0xff;
2508         c->Request.CDB[6] = 0;  // (sect >> 24) & 0xff; MSB
2509         c->Request.CDB[7] = (creq->nr_sectors >> 8) & 0xff;
2510         c->Request.CDB[8] = creq->nr_sectors & 0xff;
2511         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
2512
2513         spin_lock_irq(q->queue_lock);
2514
2515         addQ(&(h->reqQ), c);
2516         h->Qdepth++;
2517         if (h->Qdepth > h->maxQsinceinit)
2518                 h->maxQsinceinit = h->Qdepth;
2519
2520         goto queue;
2521       full:
2522         blk_stop_queue(q);
2523       startio:
2524         /* We will already have the driver lock here so not need
2525          * to lock it.
2526          */
2527         start_io(h);
2528 }
2529
2530 static inline unsigned long get_next_completion(ctlr_info_t *h)
2531 {
2532 #ifdef CONFIG_CISS_SCSI_TAPE
2533         /* Any rejects from sendcmd() lying around? Process them first */
2534         if (h->scsi_rejects.ncompletions == 0)
2535                 return h->access.command_completed(h);
2536         else {
2537                 struct sendcmd_reject_list *srl;
2538                 int n;
2539                 srl = &h->scsi_rejects;
2540                 n = --srl->ncompletions;
2541                 /* printk("cciss%d: processing saved reject\n", h->ctlr); */
2542                 printk("p");
2543                 return srl->complete[n];
2544         }
2545 #else
2546         return h->access.command_completed(h);
2547 #endif
2548 }
2549
2550 static inline int interrupt_pending(ctlr_info_t *h)
2551 {
2552 #ifdef CONFIG_CISS_SCSI_TAPE
2553         return (h->access.intr_pending(h)
2554                 || (h->scsi_rejects.ncompletions > 0));
2555 #else
2556         return h->access.intr_pending(h);
2557 #endif
2558 }
2559
2560 static inline long interrupt_not_for_us(ctlr_info_t *h)
2561 {
2562 #ifdef CONFIG_CISS_SCSI_TAPE
2563         return (((h->access.intr_pending(h) == 0) ||
2564                  (h->interrupts_enabled == 0))
2565                 && (h->scsi_rejects.ncompletions == 0));
2566 #else
2567         return (((h->access.intr_pending(h) == 0) ||
2568                  (h->interrupts_enabled == 0)));
2569 #endif
2570 }
2571
2572 static irqreturn_t do_cciss_intr(int irq, void *dev_id, struct pt_regs *regs)
2573 {
2574         ctlr_info_t *h = dev_id;
2575         CommandList_struct *c;
2576         unsigned long flags;
2577         __u32 a, a1, a2;
2578
2579         if (interrupt_not_for_us(h))
2580                 return IRQ_NONE;
2581         /*
2582          * If there are completed commands in the completion queue,
2583          * we had better do something about it.
2584          */
2585         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2586         while (interrupt_pending(h)) {
2587                 while ((a = get_next_completion(h)) != FIFO_EMPTY) {
2588                         a1 = a;
2589                         if ((a & 0x04)) {
2590                                 a2 = (a >> 3);
2591                                 if (a2 >= NR_CMDS) {
2592                                         printk(KERN_WARNING
2593                                                "cciss: controller cciss%d failed, stopping.\n",
2594                                                h->ctlr);
2595                                         fail_all_cmds(h->ctlr);
2596                                         return IRQ_HANDLED;
2597                                 }
2598
2599                                 c = h->cmd_pool + a2;
2600                                 a = c->busaddr;
2601
2602                         } else {
2603                                 a &= ~3;
2604                                 if ((c = h->cmpQ) == NULL) {
2605                                         printk(KERN_WARNING
2606                                                "cciss: Completion of %08x ignored\n",
2607                                                a1);
2608                                         continue;
2609                                 }
2610                                 while (c->busaddr != a) {
2611                                         c = c->next;
2612                                         if (c == h->cmpQ)
2613                                                 break;
2614                                 }
2615                         }
2616                         /*
2617                          * If we've found the command, take it off the
2618                          * completion Q and free it
2619                          */
2620                         if (c->busaddr == a) {
2621                                 removeQ(&h->cmpQ, c);
2622                                 if (c->cmd_type == CMD_RWREQ) {
2623                                         complete_command(h, c, 0);
2624                                 } else if (c->cmd_type == CMD_IOCTL_PEND) {
2625                                         complete(c->waiting);
2626                                 }
2627 #                               ifdef CONFIG_CISS_SCSI_TAPE
2628                                 else if (c->cmd_type == CMD_SCSI)
2629                                         complete_scsi_command(c, 0, a1);
2630 #                               endif
2631                                 continue;
2632                         }
2633                 }
2634         }
2635
2636         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2637         return IRQ_HANDLED;
2638 }
2639
2640 /*
2641  *  We cannot read the structure directly, for portability we must use
2642  *   the io functions.
2643  *   This is for debug only.
2644  */
2645 #ifdef CCISS_DEBUG
2646 static void print_cfg_table(CfgTable_struct *tb)
2647 {
2648         int i;
2649         char temp_name[17];
2650
2651         printk("Controller Configuration information\n");
2652         printk("------------------------------------\n");
2653         for (i = 0; i < 4; i++)
2654                 temp_name[i] = readb(&(tb->Signature[i]));
2655         temp_name[4] = '\0';
2656         printk("   Signature = %s\n", temp_name);
2657         printk("   Spec Number = %d\n", readl(&(tb->SpecValence)));
2658         printk("   Transport methods supported = 0x%x\n",
2659                readl(&(tb->TransportSupport)));
2660         printk("   Transport methods active = 0x%x\n",
2661                readl(&(tb->TransportActive)));
2662         printk("   Requested transport Method = 0x%x\n",
2663                readl(&(tb->HostWrite.TransportRequest)));
2664         printk("   Coalesce Interrupt Delay = 0x%x\n",
2665                readl(&(tb->HostWrite.CoalIntDelay)));
2666         printk("   Coalesce Interrupt Count = 0x%x\n",
2667                readl(&(tb->HostWrite.CoalIntCount)));
2668         printk("   Max outstanding commands = 0x%d\n",
2669                readl(&(tb->CmdsOutMax)));
2670         printk("   Bus Types = 0x%x\n", readl(&(tb->BusTypes)));
2671         for (i = 0; i < 16; i++)
2672                 temp_name[i] = readb(&(tb->ServerName[i]));
2673         temp_name[16] = '\0';
2674         printk("   Server Name = %s\n", temp_name);
2675         printk("   Heartbeat Counter = 0x%x\n\n\n", readl(&(tb->HeartBeat)));
2676 }
2677 #endif                          /* CCISS_DEBUG */
2678
2679 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
2680 {
2681         int i, offset, mem_type, bar_type;
2682         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
2683                 return 0;
2684         offset = 0;
2685         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
2686                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
2687                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
2688                         offset += 4;
2689                 else {
2690                         mem_type = pci_resource_flags(pdev, i) &
2691                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
2692                         switch (mem_type) {
2693                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
2694                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
2695                                 offset += 4;    /* 32 bit */
2696                                 break;
2697                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
2698                                 offset += 8;
2699                                 break;
2700                         default:        /* reserved in PCI 2.2 */
2701                                 printk(KERN_WARNING
2702                                        "Base address is invalid\n");
2703                                 return -1;
2704                                 break;
2705                         }
2706                 }
2707                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
2708                         return i + 1;
2709         }
2710         return -1;
2711 }
2712
2713 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
2714  * controllers that are capable. If not, we use IO-APIC mode.
2715  */
2716
2717 static void __devinit cciss_interrupt_mode(ctlr_info_t *c,
2718                                            struct pci_dev *pdev, __u32 board_id)
2719 {
2720 #ifdef CONFIG_PCI_MSI
2721         int err;
2722         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
2723         {0, 2}, {0, 3}
2724         };
2725
2726         /* Some boards advertise MSI but don't really support it */
2727         if ((board_id == 0x40700E11) ||
2728             (board_id == 0x40800E11) ||
2729             (board_id == 0x40820E11) || (board_id == 0x40830E11))
2730                 goto default_int_mode;
2731
2732         if (pci_find_capability(pdev, PCI_CAP_ID_MSIX)) {
2733                 err = pci_enable_msix(pdev, cciss_msix_entries, 4);
2734                 if (!err) {
2735                         c->intr[0] = cciss_msix_entries[0].vector;
2736                         c->intr[1] = cciss_msix_entries[1].vector;
2737                         c->intr[2] = cciss_msix_entries[2].vector;
2738                         c->intr[3] = cciss_msix_entries[3].vector;
2739                         c->msix_vector = 1;
2740                         return;
2741                 }
2742                 if (err > 0) {
2743                         printk(KERN_WARNING "cciss: only %d MSI-X vectors "
2744                                "available\n", err);
2745                 } else {
2746                         printk(KERN_WARNING "cciss: MSI-X init failed %d\n",
2747                                err);
2748                 }
2749         }
2750         if (pci_find_capability(pdev, PCI_CAP_ID_MSI)) {
2751                 if (!pci_enable_msi(pdev)) {
2752                         c->intr[SIMPLE_MODE_INT] = pdev->irq;
2753                         c->msi_vector = 1;
2754                         return;
2755                 } else {
2756                         printk(KERN_WARNING "cciss: MSI init failed\n");
2757                         c->intr[SIMPLE_MODE_INT] = pdev->irq;
2758                         return;
2759                 }
2760         }
2761       default_int_mode:
2762 #endif                          /* CONFIG_PCI_MSI */
2763         /* if we get here we're going to use the default interrupt mode */
2764         c->intr[SIMPLE_MODE_INT] = pdev->irq;
2765         return;
2766 }
2767
2768 static int cciss_pci_init(ctlr_info_t *c, struct pci_dev *pdev)
2769 {
2770         ushort subsystem_vendor_id, subsystem_device_id, command;
2771         __u32 board_id, scratchpad = 0;
2772         __u64 cfg_offset;
2773         __u32 cfg_base_addr;
2774         __u64 cfg_base_addr_index;
2775         int i, err;
2776
2777         /* check to see if controller has been disabled */
2778         /* BEFORE trying to enable it */
2779         (void)pci_read_config_word(pdev, PCI_COMMAND, &command);
2780         if (!(command & 0x02)) {
2781                 printk(KERN_WARNING
2782                        "cciss: controller appears to be disabled\n");
2783                 return -ENODEV;
2784         }
2785
2786         err = pci_enable_device(pdev);
2787         if (err) {
2788                 printk(KERN_ERR "cciss: Unable to Enable PCI device\n");
2789                 return err;
2790         }
2791
2792         err = pci_request_regions(pdev, "cciss");
2793         if (err) {
2794                 printk(KERN_ERR "cciss: Cannot obtain PCI resources, "
2795                        "aborting\n");
2796                 goto err_out_disable_pdev;
2797         }
2798
2799         subsystem_vendor_id = pdev->subsystem_vendor;
2800         subsystem_device_id = pdev->subsystem_device;
2801         board_id = (((__u32) (subsystem_device_id << 16) & 0xffff0000) |
2802                     subsystem_vendor_id);
2803
2804 #ifdef CCISS_DEBUG
2805         printk("command = %x\n", command);
2806         printk("irq = %x\n", pdev->irq);
2807         printk("board_id = %x\n", board_id);
2808 #endif                          /* CCISS_DEBUG */
2809
2810 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
2811  * else we use the IO-APIC interrupt assigned to us by system ROM.
2812  */
2813         cciss_interrupt_mode(c, pdev, board_id);
2814
2815         /*
2816          * Memory base addr is first addr , the second points to the config
2817          *   table
2818          */
2819
2820         c->paddr = pci_resource_start(pdev, 0); /* addressing mode bits already removed */
2821 #ifdef CCISS_DEBUG
2822         printk("address 0 = %x\n", c->paddr);
2823 #endif                          /* CCISS_DEBUG */
2824         c->vaddr = remap_pci_mem(c->paddr, 200);
2825
2826         /* Wait for the board to become ready.  (PCI hotplug needs this.)
2827          * We poll for up to 120 secs, once per 100ms. */
2828         for (i = 0; i < 1200; i++) {
2829                 scratchpad = readl(c->vaddr + SA5_SCRATCHPAD_OFFSET);
2830                 if (scratchpad == CCISS_FIRMWARE_READY)
2831                         break;
2832                 set_current_state(TASK_INTERRUPTIBLE);
2833                 schedule_timeout(HZ / 10);      /* wait 100ms */
2834         }
2835         if (scratchpad != CCISS_FIRMWARE_READY) {
2836                 printk(KERN_WARNING "cciss: Board not ready.  Timed out.\n");
2837                 err = -ENODEV;
2838                 goto err_out_free_res;
2839         }
2840
2841         /* get the address index number */
2842         cfg_base_addr = readl(c->vaddr + SA5_CTCFG_OFFSET);
2843         cfg_base_addr &= (__u32) 0x0000ffff;
2844 #ifdef CCISS_DEBUG
2845         printk("cfg base address = %x\n", cfg_base_addr);
2846 #endif                          /* CCISS_DEBUG */
2847         cfg_base_addr_index = find_PCI_BAR_index(pdev, cfg_base_addr);
2848 #ifdef CCISS_DEBUG
2849         printk("cfg base address index = %x\n", cfg_base_addr_index);
2850 #endif                          /* CCISS_DEBUG */
2851         if (cfg_base_addr_index == -1) {
2852                 printk(KERN_WARNING "cciss: Cannot find cfg_base_addr_index\n");
2853                 err = -ENODEV;
2854                 goto err_out_free_res;
2855         }
2856
2857         cfg_offset = readl(c->vaddr + SA5_CTMEM_OFFSET);
2858 #ifdef CCISS_DEBUG
2859         printk("cfg offset = %x\n", cfg_offset);
2860 #endif                          /* CCISS_DEBUG */
2861         c->cfgtable = remap_pci_mem(pci_resource_start(pdev,
2862                                                        cfg_base_addr_index) +
2863                                     cfg_offset, sizeof(CfgTable_struct));
2864         c->board_id = board_id;
2865
2866 #ifdef CCISS_DEBUG
2867         print_cfg_table(c->cfgtable);
2868 #endif                          /* CCISS_DEBUG */
2869
2870         for (i = 0; i < ARRAY_SIZE(products); i++) {
2871                 if (board_id == products[i].board_id) {
2872                         c->product_name = products[i].product_name;
2873                         c->access = *(products[i].access);
2874                         break;
2875                 }
2876         }
2877         if (i == ARRAY_SIZE(products)) {
2878                 printk(KERN_WARNING "cciss: Sorry, I don't know how"
2879                        " to access the Smart Array controller %08lx\n",
2880                        (unsigned long)board_id);
2881                 err = -ENODEV;
2882                 goto err_out_free_res;
2883         }
2884         if ((readb(&c->cfgtable->Signature[0]) != 'C') ||
2885             (readb(&c->cfgtable->Signature[1]) != 'I') ||
2886             (readb(&c->cfgtable->Signature[2]) != 'S') ||
2887             (readb(&c->cfgtable->Signature[3]) != 'S')) {
2888                 printk("Does not appear to be a valid CISS config table\n");
2889                 err = -ENODEV;
2890                 goto err_out_free_res;
2891         }
2892 #ifdef CONFIG_X86
2893         {
2894                 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
2895                 __u32 prefetch;
2896                 prefetch = readl(&(c->cfgtable->SCSI_Prefetch));
2897                 prefetch |= 0x100;
2898                 writel(prefetch, &(c->cfgtable->SCSI_Prefetch));
2899         }
2900 #endif
2901
2902 #ifdef CCISS_DEBUG
2903         printk("Trying to put board into Simple mode\n");
2904 #endif                          /* CCISS_DEBUG */
2905         c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
2906         /* Update the field, and then ring the doorbell */
2907         writel(CFGTBL_Trans_Simple, &(c->cfgtable->HostWrite.TransportRequest));
2908         writel(CFGTBL_ChangeReq, c->vaddr + SA5_DOORBELL);
2909
2910         /* under certain very rare conditions, this can take awhile.
2911          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
2912          * as we enter this code.) */
2913         for (i = 0; i < MAX_CONFIG_WAIT; i++) {
2914                 if (!(readl(c->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
2915                         break;
2916                 /* delay and try again */
2917                 set_current_state(TASK_INTERRUPTIBLE);
2918                 schedule_timeout(10);
2919         }
2920
2921 #ifdef CCISS_DEBUG
2922         printk(KERN_DEBUG "I counter got to %d %x\n", i,
2923                readl(c->vaddr + SA5_DOORBELL));
2924 #endif                          /* CCISS_DEBUG */
2925 #ifdef CCISS_DEBUG
2926         print_cfg_table(c->cfgtable);
2927 #endif                          /* CCISS_DEBUG */
2928
2929         if (!(readl(&(c->cfgtable->TransportActive)) & CFGTBL_Trans_Simple)) {
2930                 printk(KERN_WARNING "cciss: unable to get board into"
2931                        " simple mode\n");
2932                 err = -ENODEV;
2933                 goto err_out_free_res;
2934         }
2935         return 0;
2936
2937       err_out_free_res:
2938         pci_release_regions(pdev);
2939
2940       err_out_disable_pdev:
2941         pci_disable_device(pdev);
2942         return err;
2943 }
2944
2945 /*
2946  * Gets information about the local volumes attached to the controller.
2947  */
2948 static void cciss_getgeometry(int cntl_num)
2949 {
2950         ReportLunData_struct *ld_buff;
2951         ReadCapdata_struct *size_buff;
2952         InquiryData_struct *inq_buff;
2953         int return_code;
2954         int i;
2955         int listlength = 0;
2956         __u32 lunid = 0;
2957         int block_size;
2958         int total_size;
2959
2960         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2961         if (ld_buff == NULL) {
2962                 printk(KERN_ERR "cciss: out of memory\n");
2963                 return;
2964         }
2965         size_buff = kmalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2966         if (size_buff == NULL) {
2967                 printk(KERN_ERR "cciss: out of memory\n");
2968                 kfree(ld_buff);
2969                 return;
2970         }
2971         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2972         if (inq_buff == NULL) {
2973                 printk(KERN_ERR "cciss: out of memory\n");
2974                 kfree(ld_buff);
2975                 kfree(size_buff);
2976                 return;
2977         }
2978         /* Get the firmware version */
2979         return_code = sendcmd(CISS_INQUIRY, cntl_num, inq_buff,
2980                               sizeof(InquiryData_struct), 0, 0, 0, NULL,
2981                               TYPE_CMD);
2982         if (return_code == IO_OK) {
2983                 hba[cntl_num]->firm_ver[0] = inq_buff->data_byte[32];
2984                 hba[cntl_num]->firm_ver[1] = inq_buff->data_byte[33];
2985                 hba[cntl_num]->firm_ver[2] = inq_buff->data_byte[34];
2986                 hba[cntl_num]->firm_ver[3] = inq_buff->data_byte[35];
2987         } else {                /* send command failed */
2988
2989                 printk(KERN_WARNING "cciss: unable to determine firmware"
2990                        " version of controller\n");
2991         }
2992         /* Get the number of logical volumes */
2993         return_code = sendcmd(CISS_REPORT_LOG, cntl_num, ld_buff,
2994                               sizeof(ReportLunData_struct), 0, 0, 0, NULL,
2995                               TYPE_CMD);
2996
2997         if (return_code == IO_OK) {
2998 #ifdef CCISS_DEBUG
2999                 printk("LUN Data\n--------------------------\n");
3000 #endif                          /* CCISS_DEBUG */
3001
3002                 listlength |=
3003                     (0xff & (unsigned int)(ld_buff->LUNListLength[0])) << 24;
3004                 listlength |=
3005                     (0xff & (unsigned int)(ld_buff->LUNListLength[1])) << 16;
3006                 listlength |=
3007                     (0xff & (unsigned int)(ld_buff->LUNListLength[2])) << 8;
3008                 listlength |= 0xff & (unsigned int)(ld_buff->LUNListLength[3]);
3009         } else {                /* reading number of logical volumes failed */
3010
3011                 printk(KERN_WARNING "cciss: report logical volume"
3012                        " command failed\n");
3013                 listlength = 0;
3014         }
3015         hba[cntl_num]->num_luns = listlength / 8;       // 8 bytes pre entry
3016         if (hba[cntl_num]->num_luns > CISS_MAX_LUN) {
3017                 printk(KERN_ERR
3018                        "ciss:  only %d number of logical volumes supported\n",
3019                        CISS_MAX_LUN);
3020                 hba[cntl_num]->num_luns = CISS_MAX_LUN;
3021         }
3022 #ifdef CCISS_DEBUG
3023         printk(KERN_DEBUG "Length = %x %x %x %x = %d\n",
3024                ld_buff->LUNListLength[0], ld_buff->LUNListLength[1],
3025                ld_buff->LUNListLength[2], ld_buff->LUNListLength[3],
3026                hba[cntl_num]->num_luns);
3027 #endif                          /* CCISS_DEBUG */
3028
3029         hba[cntl_num]->highest_lun = hba[cntl_num]->num_luns - 1;
3030 //      for(i=0; i<  hba[cntl_num]->num_luns; i++)
3031         for (i = 0; i < CISS_MAX_LUN; i++) {
3032                 if (i < hba[cntl_num]->num_luns) {
3033                         lunid = (0xff & (unsigned int)(ld_buff->LUN[i][3]))
3034                             << 24;
3035                         lunid |= (0xff & (unsigned int)(ld_buff->LUN[i][2]))
3036                             << 16;
3037                         lunid |= (0xff & (unsigned int)(ld_buff->LUN[i][1]))
3038                             << 8;
3039                         lunid |= 0xff & (unsigned int)(ld_buff->LUN[i][0]);
3040
3041                         hba[cntl_num]->drv[i].LunID = lunid;
3042
3043 #ifdef CCISS_DEBUG
3044                         printk(KERN_DEBUG "LUN[%d]:  %x %x %x %x = %x\n", i,
3045                                ld_buff->LUN[i][0], ld_buff->LUN[i][1],
3046                                ld_buff->LUN[i][2], ld_buff->LUN[i][3],
3047                                hba[cntl_num]->drv[i].LunID);
3048 #endif                          /* CCISS_DEBUG */
3049                         cciss_read_capacity(cntl_num, i, size_buff, 0,
3050                                             &total_size, &block_size);
3051                         cciss_geometry_inquiry(cntl_num, i, 0, total_size,
3052                                                block_size, inq_buff,
3053                                                &hba[cntl_num]->drv[i]);
3054                 } else {
3055                         /* initialize raid_level to indicate a free space */
3056                         hba[cntl_num]->drv[i].raid_level = -1;
3057                 }
3058         }
3059         kfree(ld_buff);
3060         kfree(size_buff);
3061         kfree(inq_buff);
3062 }
3063
3064 /* Function to find the first free pointer into our hba[] array */
3065 /* Returns -1 if no free entries are left.  */
3066 static int alloc_cciss_hba(void)
3067 {
3068         struct gendisk *disk[NWD];
3069         int i, n;
3070         for (n = 0; n < NWD; n++) {
3071                 disk[n] = alloc_disk(1 << NWD_SHIFT);
3072                 if (!disk[n])
3073                         goto out;
3074         }
3075
3076         for (i = 0; i < MAX_CTLR; i++) {
3077                 if (!hba[i]) {
3078                         ctlr_info_t *p;
3079                         p = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
3080                         if (!p)
3081                                 goto Enomem;
3082                         for (n = 0; n < NWD; n++)
3083                                 p->gendisk[n] = disk[n];
3084                         hba[i] = p;
3085                         return i;
3086                 }
3087         }
3088         printk(KERN_WARNING "cciss: This driver supports a maximum"
3089                " of %d controllers.\n", MAX_CTLR);
3090         goto out;
3091       Enomem:
3092         printk(KERN_ERR "cciss: out of memory.\n");
3093       out:
3094         while (n--)
3095                 put_disk(disk[n]);
3096         return -1;
3097 }
3098
3099 static void free_hba(int i)
3100 {
3101         ctlr_info_t *p = hba[i];
3102         int n;
3103
3104         hba[i] = NULL;
3105         for (n = 0; n < NWD; n++)
3106                 put_disk(p->gendisk[n]);
3107         kfree(p);
3108 }
3109
3110 /*
3111  *  This is it.  Find all the controllers and register them.  I really hate
3112  *  stealing all these major device numbers.
3113  *  returns the number of block devices registered.
3114  */
3115 static int __devinit cciss_init_one(struct pci_dev *pdev,
3116                                     const struct pci_device_id *ent)
3117 {
3118         request_queue_t *q;
3119         int i;
3120         int j;
3121         int rc;
3122         int dac;
3123
3124         i = alloc_cciss_hba();
3125         if (i < 0)
3126                 return -1;
3127
3128         hba[i]->busy_initializing = 1;
3129
3130         if (cciss_pci_init(hba[i], pdev) != 0)
3131                 goto clean1;
3132
3133         sprintf(hba[i]->devname, "cciss%d", i);
3134         hba[i]->ctlr = i;
3135         hba[i]->pdev = pdev;
3136
3137         /* configure PCI DMA stuff */
3138         if (!pci_set_dma_mask(pdev, DMA_64BIT_MASK))
3139                 dac = 1;
3140         else if (!pci_set_dma_mask(pdev, DMA_32BIT_MASK))
3141                 dac = 0;
3142         else {
3143                 printk(KERN_ERR "cciss: no suitable DMA available\n");
3144                 goto clean1;
3145         }
3146
3147         /*
3148          * register with the major number, or get a dynamic major number
3149          * by passing 0 as argument.  This is done for greater than
3150          * 8 controller support.
3151          */
3152         if (i < MAX_CTLR_ORIG)
3153                 hba[i]->major = COMPAQ_CISS_MAJOR + i;
3154         rc = register_blkdev(hba[i]->major, hba[i]->devname);
3155         if (rc == -EBUSY || rc == -EINVAL) {
3156                 printk(KERN_ERR
3157                        "cciss:  Unable to get major number %d for %s "
3158                        "on hba %d\n", hba[i]->major, hba[i]->devname, i);
3159                 goto clean1;
3160         } else {
3161                 if (i >= MAX_CTLR_ORIG)
3162                         hba[i]->major = rc;
3163         }
3164
3165         /* make sure the board interrupts are off */
3166         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_OFF);
3167         if (request_irq(hba[i]->intr[SIMPLE_MODE_INT], do_cciss_intr,
3168                         IRQF_DISABLED | IRQF_SHARED, hba[i]->devname, hba[i])) {
3169                 printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
3170                        hba[i]->intr[SIMPLE_MODE_INT], hba[i]->devname);
3171                 goto clean2;
3172         }
3173
3174         printk(KERN_INFO "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
3175                hba[i]->devname, pdev->device, pci_name(pdev),
3176                hba[i]->intr[SIMPLE_MODE_INT], dac ? "" : " not");
3177
3178         hba[i]->cmd_pool_bits =
3179             kmalloc(((NR_CMDS + BITS_PER_LONG -
3180                       1) / BITS_PER_LONG) * sizeof(unsigned long), GFP_KERNEL);
3181         hba[i]->cmd_pool = (CommandList_struct *)
3182             pci_alloc_consistent(hba[i]->pdev,
3183                     NR_CMDS * sizeof(CommandList_struct),
3184                     &(hba[i]->cmd_pool_dhandle));
3185         hba[i]->errinfo_pool = (ErrorInfo_struct *)
3186             pci_alloc_consistent(hba[i]->pdev,
3187                     NR_CMDS * sizeof(ErrorInfo_struct),
3188                     &(hba[i]->errinfo_pool_dhandle));
3189         if ((hba[i]->cmd_pool_bits == NULL)
3190             || (hba[i]->cmd_pool == NULL)
3191             || (hba[i]->errinfo_pool == NULL)) {
3192                 printk(KERN_ERR "cciss: out of memory");
3193                 goto clean4;
3194         }
3195 #ifdef CONFIG_CISS_SCSI_TAPE
3196         hba[i]->scsi_rejects.complete =
3197             kmalloc(sizeof(hba[i]->scsi_rejects.complete[0]) *
3198                     (NR_CMDS + 5), GFP_KERNEL);
3199         if (hba[i]->scsi_rejects.complete == NULL) {
3200                 printk(KERN_ERR "cciss: out of memory");
3201                 goto clean4;
3202         }
3203 #endif
3204         spin_lock_init(&hba[i]->lock);
3205
3206         /* Initialize the pdev driver private data.
3207            have it point to hba[i].  */
3208         pci_set_drvdata(pdev, hba[i]);
3209         /* command and error info recs zeroed out before
3210            they are used */
3211         memset(hba[i]->cmd_pool_bits, 0,
3212                ((NR_CMDS + BITS_PER_LONG -
3213                  1) / BITS_PER_LONG) * sizeof(unsigned long));
3214
3215 #ifdef CCISS_DEBUG
3216         printk(KERN_DEBUG "Scanning for drives on controller cciss%d\n", i);
3217 #endif                          /* CCISS_DEBUG */
3218
3219         cciss_getgeometry(i);
3220
3221         cciss_scsi_setup(i);
3222
3223         /* Turn the interrupts on so we can service requests */
3224         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_ON);
3225
3226         cciss_procinit(i);
3227         hba[i]->busy_initializing = 0;
3228
3229         for (j = 0; j < NWD; j++) {     /* mfm */
3230                 drive_info_struct *drv = &(hba[i]->drv[j]);
3231                 struct gendisk *disk = hba[i]->gendisk[j];
3232
3233                 q = blk_init_queue(do_cciss_request, &hba[i]->lock);
3234                 if (!q) {
3235                         printk(KERN_ERR
3236                                "cciss:  unable to allocate queue for disk %d\n",
3237                                j);
3238                         break;
3239                 }
3240                 drv->queue = q;
3241
3242                 q->backing_dev_info.ra_pages = READ_AHEAD;
3243                 blk_queue_bounce_limit(q, hba[i]->pdev->dma_mask);
3244
3245                 /* This is a hardware imposed limit. */
3246                 blk_queue_max_hw_segments(q, MAXSGENTRIES);
3247
3248                 /* This is a limit in the driver and could be eliminated. */
3249                 blk_queue_max_phys_segments(q, MAXSGENTRIES);
3250
3251                 blk_queue_max_sectors(q, 512);
3252
3253                 blk_queue_softirq_done(q, cciss_softirq_done);
3254
3255                 q->queuedata = hba[i];
3256                 sprintf(disk->disk_name, "cciss/c%dd%d", i, j);
3257                 disk->major = hba[i]->major;
3258                 disk->first_minor = j << NWD_SHIFT;
3259                 disk->fops = &cciss_fops;
3260                 disk->queue = q;
3261                 disk->private_data = drv;
3262                 disk->driverfs_dev = &pdev->dev;
3263                 /* we must register the controller even if no disks exist */
3264                 /* this is for the online array utilities */
3265                 if (!drv->heads && j)
3266                         continue;
3267                 blk_queue_hardsect_size(q, drv->block_size);
3268                 set_capacity(disk, drv->nr_blocks);
3269                 add_disk(disk);
3270         }
3271
3272         return 1;
3273
3274       clean4:
3275 #ifdef CONFIG_CISS_SCSI_TAPE
3276         kfree(hba[i]->scsi_rejects.complete);
3277 #endif
3278         kfree(hba[i]->cmd_pool_bits);
3279         if (hba[i]->cmd_pool)
3280                 pci_free_consistent(hba[i]->pdev,
3281                                     NR_CMDS * sizeof(CommandList_struct),
3282                                     hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
3283         if (hba[i]->errinfo_pool)
3284                 pci_free_consistent(hba[i]->pdev,
3285                                     NR_CMDS * sizeof(ErrorInfo_struct),
3286                                     hba[i]->errinfo_pool,
3287                                     hba[i]->errinfo_pool_dhandle);
3288         free_irq(hba[i]->intr[SIMPLE_MODE_INT], hba[i]);
3289       clean2:
3290         unregister_blkdev(hba[i]->major, hba[i]->devname);
3291       clean1:
3292         hba[i]->busy_initializing = 0;
3293         free_hba(i);
3294         return -1;
3295 }
3296
3297 static void __devexit cciss_remove_one(struct pci_dev *pdev)
3298 {
3299         ctlr_info_t *tmp_ptr;
3300         int i, j;
3301         char flush_buf[4];
3302         int return_code;
3303
3304         if (pci_get_drvdata(pdev) == NULL) {
3305                 printk(KERN_ERR "cciss: Unable to remove device \n");
3306                 return;
3307         }
3308         tmp_ptr = pci_get_drvdata(pdev);
3309         i = tmp_ptr->ctlr;
3310         if (hba[i] == NULL) {
3311                 printk(KERN_ERR "cciss: device appears to "
3312                        "already be removed \n");
3313                 return;
3314         }
3315         /* Turn board interrupts off  and send the flush cache command */
3316         /* sendcmd will turn off interrupt, and send the flush...
3317          * To write all data in the battery backed cache to disks */
3318         memset(flush_buf, 0, 4);
3319         return_code = sendcmd(CCISS_CACHE_FLUSH, i, flush_buf, 4, 0, 0, 0, NULL,
3320                               TYPE_CMD);
3321         if (return_code != IO_OK) {
3322                 printk(KERN_WARNING "Error Flushing cache on controller %d\n",
3323                        i);
3324         }
3325         free_irq(hba[i]->intr[2], hba[i]);
3326
3327 #ifdef CONFIG_PCI_MSI
3328         if (hba[i]->msix_vector)
3329                 pci_disable_msix(hba[i]->pdev);
3330         else if (hba[i]->msi_vector)
3331                 pci_disable_msi(hba[i]->pdev);
3332 #endif                          /* CONFIG_PCI_MSI */
3333
3334         iounmap(hba[i]->vaddr);
3335         cciss_unregister_scsi(i);       /* unhook from SCSI subsystem */
3336         unregister_blkdev(hba[i]->major, hba[i]->devname);
3337         remove_proc_entry(hba[i]->devname, proc_cciss);
3338
3339         /* remove it from the disk list */
3340         for (j = 0; j < NWD; j++) {
3341                 struct gendisk *disk = hba[i]->gendisk[j];
3342                 if (disk) {
3343                         request_queue_t *q = disk->queue;
3344
3345                         if (disk->flags & GENHD_FL_UP)
3346                                 del_gendisk(disk);
3347                         if (q)
3348                                 blk_cleanup_queue(q);
3349                 }
3350         }
3351
3352         pci_free_consistent(hba[i]->pdev, NR_CMDS * sizeof(CommandList_struct),
3353                             hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
3354         pci_free_consistent(hba[i]->pdev, NR_CMDS * sizeof(ErrorInfo_struct),
3355                             hba[i]->errinfo_pool, hba[i]->errinfo_pool_dhandle);
3356         kfree(hba[i]->cmd_pool_bits);
3357 #ifdef CONFIG_CISS_SCSI_TAPE
3358         kfree(hba[i]->scsi_rejects.complete);
3359 #endif
3360         pci_release_regions(pdev);
3361         pci_disable_device(pdev);
3362         pci_set_drvdata(pdev, NULL);
3363         free_hba(i);
3364 }
3365
3366 static struct pci_driver cciss_pci_driver = {
3367         .name = "cciss",
3368         .probe = cciss_init_one,
3369         .remove = __devexit_p(cciss_remove_one),
3370         .id_table = cciss_pci_device_id,        /* id_table */
3371 };
3372
3373 /*
3374  *  This is it.  Register the PCI driver information for the cards we control
3375  *  the OS will call our registered routines when it finds one of our cards.
3376  */
3377 static int __init cciss_init(void)
3378 {
3379         printk(KERN_INFO DRIVER_NAME "\n");
3380
3381         /* Register for our PCI devices */
3382         return pci_register_driver(&cciss_pci_driver);
3383 }
3384
3385 static void __exit cciss_cleanup(void)
3386 {
3387         int i;
3388
3389         pci_unregister_driver(&cciss_pci_driver);
3390         /* double check that all controller entrys have been removed */
3391         for (i = 0; i < MAX_CTLR; i++) {
3392                 if (hba[i] != NULL) {
3393                         printk(KERN_WARNING "cciss: had to remove"
3394                                " controller %d\n", i);
3395                         cciss_remove_one(hba[i]->pdev);
3396                 }
3397         }
3398         remove_proc_entry("cciss", proc_root_driver);
3399 }
3400
3401 static void fail_all_cmds(unsigned long ctlr)
3402 {
3403         /* If we get here, the board is apparently dead. */
3404         ctlr_info_t *h = hba[ctlr];
3405         CommandList_struct *c;
3406         unsigned long flags;
3407
3408         printk(KERN_WARNING "cciss%d: controller not responding.\n", h->ctlr);
3409         h->alive = 0;           /* the controller apparently died... */
3410
3411         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
3412
3413         pci_disable_device(h->pdev);    /* Make sure it is really dead. */
3414
3415         /* move everything off the request queue onto the completed queue */
3416         while ((c = h->reqQ) != NULL) {
3417                 removeQ(&(h->reqQ), c);
3418                 h->Qdepth--;
3419                 addQ(&(h->cmpQ), c);
3420         }
3421
3422         /* Now, fail everything on the completed queue with a HW error */
3423         while ((c = h->cmpQ) != NULL) {
3424                 removeQ(&h->cmpQ, c);
3425                 c->err_info->CommandStatus = CMD_HARDWARE_ERR;
3426                 if (c->cmd_type == CMD_RWREQ) {
3427                         complete_command(h, c, 0);
3428                 } else if (c->cmd_type == CMD_IOCTL_PEND)
3429                         complete(c->waiting);
3430 #ifdef CONFIG_CISS_SCSI_TAPE
3431                 else if (c->cmd_type == CMD_SCSI)
3432                         complete_scsi_command(c, 0, 0);
3433 #endif
3434         }
3435         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
3436         return;
3437 }
3438
3439 module_init(cciss_init);
3440 module_exit(cciss_cleanup);