]> www.pilppa.org Git - linux-2.6-omap-h63xx.git/blob - drivers/block/cciss.c
blk_end_request: changing cciss (take 4)
[linux-2.6-omap-h63xx.git] / drivers / block / cciss.c
1 /*
2  *    Disk Array driver for HP Smart Array controllers.
3  *    (C) Copyright 2000, 2007 Hewlett-Packard Development Company, L.P.
4  *
5  *    This program is free software; you can redistribute it and/or modify
6  *    it under the terms of the GNU General Public License as published by
7  *    the Free Software Foundation; version 2 of the License.
8  *
9  *    This program is distributed in the hope that it will be useful,
10  *    but WITHOUT ANY WARRANTY; without even the implied warranty of
11  *    MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12  *    General Public License for more details.
13  *
14  *    You should have received a copy of the GNU General Public License
15  *    along with this program; if not, write to the Free Software
16  *    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
17  *    02111-1307, USA.
18  *
19  *    Questions/Comments/Bugfixes to iss_storagedev@hp.com
20  *
21  */
22
23 #include <linux/module.h>
24 #include <linux/interrupt.h>
25 #include <linux/types.h>
26 #include <linux/pci.h>
27 #include <linux/kernel.h>
28 #include <linux/slab.h>
29 #include <linux/delay.h>
30 #include <linux/major.h>
31 #include <linux/fs.h>
32 #include <linux/bio.h>
33 #include <linux/blkpg.h>
34 #include <linux/timer.h>
35 #include <linux/proc_fs.h>
36 #include <linux/init.h>
37 #include <linux/hdreg.h>
38 #include <linux/spinlock.h>
39 #include <linux/compat.h>
40 #include <linux/blktrace_api.h>
41 #include <asm/uaccess.h>
42 #include <asm/io.h>
43
44 #include <linux/dma-mapping.h>
45 #include <linux/blkdev.h>
46 #include <linux/genhd.h>
47 #include <linux/completion.h>
48 #include <scsi/scsi.h>
49 #include <scsi/sg.h>
50 #include <scsi/scsi_ioctl.h>
51 #include <linux/cdrom.h>
52
53 #define CCISS_DRIVER_VERSION(maj,min,submin) ((maj<<16)|(min<<8)|(submin))
54 #define DRIVER_NAME "HP CISS Driver (v 3.6.14)"
55 #define DRIVER_VERSION CCISS_DRIVER_VERSION(3,6,14)
56
57 /* Embedded module documentation macros - see modules.h */
58 MODULE_AUTHOR("Hewlett-Packard Company");
59 MODULE_DESCRIPTION("Driver for HP Controller SA5xxx SA6xxx version 3.6.14");
60 MODULE_SUPPORTED_DEVICE("HP SA5i SA5i+ SA532 SA5300 SA5312 SA641 SA642 SA6400"
61                         " SA6i P600 P800 P400 P400i E200 E200i E500");
62 MODULE_VERSION("3.6.14");
63 MODULE_LICENSE("GPL");
64
65 #include "cciss_cmd.h"
66 #include "cciss.h"
67 #include <linux/cciss_ioctl.h>
68
69 /* define the PCI info for the cards we can control */
70 static const struct pci_device_id cciss_pci_device_id[] = {
71         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISS,  0x0E11, 0x4070},
72         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4080},
73         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4082},
74         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4083},
75         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x4091},
76         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409A},
77         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409B},
78         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409C},
79         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409D},
80         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSA,     0x103C, 0x3225},
81         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3223},
82         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3234},
83         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3235},
84         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3211},
85         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3212},
86         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3213},
87         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3214},
88         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3215},
89         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3237},
90         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x323D},
91         {PCI_VENDOR_ID_HP,     PCI_ANY_ID,      PCI_ANY_ID, PCI_ANY_ID,
92                 PCI_CLASS_STORAGE_RAID << 8, 0xffff << 8, 0},
93         {0,}
94 };
95
96 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
97
98 /*  board_id = Subsystem Device ID & Vendor ID
99  *  product = Marketing Name for the board
100  *  access = Address of the struct of function pointers
101  *  nr_cmds = Number of commands supported by controller
102  */
103 static struct board_type products[] = {
104         {0x40700E11, "Smart Array 5300", &SA5_access, 512},
105         {0x40800E11, "Smart Array 5i", &SA5B_access, 512},
106         {0x40820E11, "Smart Array 532", &SA5B_access, 512},
107         {0x40830E11, "Smart Array 5312", &SA5B_access, 512},
108         {0x409A0E11, "Smart Array 641", &SA5_access, 512},
109         {0x409B0E11, "Smart Array 642", &SA5_access, 512},
110         {0x409C0E11, "Smart Array 6400", &SA5_access, 512},
111         {0x409D0E11, "Smart Array 6400 EM", &SA5_access, 512},
112         {0x40910E11, "Smart Array 6i", &SA5_access, 512},
113         {0x3225103C, "Smart Array P600", &SA5_access, 512},
114         {0x3223103C, "Smart Array P800", &SA5_access, 512},
115         {0x3234103C, "Smart Array P400", &SA5_access, 512},
116         {0x3235103C, "Smart Array P400i", &SA5_access, 512},
117         {0x3211103C, "Smart Array E200i", &SA5_access, 120},
118         {0x3212103C, "Smart Array E200", &SA5_access, 120},
119         {0x3213103C, "Smart Array E200i", &SA5_access, 120},
120         {0x3214103C, "Smart Array E200i", &SA5_access, 120},
121         {0x3215103C, "Smart Array E200i", &SA5_access, 120},
122         {0x3237103C, "Smart Array E500", &SA5_access, 512},
123         {0x323D103C, "Smart Array P700m", &SA5_access, 512},
124         {0xFFFF103C, "Unknown Smart Array", &SA5_access, 120},
125 };
126
127 /* How long to wait (in milliseconds) for board to go into simple mode */
128 #define MAX_CONFIG_WAIT 30000
129 #define MAX_IOCTL_CONFIG_WAIT 1000
130
131 /*define how many times we will try a command because of bus resets */
132 #define MAX_CMD_RETRIES 3
133
134 #define READ_AHEAD       1024
135 #define MAX_CTLR        32
136
137 /* Originally cciss driver only supports 8 major numbers */
138 #define MAX_CTLR_ORIG   8
139
140 static ctlr_info_t *hba[MAX_CTLR];
141
142 static void do_cciss_request(struct request_queue *q);
143 static irqreturn_t do_cciss_intr(int irq, void *dev_id);
144 static int cciss_open(struct inode *inode, struct file *filep);
145 static int cciss_release(struct inode *inode, struct file *filep);
146 static int cciss_ioctl(struct inode *inode, struct file *filep,
147                        unsigned int cmd, unsigned long arg);
148 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
149
150 static int cciss_revalidate(struct gendisk *disk);
151 static int rebuild_lun_table(ctlr_info_t *h, struct gendisk *del_disk);
152 static int deregister_disk(struct gendisk *disk, drive_info_struct *drv,
153                            int clear_all);
154
155 static void cciss_read_capacity(int ctlr, int logvol, int withirq,
156                         sector_t *total_size, unsigned int *block_size);
157 static void cciss_read_capacity_16(int ctlr, int logvol, int withirq,
158                         sector_t *total_size, unsigned int *block_size);
159 static void cciss_geometry_inquiry(int ctlr, int logvol,
160                         int withirq, sector_t total_size,
161                         unsigned int block_size, InquiryData_struct *inq_buff,
162                                    drive_info_struct *drv);
163 static void cciss_getgeometry(int cntl_num);
164 static void __devinit cciss_interrupt_mode(ctlr_info_t *, struct pci_dev *,
165                                            __u32);
166 static void start_io(ctlr_info_t *h);
167 static int sendcmd(__u8 cmd, int ctlr, void *buff, size_t size,
168                    unsigned int use_unit_num, unsigned int log_unit,
169                    __u8 page_code, unsigned char *scsi3addr, int cmd_type);
170 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
171                            unsigned int use_unit_num, unsigned int log_unit,
172                            __u8 page_code, int cmd_type);
173
174 static void fail_all_cmds(unsigned long ctlr);
175
176 #ifdef CONFIG_PROC_FS
177 static int cciss_proc_get_info(char *buffer, char **start, off_t offset,
178                                int length, int *eof, void *data);
179 static void cciss_procinit(int i);
180 #else
181 static void cciss_procinit(int i)
182 {
183 }
184 #endif                          /* CONFIG_PROC_FS */
185
186 #ifdef CONFIG_COMPAT
187 static long cciss_compat_ioctl(struct file *f, unsigned cmd, unsigned long arg);
188 #endif
189
190 static struct block_device_operations cciss_fops = {
191         .owner = THIS_MODULE,
192         .open = cciss_open,
193         .release = cciss_release,
194         .ioctl = cciss_ioctl,
195         .getgeo = cciss_getgeo,
196 #ifdef CONFIG_COMPAT
197         .compat_ioctl = cciss_compat_ioctl,
198 #endif
199         .revalidate_disk = cciss_revalidate,
200 };
201
202 /*
203  * Enqueuing and dequeuing functions for cmdlists.
204  */
205 static inline void addQ(CommandList_struct **Qptr, CommandList_struct *c)
206 {
207         if (*Qptr == NULL) {
208                 *Qptr = c;
209                 c->next = c->prev = c;
210         } else {
211                 c->prev = (*Qptr)->prev;
212                 c->next = (*Qptr);
213                 (*Qptr)->prev->next = c;
214                 (*Qptr)->prev = c;
215         }
216 }
217
218 static inline CommandList_struct *removeQ(CommandList_struct **Qptr,
219                                           CommandList_struct *c)
220 {
221         if (c && c->next != c) {
222                 if (*Qptr == c)
223                         *Qptr = c->next;
224                 c->prev->next = c->next;
225                 c->next->prev = c->prev;
226         } else {
227                 *Qptr = NULL;
228         }
229         return c;
230 }
231
232 #include "cciss_scsi.c"         /* For SCSI tape support */
233
234 #define RAID_UNKNOWN 6
235
236 #ifdef CONFIG_PROC_FS
237
238 /*
239  * Report information about this controller.
240  */
241 #define ENG_GIG 1000000000
242 #define ENG_GIG_FACTOR (ENG_GIG/512)
243 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
244         "UNKNOWN"
245 };
246
247 static struct proc_dir_entry *proc_cciss;
248
249 static int cciss_proc_get_info(char *buffer, char **start, off_t offset,
250                                int length, int *eof, void *data)
251 {
252         off_t pos = 0;
253         off_t len = 0;
254         int size, i, ctlr;
255         ctlr_info_t *h = (ctlr_info_t *) data;
256         drive_info_struct *drv;
257         unsigned long flags;
258         sector_t vol_sz, vol_sz_frac;
259
260         ctlr = h->ctlr;
261
262         /* prevent displaying bogus info during configuration
263          * or deconfiguration of a logical volume
264          */
265         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
266         if (h->busy_configuring) {
267                 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
268                 return -EBUSY;
269         }
270         h->busy_configuring = 1;
271         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
272
273         size = sprintf(buffer, "%s: HP %s Controller\n"
274                        "Board ID: 0x%08lx\n"
275                        "Firmware Version: %c%c%c%c\n"
276                        "IRQ: %d\n"
277                        "Logical drives: %d\n"
278                        "Max sectors: %d\n"
279                        "Current Q depth: %d\n"
280                        "Current # commands on controller: %d\n"
281                        "Max Q depth since init: %d\n"
282                        "Max # commands on controller since init: %d\n"
283                        "Max SG entries since init: %d\n\n",
284                        h->devname,
285                        h->product_name,
286                        (unsigned long)h->board_id,
287                        h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
288                        h->firm_ver[3], (unsigned int)h->intr[SIMPLE_MODE_INT],
289                        h->num_luns,
290                        h->cciss_max_sectors,
291                        h->Qdepth, h->commands_outstanding,
292                        h->maxQsinceinit, h->max_outstanding, h->maxSG);
293
294         pos += size;
295         len += size;
296         cciss_proc_tape_report(ctlr, buffer, &pos, &len);
297         for (i = 0; i <= h->highest_lun; i++) {
298
299                 drv = &h->drv[i];
300                 if (drv->heads == 0)
301                         continue;
302
303                 vol_sz = drv->nr_blocks;
304                 vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
305                 vol_sz_frac *= 100;
306                 sector_div(vol_sz_frac, ENG_GIG_FACTOR);
307
308                 if (drv->raid_level > 5)
309                         drv->raid_level = RAID_UNKNOWN;
310                 size = sprintf(buffer + len, "cciss/c%dd%d:"
311                                "\t%4u.%02uGB\tRAID %s\n",
312                                ctlr, i, (int)vol_sz, (int)vol_sz_frac,
313                                raid_label[drv->raid_level]);
314                 pos += size;
315                 len += size;
316         }
317
318         *eof = 1;
319         *start = buffer + offset;
320         len -= offset;
321         if (len > length)
322                 len = length;
323         h->busy_configuring = 0;
324         return len;
325 }
326
327 static int
328 cciss_proc_write(struct file *file, const char __user *buffer,
329                  unsigned long count, void *data)
330 {
331         unsigned char cmd[80];
332         int len;
333 #ifdef CONFIG_CISS_SCSI_TAPE
334         ctlr_info_t *h = (ctlr_info_t *) data;
335         int rc;
336 #endif
337
338         if (count > sizeof(cmd) - 1)
339                 return -EINVAL;
340         if (copy_from_user(cmd, buffer, count))
341                 return -EFAULT;
342         cmd[count] = '\0';
343         len = strlen(cmd);      // above 3 lines ensure safety
344         if (len && cmd[len - 1] == '\n')
345                 cmd[--len] = '\0';
346 #       ifdef CONFIG_CISS_SCSI_TAPE
347         if (strcmp("engage scsi", cmd) == 0) {
348                 rc = cciss_engage_scsi(h->ctlr);
349                 if (rc != 0)
350                         return -rc;
351                 return count;
352         }
353         /* might be nice to have "disengage" too, but it's not
354            safely possible. (only 1 module use count, lock issues.) */
355 #       endif
356         return -EINVAL;
357 }
358
359 /*
360  * Get us a file in /proc/cciss that says something about each controller.
361  * Create /proc/cciss if it doesn't exist yet.
362  */
363 static void __devinit cciss_procinit(int i)
364 {
365         struct proc_dir_entry *pde;
366
367         if (proc_cciss == NULL) {
368                 proc_cciss = proc_mkdir("cciss", proc_root_driver);
369                 if (!proc_cciss)
370                         return;
371         }
372
373         pde = create_proc_read_entry(hba[i]->devname,
374                                      S_IWUSR | S_IRUSR | S_IRGRP | S_IROTH,
375                                      proc_cciss, cciss_proc_get_info, hba[i]);
376         pde->write_proc = cciss_proc_write;
377 }
378 #endif                          /* CONFIG_PROC_FS */
379
380 /*
381  * For operations that cannot sleep, a command block is allocated at init,
382  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
383  * which ones are free or in use.  For operations that can wait for kmalloc
384  * to possible sleep, this routine can be called with get_from_pool set to 0.
385  * cmd_free() MUST be called with a got_from_pool set to 0 if cmd_alloc was.
386  */
387 static CommandList_struct *cmd_alloc(ctlr_info_t *h, int get_from_pool)
388 {
389         CommandList_struct *c;
390         int i;
391         u64bit temp64;
392         dma_addr_t cmd_dma_handle, err_dma_handle;
393
394         if (!get_from_pool) {
395                 c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
396                         sizeof(CommandList_struct), &cmd_dma_handle);
397                 if (c == NULL)
398                         return NULL;
399                 memset(c, 0, sizeof(CommandList_struct));
400
401                 c->cmdindex = -1;
402
403                 c->err_info = (ErrorInfo_struct *)
404                     pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
405                             &err_dma_handle);
406
407                 if (c->err_info == NULL) {
408                         pci_free_consistent(h->pdev,
409                                 sizeof(CommandList_struct), c, cmd_dma_handle);
410                         return NULL;
411                 }
412                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
413         } else {                /* get it out of the controllers pool */
414
415                 do {
416                         i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
417                         if (i == h->nr_cmds)
418                                 return NULL;
419                 } while (test_and_set_bit
420                          (i & (BITS_PER_LONG - 1),
421                           h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
422 #ifdef CCISS_DEBUG
423                 printk(KERN_DEBUG "cciss: using command buffer %d\n", i);
424 #endif
425                 c = h->cmd_pool + i;
426                 memset(c, 0, sizeof(CommandList_struct));
427                 cmd_dma_handle = h->cmd_pool_dhandle
428                     + i * sizeof(CommandList_struct);
429                 c->err_info = h->errinfo_pool + i;
430                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
431                 err_dma_handle = h->errinfo_pool_dhandle
432                     + i * sizeof(ErrorInfo_struct);
433                 h->nr_allocs++;
434
435                 c->cmdindex = i;
436         }
437
438         c->busaddr = (__u32) cmd_dma_handle;
439         temp64.val = (__u64) err_dma_handle;
440         c->ErrDesc.Addr.lower = temp64.val32.lower;
441         c->ErrDesc.Addr.upper = temp64.val32.upper;
442         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
443
444         c->ctlr = h->ctlr;
445         return c;
446 }
447
448 /*
449  * Frees a command block that was previously allocated with cmd_alloc().
450  */
451 static void cmd_free(ctlr_info_t *h, CommandList_struct *c, int got_from_pool)
452 {
453         int i;
454         u64bit temp64;
455
456         if (!got_from_pool) {
457                 temp64.val32.lower = c->ErrDesc.Addr.lower;
458                 temp64.val32.upper = c->ErrDesc.Addr.upper;
459                 pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
460                                     c->err_info, (dma_addr_t) temp64.val);
461                 pci_free_consistent(h->pdev, sizeof(CommandList_struct),
462                                     c, (dma_addr_t) c->busaddr);
463         } else {
464                 i = c - h->cmd_pool;
465                 clear_bit(i & (BITS_PER_LONG - 1),
466                           h->cmd_pool_bits + (i / BITS_PER_LONG));
467                 h->nr_frees++;
468         }
469 }
470
471 static inline ctlr_info_t *get_host(struct gendisk *disk)
472 {
473         return disk->queue->queuedata;
474 }
475
476 static inline drive_info_struct *get_drv(struct gendisk *disk)
477 {
478         return disk->private_data;
479 }
480
481 /*
482  * Open.  Make sure the device is really there.
483  */
484 static int cciss_open(struct inode *inode, struct file *filep)
485 {
486         ctlr_info_t *host = get_host(inode->i_bdev->bd_disk);
487         drive_info_struct *drv = get_drv(inode->i_bdev->bd_disk);
488
489 #ifdef CCISS_DEBUG
490         printk(KERN_DEBUG "cciss_open %s\n", inode->i_bdev->bd_disk->disk_name);
491 #endif                          /* CCISS_DEBUG */
492
493         if (host->busy_initializing || drv->busy_configuring)
494                 return -EBUSY;
495         /*
496          * Root is allowed to open raw volume zero even if it's not configured
497          * so array config can still work. Root is also allowed to open any
498          * volume that has a LUN ID, so it can issue IOCTL to reread the
499          * disk information.  I don't think I really like this
500          * but I'm already using way to many device nodes to claim another one
501          * for "raw controller".
502          */
503         if (drv->heads == 0) {
504                 if (iminor(inode) != 0) {       /* not node 0? */
505                         /* if not node 0 make sure it is a partition = 0 */
506                         if (iminor(inode) & 0x0f) {
507                                 return -ENXIO;
508                                 /* if it is, make sure we have a LUN ID */
509                         } else if (drv->LunID == 0) {
510                                 return -ENXIO;
511                         }
512                 }
513                 if (!capable(CAP_SYS_ADMIN))
514                         return -EPERM;
515         }
516         drv->usage_count++;
517         host->usage_count++;
518         return 0;
519 }
520
521 /*
522  * Close.  Sync first.
523  */
524 static int cciss_release(struct inode *inode, struct file *filep)
525 {
526         ctlr_info_t *host = get_host(inode->i_bdev->bd_disk);
527         drive_info_struct *drv = get_drv(inode->i_bdev->bd_disk);
528
529 #ifdef CCISS_DEBUG
530         printk(KERN_DEBUG "cciss_release %s\n",
531                inode->i_bdev->bd_disk->disk_name);
532 #endif                          /* CCISS_DEBUG */
533
534         drv->usage_count--;
535         host->usage_count--;
536         return 0;
537 }
538
539 #ifdef CONFIG_COMPAT
540
541 static int do_ioctl(struct file *f, unsigned cmd, unsigned long arg)
542 {
543         int ret;
544         lock_kernel();
545         ret = cciss_ioctl(f->f_path.dentry->d_inode, f, cmd, arg);
546         unlock_kernel();
547         return ret;
548 }
549
550 static int cciss_ioctl32_passthru(struct file *f, unsigned cmd,
551                                   unsigned long arg);
552 static int cciss_ioctl32_big_passthru(struct file *f, unsigned cmd,
553                                       unsigned long arg);
554
555 static long cciss_compat_ioctl(struct file *f, unsigned cmd, unsigned long arg)
556 {
557         switch (cmd) {
558         case CCISS_GETPCIINFO:
559         case CCISS_GETINTINFO:
560         case CCISS_SETINTINFO:
561         case CCISS_GETNODENAME:
562         case CCISS_SETNODENAME:
563         case CCISS_GETHEARTBEAT:
564         case CCISS_GETBUSTYPES:
565         case CCISS_GETFIRMVER:
566         case CCISS_GETDRIVVER:
567         case CCISS_REVALIDVOLS:
568         case CCISS_DEREGDISK:
569         case CCISS_REGNEWDISK:
570         case CCISS_REGNEWD:
571         case CCISS_RESCANDISK:
572         case CCISS_GETLUNINFO:
573                 return do_ioctl(f, cmd, arg);
574
575         case CCISS_PASSTHRU32:
576                 return cciss_ioctl32_passthru(f, cmd, arg);
577         case CCISS_BIG_PASSTHRU32:
578                 return cciss_ioctl32_big_passthru(f, cmd, arg);
579
580         default:
581                 return -ENOIOCTLCMD;
582         }
583 }
584
585 static int cciss_ioctl32_passthru(struct file *f, unsigned cmd,
586                                   unsigned long arg)
587 {
588         IOCTL32_Command_struct __user *arg32 =
589             (IOCTL32_Command_struct __user *) arg;
590         IOCTL_Command_struct arg64;
591         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
592         int err;
593         u32 cp;
594
595         err = 0;
596         err |=
597             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
598                            sizeof(arg64.LUN_info));
599         err |=
600             copy_from_user(&arg64.Request, &arg32->Request,
601                            sizeof(arg64.Request));
602         err |=
603             copy_from_user(&arg64.error_info, &arg32->error_info,
604                            sizeof(arg64.error_info));
605         err |= get_user(arg64.buf_size, &arg32->buf_size);
606         err |= get_user(cp, &arg32->buf);
607         arg64.buf = compat_ptr(cp);
608         err |= copy_to_user(p, &arg64, sizeof(arg64));
609
610         if (err)
611                 return -EFAULT;
612
613         err = do_ioctl(f, CCISS_PASSTHRU, (unsigned long)p);
614         if (err)
615                 return err;
616         err |=
617             copy_in_user(&arg32->error_info, &p->error_info,
618                          sizeof(arg32->error_info));
619         if (err)
620                 return -EFAULT;
621         return err;
622 }
623
624 static int cciss_ioctl32_big_passthru(struct file *file, unsigned cmd,
625                                       unsigned long arg)
626 {
627         BIG_IOCTL32_Command_struct __user *arg32 =
628             (BIG_IOCTL32_Command_struct __user *) arg;
629         BIG_IOCTL_Command_struct arg64;
630         BIG_IOCTL_Command_struct __user *p =
631             compat_alloc_user_space(sizeof(arg64));
632         int err;
633         u32 cp;
634
635         err = 0;
636         err |=
637             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
638                            sizeof(arg64.LUN_info));
639         err |=
640             copy_from_user(&arg64.Request, &arg32->Request,
641                            sizeof(arg64.Request));
642         err |=
643             copy_from_user(&arg64.error_info, &arg32->error_info,
644                            sizeof(arg64.error_info));
645         err |= get_user(arg64.buf_size, &arg32->buf_size);
646         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
647         err |= get_user(cp, &arg32->buf);
648         arg64.buf = compat_ptr(cp);
649         err |= copy_to_user(p, &arg64, sizeof(arg64));
650
651         if (err)
652                 return -EFAULT;
653
654         err = do_ioctl(file, CCISS_BIG_PASSTHRU, (unsigned long)p);
655         if (err)
656                 return err;
657         err |=
658             copy_in_user(&arg32->error_info, &p->error_info,
659                          sizeof(arg32->error_info));
660         if (err)
661                 return -EFAULT;
662         return err;
663 }
664 #endif
665
666 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
667 {
668         drive_info_struct *drv = get_drv(bdev->bd_disk);
669
670         if (!drv->cylinders)
671                 return -ENXIO;
672
673         geo->heads = drv->heads;
674         geo->sectors = drv->sectors;
675         geo->cylinders = drv->cylinders;
676         return 0;
677 }
678
679 /*
680  * ioctl
681  */
682 static int cciss_ioctl(struct inode *inode, struct file *filep,
683                        unsigned int cmd, unsigned long arg)
684 {
685         struct block_device *bdev = inode->i_bdev;
686         struct gendisk *disk = bdev->bd_disk;
687         ctlr_info_t *host = get_host(disk);
688         drive_info_struct *drv = get_drv(disk);
689         int ctlr = host->ctlr;
690         void __user *argp = (void __user *)arg;
691
692 #ifdef CCISS_DEBUG
693         printk(KERN_DEBUG "cciss_ioctl: Called with cmd=%x %lx\n", cmd, arg);
694 #endif                          /* CCISS_DEBUG */
695
696         switch (cmd) {
697         case CCISS_GETPCIINFO:
698                 {
699                         cciss_pci_info_struct pciinfo;
700
701                         if (!arg)
702                                 return -EINVAL;
703                         pciinfo.domain = pci_domain_nr(host->pdev->bus);
704                         pciinfo.bus = host->pdev->bus->number;
705                         pciinfo.dev_fn = host->pdev->devfn;
706                         pciinfo.board_id = host->board_id;
707                         if (copy_to_user
708                             (argp, &pciinfo, sizeof(cciss_pci_info_struct)))
709                                 return -EFAULT;
710                         return 0;
711                 }
712         case CCISS_GETINTINFO:
713                 {
714                         cciss_coalint_struct intinfo;
715                         if (!arg)
716                                 return -EINVAL;
717                         intinfo.delay =
718                             readl(&host->cfgtable->HostWrite.CoalIntDelay);
719                         intinfo.count =
720                             readl(&host->cfgtable->HostWrite.CoalIntCount);
721                         if (copy_to_user
722                             (argp, &intinfo, sizeof(cciss_coalint_struct)))
723                                 return -EFAULT;
724                         return 0;
725                 }
726         case CCISS_SETINTINFO:
727                 {
728                         cciss_coalint_struct intinfo;
729                         unsigned long flags;
730                         int i;
731
732                         if (!arg)
733                                 return -EINVAL;
734                         if (!capable(CAP_SYS_ADMIN))
735                                 return -EPERM;
736                         if (copy_from_user
737                             (&intinfo, argp, sizeof(cciss_coalint_struct)))
738                                 return -EFAULT;
739                         if ((intinfo.delay == 0) && (intinfo.count == 0))
740                         {
741 //                      printk("cciss_ioctl: delay and count cannot be 0\n");
742                                 return -EINVAL;
743                         }
744                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
745                         /* Update the field, and then ring the doorbell */
746                         writel(intinfo.delay,
747                                &(host->cfgtable->HostWrite.CoalIntDelay));
748                         writel(intinfo.count,
749                                &(host->cfgtable->HostWrite.CoalIntCount));
750                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
751
752                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
753                                 if (!(readl(host->vaddr + SA5_DOORBELL)
754                                       & CFGTBL_ChangeReq))
755                                         break;
756                                 /* delay and try again */
757                                 udelay(1000);
758                         }
759                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
760                         if (i >= MAX_IOCTL_CONFIG_WAIT)
761                                 return -EAGAIN;
762                         return 0;
763                 }
764         case CCISS_GETNODENAME:
765                 {
766                         NodeName_type NodeName;
767                         int i;
768
769                         if (!arg)
770                                 return -EINVAL;
771                         for (i = 0; i < 16; i++)
772                                 NodeName[i] =
773                                     readb(&host->cfgtable->ServerName[i]);
774                         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
775                                 return -EFAULT;
776                         return 0;
777                 }
778         case CCISS_SETNODENAME:
779                 {
780                         NodeName_type NodeName;
781                         unsigned long flags;
782                         int i;
783
784                         if (!arg)
785                                 return -EINVAL;
786                         if (!capable(CAP_SYS_ADMIN))
787                                 return -EPERM;
788
789                         if (copy_from_user
790                             (NodeName, argp, sizeof(NodeName_type)))
791                                 return -EFAULT;
792
793                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
794
795                         /* Update the field, and then ring the doorbell */
796                         for (i = 0; i < 16; i++)
797                                 writeb(NodeName[i],
798                                        &host->cfgtable->ServerName[i]);
799
800                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
801
802                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
803                                 if (!(readl(host->vaddr + SA5_DOORBELL)
804                                       & CFGTBL_ChangeReq))
805                                         break;
806                                 /* delay and try again */
807                                 udelay(1000);
808                         }
809                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
810                         if (i >= MAX_IOCTL_CONFIG_WAIT)
811                                 return -EAGAIN;
812                         return 0;
813                 }
814
815         case CCISS_GETHEARTBEAT:
816                 {
817                         Heartbeat_type heartbeat;
818
819                         if (!arg)
820                                 return -EINVAL;
821                         heartbeat = readl(&host->cfgtable->HeartBeat);
822                         if (copy_to_user
823                             (argp, &heartbeat, sizeof(Heartbeat_type)))
824                                 return -EFAULT;
825                         return 0;
826                 }
827         case CCISS_GETBUSTYPES:
828                 {
829                         BusTypes_type BusTypes;
830
831                         if (!arg)
832                                 return -EINVAL;
833                         BusTypes = readl(&host->cfgtable->BusTypes);
834                         if (copy_to_user
835                             (argp, &BusTypes, sizeof(BusTypes_type)))
836                                 return -EFAULT;
837                         return 0;
838                 }
839         case CCISS_GETFIRMVER:
840                 {
841                         FirmwareVer_type firmware;
842
843                         if (!arg)
844                                 return -EINVAL;
845                         memcpy(firmware, host->firm_ver, 4);
846
847                         if (copy_to_user
848                             (argp, firmware, sizeof(FirmwareVer_type)))
849                                 return -EFAULT;
850                         return 0;
851                 }
852         case CCISS_GETDRIVVER:
853                 {
854                         DriverVer_type DriverVer = DRIVER_VERSION;
855
856                         if (!arg)
857                                 return -EINVAL;
858
859                         if (copy_to_user
860                             (argp, &DriverVer, sizeof(DriverVer_type)))
861                                 return -EFAULT;
862                         return 0;
863                 }
864
865         case CCISS_REVALIDVOLS:
866                 return rebuild_lun_table(host, NULL);
867
868         case CCISS_GETLUNINFO:{
869                         LogvolInfo_struct luninfo;
870
871                         luninfo.LunID = drv->LunID;
872                         luninfo.num_opens = drv->usage_count;
873                         luninfo.num_parts = 0;
874                         if (copy_to_user(argp, &luninfo,
875                                          sizeof(LogvolInfo_struct)))
876                                 return -EFAULT;
877                         return 0;
878                 }
879         case CCISS_DEREGDISK:
880                 return rebuild_lun_table(host, disk);
881
882         case CCISS_REGNEWD:
883                 return rebuild_lun_table(host, NULL);
884
885         case CCISS_PASSTHRU:
886                 {
887                         IOCTL_Command_struct iocommand;
888                         CommandList_struct *c;
889                         char *buff = NULL;
890                         u64bit temp64;
891                         unsigned long flags;
892                         DECLARE_COMPLETION_ONSTACK(wait);
893
894                         if (!arg)
895                                 return -EINVAL;
896
897                         if (!capable(CAP_SYS_RAWIO))
898                                 return -EPERM;
899
900                         if (copy_from_user
901                             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
902                                 return -EFAULT;
903                         if ((iocommand.buf_size < 1) &&
904                             (iocommand.Request.Type.Direction != XFER_NONE)) {
905                                 return -EINVAL;
906                         }
907 #if 0                           /* 'buf_size' member is 16-bits, and always smaller than kmalloc limit */
908                         /* Check kmalloc limits */
909                         if (iocommand.buf_size > 128000)
910                                 return -EINVAL;
911 #endif
912                         if (iocommand.buf_size > 0) {
913                                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
914                                 if (buff == NULL)
915                                         return -EFAULT;
916                         }
917                         if (iocommand.Request.Type.Direction == XFER_WRITE) {
918                                 /* Copy the data into the buffer we created */
919                                 if (copy_from_user
920                                     (buff, iocommand.buf, iocommand.buf_size)) {
921                                         kfree(buff);
922                                         return -EFAULT;
923                                 }
924                         } else {
925                                 memset(buff, 0, iocommand.buf_size);
926                         }
927                         if ((c = cmd_alloc(host, 0)) == NULL) {
928                                 kfree(buff);
929                                 return -ENOMEM;
930                         }
931                         // Fill in the command type
932                         c->cmd_type = CMD_IOCTL_PEND;
933                         // Fill in Command Header
934                         c->Header.ReplyQueue = 0;       // unused in simple mode
935                         if (iocommand.buf_size > 0)     // buffer to fill
936                         {
937                                 c->Header.SGList = 1;
938                                 c->Header.SGTotal = 1;
939                         } else  // no buffers to fill
940                         {
941                                 c->Header.SGList = 0;
942                                 c->Header.SGTotal = 0;
943                         }
944                         c->Header.LUN = iocommand.LUN_info;
945                         c->Header.Tag.lower = c->busaddr;       // use the kernel address the cmd block for tag
946
947                         // Fill in Request block
948                         c->Request = iocommand.Request;
949
950                         // Fill in the scatter gather information
951                         if (iocommand.buf_size > 0) {
952                                 temp64.val = pci_map_single(host->pdev, buff,
953                                         iocommand.buf_size,
954                                         PCI_DMA_BIDIRECTIONAL);
955                                 c->SG[0].Addr.lower = temp64.val32.lower;
956                                 c->SG[0].Addr.upper = temp64.val32.upper;
957                                 c->SG[0].Len = iocommand.buf_size;
958                                 c->SG[0].Ext = 0;       // we are not chaining
959                         }
960                         c->waiting = &wait;
961
962                         /* Put the request on the tail of the request queue */
963                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
964                         addQ(&host->reqQ, c);
965                         host->Qdepth++;
966                         start_io(host);
967                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
968
969                         wait_for_completion(&wait);
970
971                         /* unlock the buffers from DMA */
972                         temp64.val32.lower = c->SG[0].Addr.lower;
973                         temp64.val32.upper = c->SG[0].Addr.upper;
974                         pci_unmap_single(host->pdev, (dma_addr_t) temp64.val,
975                                          iocommand.buf_size,
976                                          PCI_DMA_BIDIRECTIONAL);
977
978                         /* Copy the error information out */
979                         iocommand.error_info = *(c->err_info);
980                         if (copy_to_user
981                             (argp, &iocommand, sizeof(IOCTL_Command_struct))) {
982                                 kfree(buff);
983                                 cmd_free(host, c, 0);
984                                 return -EFAULT;
985                         }
986
987                         if (iocommand.Request.Type.Direction == XFER_READ) {
988                                 /* Copy the data out of the buffer we created */
989                                 if (copy_to_user
990                                     (iocommand.buf, buff, iocommand.buf_size)) {
991                                         kfree(buff);
992                                         cmd_free(host, c, 0);
993                                         return -EFAULT;
994                                 }
995                         }
996                         kfree(buff);
997                         cmd_free(host, c, 0);
998                         return 0;
999                 }
1000         case CCISS_BIG_PASSTHRU:{
1001                         BIG_IOCTL_Command_struct *ioc;
1002                         CommandList_struct *c;
1003                         unsigned char **buff = NULL;
1004                         int *buff_size = NULL;
1005                         u64bit temp64;
1006                         unsigned long flags;
1007                         BYTE sg_used = 0;
1008                         int status = 0;
1009                         int i;
1010                         DECLARE_COMPLETION_ONSTACK(wait);
1011                         __u32 left;
1012                         __u32 sz;
1013                         BYTE __user *data_ptr;
1014
1015                         if (!arg)
1016                                 return -EINVAL;
1017                         if (!capable(CAP_SYS_RAWIO))
1018                                 return -EPERM;
1019                         ioc = (BIG_IOCTL_Command_struct *)
1020                             kmalloc(sizeof(*ioc), GFP_KERNEL);
1021                         if (!ioc) {
1022                                 status = -ENOMEM;
1023                                 goto cleanup1;
1024                         }
1025                         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1026                                 status = -EFAULT;
1027                                 goto cleanup1;
1028                         }
1029                         if ((ioc->buf_size < 1) &&
1030                             (ioc->Request.Type.Direction != XFER_NONE)) {
1031                                 status = -EINVAL;
1032                                 goto cleanup1;
1033                         }
1034                         /* Check kmalloc limits  using all SGs */
1035                         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1036                                 status = -EINVAL;
1037                                 goto cleanup1;
1038                         }
1039                         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1040                                 status = -EINVAL;
1041                                 goto cleanup1;
1042                         }
1043                         buff =
1044                             kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1045                         if (!buff) {
1046                                 status = -ENOMEM;
1047                                 goto cleanup1;
1048                         }
1049                         buff_size = kmalloc(MAXSGENTRIES * sizeof(int),
1050                                                    GFP_KERNEL);
1051                         if (!buff_size) {
1052                                 status = -ENOMEM;
1053                                 goto cleanup1;
1054                         }
1055                         left = ioc->buf_size;
1056                         data_ptr = ioc->buf;
1057                         while (left) {
1058                                 sz = (left >
1059                                       ioc->malloc_size) ? ioc->
1060                                     malloc_size : left;
1061                                 buff_size[sg_used] = sz;
1062                                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1063                                 if (buff[sg_used] == NULL) {
1064                                         status = -ENOMEM;
1065                                         goto cleanup1;
1066                                 }
1067                                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1068                                         if (copy_from_user
1069                                             (buff[sg_used], data_ptr, sz)) {
1070                                                 status = -ENOMEM;
1071                                                 goto cleanup1;
1072                                         }
1073                                 } else {
1074                                         memset(buff[sg_used], 0, sz);
1075                                 }
1076                                 left -= sz;
1077                                 data_ptr += sz;
1078                                 sg_used++;
1079                         }
1080                         if ((c = cmd_alloc(host, 0)) == NULL) {
1081                                 status = -ENOMEM;
1082                                 goto cleanup1;
1083                         }
1084                         c->cmd_type = CMD_IOCTL_PEND;
1085                         c->Header.ReplyQueue = 0;
1086
1087                         if (ioc->buf_size > 0) {
1088                                 c->Header.SGList = sg_used;
1089                                 c->Header.SGTotal = sg_used;
1090                         } else {
1091                                 c->Header.SGList = 0;
1092                                 c->Header.SGTotal = 0;
1093                         }
1094                         c->Header.LUN = ioc->LUN_info;
1095                         c->Header.Tag.lower = c->busaddr;
1096
1097                         c->Request = ioc->Request;
1098                         if (ioc->buf_size > 0) {
1099                                 int i;
1100                                 for (i = 0; i < sg_used; i++) {
1101                                         temp64.val =
1102                                             pci_map_single(host->pdev, buff[i],
1103                                                     buff_size[i],
1104                                                     PCI_DMA_BIDIRECTIONAL);
1105                                         c->SG[i].Addr.lower =
1106                                             temp64.val32.lower;
1107                                         c->SG[i].Addr.upper =
1108                                             temp64.val32.upper;
1109                                         c->SG[i].Len = buff_size[i];
1110                                         c->SG[i].Ext = 0;       /* we are not chaining */
1111                                 }
1112                         }
1113                         c->waiting = &wait;
1114                         /* Put the request on the tail of the request queue */
1115                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1116                         addQ(&host->reqQ, c);
1117                         host->Qdepth++;
1118                         start_io(host);
1119                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1120                         wait_for_completion(&wait);
1121                         /* unlock the buffers from DMA */
1122                         for (i = 0; i < sg_used; i++) {
1123                                 temp64.val32.lower = c->SG[i].Addr.lower;
1124                                 temp64.val32.upper = c->SG[i].Addr.upper;
1125                                 pci_unmap_single(host->pdev,
1126                                         (dma_addr_t) temp64.val, buff_size[i],
1127                                         PCI_DMA_BIDIRECTIONAL);
1128                         }
1129                         /* Copy the error information out */
1130                         ioc->error_info = *(c->err_info);
1131                         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1132                                 cmd_free(host, c, 0);
1133                                 status = -EFAULT;
1134                                 goto cleanup1;
1135                         }
1136                         if (ioc->Request.Type.Direction == XFER_READ) {
1137                                 /* Copy the data out of the buffer we created */
1138                                 BYTE __user *ptr = ioc->buf;
1139                                 for (i = 0; i < sg_used; i++) {
1140                                         if (copy_to_user
1141                                             (ptr, buff[i], buff_size[i])) {
1142                                                 cmd_free(host, c, 0);
1143                                                 status = -EFAULT;
1144                                                 goto cleanup1;
1145                                         }
1146                                         ptr += buff_size[i];
1147                                 }
1148                         }
1149                         cmd_free(host, c, 0);
1150                         status = 0;
1151                       cleanup1:
1152                         if (buff) {
1153                                 for (i = 0; i < sg_used; i++)
1154                                         kfree(buff[i]);
1155                                 kfree(buff);
1156                         }
1157                         kfree(buff_size);
1158                         kfree(ioc);
1159                         return status;
1160                 }
1161
1162         /* scsi_cmd_ioctl handles these, below, though some are not */
1163         /* very meaningful for cciss.  SG_IO is the main one people want. */
1164
1165         case SG_GET_VERSION_NUM:
1166         case SG_SET_TIMEOUT:
1167         case SG_GET_TIMEOUT:
1168         case SG_GET_RESERVED_SIZE:
1169         case SG_SET_RESERVED_SIZE:
1170         case SG_EMULATED_HOST:
1171         case SG_IO:
1172         case SCSI_IOCTL_SEND_COMMAND:
1173                 return scsi_cmd_ioctl(filep, disk->queue, disk, cmd, argp);
1174
1175         /* scsi_cmd_ioctl would normally handle these, below, but */
1176         /* they aren't a good fit for cciss, as CD-ROMs are */
1177         /* not supported, and we don't have any bus/target/lun */
1178         /* which we present to the kernel. */
1179
1180         case CDROM_SEND_PACKET:
1181         case CDROMCLOSETRAY:
1182         case CDROMEJECT:
1183         case SCSI_IOCTL_GET_IDLUN:
1184         case SCSI_IOCTL_GET_BUS_NUMBER:
1185         default:
1186                 return -ENOTTY;
1187         }
1188 }
1189
1190 static void cciss_check_queues(ctlr_info_t *h)
1191 {
1192         int start_queue = h->next_to_run;
1193         int i;
1194
1195         /* check to see if we have maxed out the number of commands that can
1196          * be placed on the queue.  If so then exit.  We do this check here
1197          * in case the interrupt we serviced was from an ioctl and did not
1198          * free any new commands.
1199          */
1200         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1201                 return;
1202
1203         /* We have room on the queue for more commands.  Now we need to queue
1204          * them up.  We will also keep track of the next queue to run so
1205          * that every queue gets a chance to be started first.
1206          */
1207         for (i = 0; i < h->highest_lun + 1; i++) {
1208                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1209                 /* make sure the disk has been added and the drive is real
1210                  * because this can be called from the middle of init_one.
1211                  */
1212                 if (!(h->drv[curr_queue].queue) || !(h->drv[curr_queue].heads))
1213                         continue;
1214                 blk_start_queue(h->gendisk[curr_queue]->queue);
1215
1216                 /* check to see if we have maxed out the number of commands
1217                  * that can be placed on the queue.
1218                  */
1219                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1220                         if (curr_queue == start_queue) {
1221                                 h->next_to_run =
1222                                     (start_queue + 1) % (h->highest_lun + 1);
1223                                 break;
1224                         } else {
1225                                 h->next_to_run = curr_queue;
1226                                 break;
1227                         }
1228                 } else {
1229                         curr_queue = (curr_queue + 1) % (h->highest_lun + 1);
1230                 }
1231         }
1232 }
1233
1234 static void cciss_softirq_done(struct request *rq)
1235 {
1236         CommandList_struct *cmd = rq->completion_data;
1237         ctlr_info_t *h = hba[cmd->ctlr];
1238         unsigned long flags;
1239         u64bit temp64;
1240         int i, ddir;
1241
1242         if (cmd->Request.Type.Direction == XFER_READ)
1243                 ddir = PCI_DMA_FROMDEVICE;
1244         else
1245                 ddir = PCI_DMA_TODEVICE;
1246
1247         /* command did not need to be retried */
1248         /* unmap the DMA mapping for all the scatter gather elements */
1249         for (i = 0; i < cmd->Header.SGList; i++) {
1250                 temp64.val32.lower = cmd->SG[i].Addr.lower;
1251                 temp64.val32.upper = cmd->SG[i].Addr.upper;
1252                 pci_unmap_page(h->pdev, temp64.val, cmd->SG[i].Len, ddir);
1253         }
1254
1255 #ifdef CCISS_DEBUG
1256         printk("Done with %p\n", rq);
1257 #endif                          /* CCISS_DEBUG */
1258
1259         if (blk_end_request(rq, (rq->errors == 0) ? 0 : -EIO, blk_rq_bytes(rq)))
1260                 BUG();
1261
1262         spin_lock_irqsave(&h->lock, flags);
1263         cmd_free(h, cmd, 1);
1264         cciss_check_queues(h);
1265         spin_unlock_irqrestore(&h->lock, flags);
1266 }
1267
1268 /* This function will check the usage_count of the drive to be updated/added.
1269  * If the usage_count is zero then the drive information will be updated and
1270  * the disk will be re-registered with the kernel.  If not then it will be
1271  * left alone for the next reboot.  The exception to this is disk 0 which
1272  * will always be left registered with the kernel since it is also the
1273  * controller node.  Any changes to disk 0 will show up on the next
1274  * reboot.
1275  */
1276 static void cciss_update_drive_info(int ctlr, int drv_index)
1277 {
1278         ctlr_info_t *h = hba[ctlr];
1279         struct gendisk *disk;
1280         InquiryData_struct *inq_buff = NULL;
1281         unsigned int block_size;
1282         sector_t total_size;
1283         unsigned long flags = 0;
1284         int ret = 0;
1285
1286         /* if the disk already exists then deregister it before proceeding */
1287         if (h->drv[drv_index].raid_level != -1) {
1288                 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1289                 h->drv[drv_index].busy_configuring = 1;
1290                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1291                 ret = deregister_disk(h->gendisk[drv_index],
1292                                       &h->drv[drv_index], 0);
1293                 h->drv[drv_index].busy_configuring = 0;
1294         }
1295
1296         /* If the disk is in use return */
1297         if (ret)
1298                 return;
1299
1300         /* Get information about the disk and modify the driver structure */
1301         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1302         if (inq_buff == NULL)
1303                 goto mem_msg;
1304
1305         /* testing to see if 16-byte CDBs are already being used */
1306         if (h->cciss_read == CCISS_READ_16) {
1307                 cciss_read_capacity_16(h->ctlr, drv_index, 1,
1308                         &total_size, &block_size);
1309                 goto geo_inq;
1310         }
1311
1312         cciss_read_capacity(ctlr, drv_index, 1,
1313                             &total_size, &block_size);
1314
1315         /* if read_capacity returns all F's this volume is >2TB in size */
1316         /* so we switch to 16-byte CDB's for all read/write ops */
1317         if (total_size == 0xFFFFFFFFULL) {
1318                 cciss_read_capacity_16(ctlr, drv_index, 1,
1319                 &total_size, &block_size);
1320                 h->cciss_read = CCISS_READ_16;
1321                 h->cciss_write = CCISS_WRITE_16;
1322         } else {
1323                 h->cciss_read = CCISS_READ_10;
1324                 h->cciss_write = CCISS_WRITE_10;
1325         }
1326 geo_inq:
1327         cciss_geometry_inquiry(ctlr, drv_index, 1, total_size, block_size,
1328                                inq_buff, &h->drv[drv_index]);
1329
1330         ++h->num_luns;
1331         disk = h->gendisk[drv_index];
1332         set_capacity(disk, h->drv[drv_index].nr_blocks);
1333
1334         /* if it's the controller it's already added */
1335         if (drv_index) {
1336                 disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1337                 sprintf(disk->disk_name, "cciss/c%dd%d", ctlr, drv_index);
1338                 disk->major = h->major;
1339                 disk->first_minor = drv_index << NWD_SHIFT;
1340                 disk->fops = &cciss_fops;
1341                 disk->private_data = &h->drv[drv_index];
1342
1343                 /* Set up queue information */
1344                 disk->queue->backing_dev_info.ra_pages = READ_AHEAD;
1345                 blk_queue_bounce_limit(disk->queue, hba[ctlr]->pdev->dma_mask);
1346
1347                 /* This is a hardware imposed limit. */
1348                 blk_queue_max_hw_segments(disk->queue, MAXSGENTRIES);
1349
1350                 /* This is a limit in the driver and could be eliminated. */
1351                 blk_queue_max_phys_segments(disk->queue, MAXSGENTRIES);
1352
1353                 blk_queue_max_sectors(disk->queue, h->cciss_max_sectors);
1354
1355                 blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1356
1357                 disk->queue->queuedata = hba[ctlr];
1358
1359                 blk_queue_hardsect_size(disk->queue,
1360                                         hba[ctlr]->drv[drv_index].block_size);
1361
1362                 h->drv[drv_index].queue = disk->queue;
1363                 add_disk(disk);
1364         }
1365
1366       freeret:
1367         kfree(inq_buff);
1368         return;
1369       mem_msg:
1370         printk(KERN_ERR "cciss: out of memory\n");
1371         goto freeret;
1372 }
1373
1374 /* This function will find the first index of the controllers drive array
1375  * that has a -1 for the raid_level and will return that index.  This is
1376  * where new drives will be added.  If the index to be returned is greater
1377  * than the highest_lun index for the controller then highest_lun is set
1378  * to this new index.  If there are no available indexes then -1 is returned.
1379  */
1380 static int cciss_find_free_drive_index(int ctlr)
1381 {
1382         int i;
1383
1384         for (i = 0; i < CISS_MAX_LUN; i++) {
1385                 if (hba[ctlr]->drv[i].raid_level == -1) {
1386                         if (i > hba[ctlr]->highest_lun)
1387                                 hba[ctlr]->highest_lun = i;
1388                         return i;
1389                 }
1390         }
1391         return -1;
1392 }
1393
1394 /* This function will add and remove logical drives from the Logical
1395  * drive array of the controller and maintain persistency of ordering
1396  * so that mount points are preserved until the next reboot.  This allows
1397  * for the removal of logical drives in the middle of the drive array
1398  * without a re-ordering of those drives.
1399  * INPUT
1400  * h            = The controller to perform the operations on
1401  * del_disk     = The disk to remove if specified.  If the value given
1402  *                is NULL then no disk is removed.
1403  */
1404 static int rebuild_lun_table(ctlr_info_t *h, struct gendisk *del_disk)
1405 {
1406         int ctlr = h->ctlr;
1407         int num_luns;
1408         ReportLunData_struct *ld_buff = NULL;
1409         drive_info_struct *drv = NULL;
1410         int return_code;
1411         int listlength = 0;
1412         int i;
1413         int drv_found;
1414         int drv_index = 0;
1415         __u32 lunid = 0;
1416         unsigned long flags;
1417
1418         /* Set busy_configuring flag for this operation */
1419         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1420         if (h->busy_configuring) {
1421                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1422                 return -EBUSY;
1423         }
1424         h->busy_configuring = 1;
1425
1426         /* if del_disk is NULL then we are being called to add a new disk
1427          * and update the logical drive table.  If it is not NULL then
1428          * we will check if the disk is in use or not.
1429          */
1430         if (del_disk != NULL) {
1431                 drv = get_drv(del_disk);
1432                 drv->busy_configuring = 1;
1433                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1434                 return_code = deregister_disk(del_disk, drv, 1);
1435                 drv->busy_configuring = 0;
1436                 h->busy_configuring = 0;
1437                 return return_code;
1438         } else {
1439                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1440                 if (!capable(CAP_SYS_RAWIO))
1441                         return -EPERM;
1442
1443                 ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
1444                 if (ld_buff == NULL)
1445                         goto mem_msg;
1446
1447                 return_code = sendcmd_withirq(CISS_REPORT_LOG, ctlr, ld_buff,
1448                                               sizeof(ReportLunData_struct), 0,
1449                                               0, 0, TYPE_CMD);
1450
1451                 if (return_code == IO_OK) {
1452                         listlength =
1453                                 be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
1454                 } else {        /* reading number of logical volumes failed */
1455                         printk(KERN_WARNING "cciss: report logical volume"
1456                                " command failed\n");
1457                         listlength = 0;
1458                         goto freeret;
1459                 }
1460
1461                 num_luns = listlength / 8;      /* 8 bytes per entry */
1462                 if (num_luns > CISS_MAX_LUN) {
1463                         num_luns = CISS_MAX_LUN;
1464                         printk(KERN_WARNING "cciss: more luns configured"
1465                                " on controller than can be handled by"
1466                                " this driver.\n");
1467                 }
1468
1469                 /* Compare controller drive array to drivers drive array.
1470                  * Check for updates in the drive information and any new drives
1471                  * on the controller.
1472                  */
1473                 for (i = 0; i < num_luns; i++) {
1474                         int j;
1475
1476                         drv_found = 0;
1477
1478                         lunid = (0xff &
1479                                  (unsigned int)(ld_buff->LUN[i][3])) << 24;
1480                         lunid |= (0xff &
1481                                   (unsigned int)(ld_buff->LUN[i][2])) << 16;
1482                         lunid |= (0xff &
1483                                   (unsigned int)(ld_buff->LUN[i][1])) << 8;
1484                         lunid |= 0xff & (unsigned int)(ld_buff->LUN[i][0]);
1485
1486                         /* Find if the LUN is already in the drive array
1487                          * of the controller.  If so then update its info
1488                          * if not is use.  If it does not exist then find
1489                          * the first free index and add it.
1490                          */
1491                         for (j = 0; j <= h->highest_lun; j++) {
1492                                 if (h->drv[j].LunID == lunid) {
1493                                         drv_index = j;
1494                                         drv_found = 1;
1495                                 }
1496                         }
1497
1498                         /* check if the drive was found already in the array */
1499                         if (!drv_found) {
1500                                 drv_index = cciss_find_free_drive_index(ctlr);
1501                                 if (drv_index == -1)
1502                                         goto freeret;
1503
1504                                 /*Check if the gendisk needs to be allocated */
1505                                 if (!h->gendisk[drv_index]){
1506                                         h->gendisk[drv_index] = alloc_disk(1 << NWD_SHIFT);
1507                                         if (!h->gendisk[drv_index]){
1508                                                 printk(KERN_ERR "cciss: could not allocate new disk %d\n", drv_index);
1509                                                 goto mem_msg;
1510                                         }
1511                                 }
1512                         }
1513                         h->drv[drv_index].LunID = lunid;
1514                         cciss_update_drive_info(ctlr, drv_index);
1515                 }               /* end for */
1516         }                       /* end else */
1517
1518       freeret:
1519         kfree(ld_buff);
1520         h->busy_configuring = 0;
1521         /* We return -1 here to tell the ACU that we have registered/updated
1522          * all of the drives that we can and to keep it from calling us
1523          * additional times.
1524          */
1525         return -1;
1526       mem_msg:
1527         printk(KERN_ERR "cciss: out of memory\n");
1528         goto freeret;
1529 }
1530
1531 /* This function will deregister the disk and it's queue from the
1532  * kernel.  It must be called with the controller lock held and the
1533  * drv structures busy_configuring flag set.  It's parameters are:
1534  *
1535  * disk = This is the disk to be deregistered
1536  * drv  = This is the drive_info_struct associated with the disk to be
1537  *        deregistered.  It contains information about the disk used
1538  *        by the driver.
1539  * clear_all = This flag determines whether or not the disk information
1540  *             is going to be completely cleared out and the highest_lun
1541  *             reset.  Sometimes we want to clear out information about
1542  *             the disk in preparation for re-adding it.  In this case
1543  *             the highest_lun should be left unchanged and the LunID
1544  *             should not be cleared.
1545 */
1546 static int deregister_disk(struct gendisk *disk, drive_info_struct *drv,
1547                            int clear_all)
1548 {
1549         int i;
1550         ctlr_info_t *h = get_host(disk);
1551
1552         if (!capable(CAP_SYS_RAWIO))
1553                 return -EPERM;
1554
1555         /* make sure logical volume is NOT is use */
1556         if (clear_all || (h->gendisk[0] == disk)) {
1557                 if (drv->usage_count > 1)
1558                         return -EBUSY;
1559         } else if (drv->usage_count > 0)
1560                 return -EBUSY;
1561
1562         /* invalidate the devices and deregister the disk.  If it is disk
1563          * zero do not deregister it but just zero out it's values.  This
1564          * allows us to delete disk zero but keep the controller registered.
1565          */
1566         if (h->gendisk[0] != disk) {
1567                 struct request_queue *q = disk->queue;
1568                 if (disk->flags & GENHD_FL_UP)
1569                         del_gendisk(disk);
1570                 if (q) {
1571                         blk_cleanup_queue(q);
1572                         /* Set drv->queue to NULL so that we do not try
1573                          * to call blk_start_queue on this queue in the
1574                          * interrupt handler
1575                          */
1576                         drv->queue = NULL;
1577                 }
1578                 /* If clear_all is set then we are deleting the logical
1579                  * drive, not just refreshing its info.  For drives
1580                  * other than disk 0 we will call put_disk.  We do not
1581                  * do this for disk 0 as we need it to be able to
1582                  * configure the controller.
1583                 */
1584                 if (clear_all){
1585                         /* This isn't pretty, but we need to find the
1586                          * disk in our array and NULL our the pointer.
1587                          * This is so that we will call alloc_disk if
1588                          * this index is used again later.
1589                         */
1590                         for (i=0; i < CISS_MAX_LUN; i++){
1591                                 if(h->gendisk[i] == disk){
1592                                         h->gendisk[i] = NULL;
1593                                         break;
1594                                 }
1595                         }
1596                         put_disk(disk);
1597                 }
1598         } else {
1599                 set_capacity(disk, 0);
1600         }
1601
1602         --h->num_luns;
1603         /* zero out the disk size info */
1604         drv->nr_blocks = 0;
1605         drv->block_size = 0;
1606         drv->heads = 0;
1607         drv->sectors = 0;
1608         drv->cylinders = 0;
1609         drv->raid_level = -1;   /* This can be used as a flag variable to
1610                                  * indicate that this element of the drive
1611                                  * array is free.
1612                                  */
1613
1614         if (clear_all) {
1615                 /* check to see if it was the last disk */
1616                 if (drv == h->drv + h->highest_lun) {
1617                         /* if so, find the new hightest lun */
1618                         int i, newhighest = -1;
1619                         for (i = 0; i < h->highest_lun; i++) {
1620                                 /* if the disk has size > 0, it is available */
1621                                 if (h->drv[i].heads)
1622                                         newhighest = i;
1623                         }
1624                         h->highest_lun = newhighest;
1625                 }
1626
1627                 drv->LunID = 0;
1628         }
1629         return 0;
1630 }
1631
1632 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,
1633                                                                                                                            1: address logical volume log_unit,
1634                                                                                                                            2: periph device address is scsi3addr */
1635                     unsigned int log_unit, __u8 page_code,
1636                     unsigned char *scsi3addr, int cmd_type)
1637 {
1638         ctlr_info_t *h = hba[ctlr];
1639         u64bit buff_dma_handle;
1640         int status = IO_OK;
1641
1642         c->cmd_type = CMD_IOCTL_PEND;
1643         c->Header.ReplyQueue = 0;
1644         if (buff != NULL) {
1645                 c->Header.SGList = 1;
1646                 c->Header.SGTotal = 1;
1647         } else {
1648                 c->Header.SGList = 0;
1649                 c->Header.SGTotal = 0;
1650         }
1651         c->Header.Tag.lower = c->busaddr;
1652
1653         c->Request.Type.Type = cmd_type;
1654         if (cmd_type == TYPE_CMD) {
1655                 switch (cmd) {
1656                 case CISS_INQUIRY:
1657                         /* If the logical unit number is 0 then, this is going
1658                            to controller so It's a physical command
1659                            mode = 0 target = 0.  So we have nothing to write.
1660                            otherwise, if use_unit_num == 1,
1661                            mode = 1(volume set addressing) target = LUNID
1662                            otherwise, if use_unit_num == 2,
1663                            mode = 0(periph dev addr) target = scsi3addr */
1664                         if (use_unit_num == 1) {
1665                                 c->Header.LUN.LogDev.VolId =
1666                                     h->drv[log_unit].LunID;
1667                                 c->Header.LUN.LogDev.Mode = 1;
1668                         } else if (use_unit_num == 2) {
1669                                 memcpy(c->Header.LUN.LunAddrBytes, scsi3addr,
1670                                        8);
1671                                 c->Header.LUN.LogDev.Mode = 0;
1672                         }
1673                         /* are we trying to read a vital product page */
1674                         if (page_code != 0) {
1675                                 c->Request.CDB[1] = 0x01;
1676                                 c->Request.CDB[2] = page_code;
1677                         }
1678                         c->Request.CDBLen = 6;
1679                         c->Request.Type.Attribute = ATTR_SIMPLE;
1680                         c->Request.Type.Direction = XFER_READ;
1681                         c->Request.Timeout = 0;
1682                         c->Request.CDB[0] = CISS_INQUIRY;
1683                         c->Request.CDB[4] = size & 0xFF;
1684                         break;
1685                 case CISS_REPORT_LOG:
1686                 case CISS_REPORT_PHYS:
1687                         /* Talking to controller so It's a physical command
1688                            mode = 00 target = 0.  Nothing to write.
1689                          */
1690                         c->Request.CDBLen = 12;
1691                         c->Request.Type.Attribute = ATTR_SIMPLE;
1692                         c->Request.Type.Direction = XFER_READ;
1693                         c->Request.Timeout = 0;
1694                         c->Request.CDB[0] = cmd;
1695                         c->Request.CDB[6] = (size >> 24) & 0xFF;        //MSB
1696                         c->Request.CDB[7] = (size >> 16) & 0xFF;
1697                         c->Request.CDB[8] = (size >> 8) & 0xFF;
1698                         c->Request.CDB[9] = size & 0xFF;
1699                         break;
1700
1701                 case CCISS_READ_CAPACITY:
1702                         c->Header.LUN.LogDev.VolId = h->drv[log_unit].LunID;
1703                         c->Header.LUN.LogDev.Mode = 1;
1704                         c->Request.CDBLen = 10;
1705                         c->Request.Type.Attribute = ATTR_SIMPLE;
1706                         c->Request.Type.Direction = XFER_READ;
1707                         c->Request.Timeout = 0;
1708                         c->Request.CDB[0] = cmd;
1709                         break;
1710                 case CCISS_READ_CAPACITY_16:
1711                         c->Header.LUN.LogDev.VolId = h->drv[log_unit].LunID;
1712                         c->Header.LUN.LogDev.Mode = 1;
1713                         c->Request.CDBLen = 16;
1714                         c->Request.Type.Attribute = ATTR_SIMPLE;
1715                         c->Request.Type.Direction = XFER_READ;
1716                         c->Request.Timeout = 0;
1717                         c->Request.CDB[0] = cmd;
1718                         c->Request.CDB[1] = 0x10;
1719                         c->Request.CDB[10] = (size >> 24) & 0xFF;
1720                         c->Request.CDB[11] = (size >> 16) & 0xFF;
1721                         c->Request.CDB[12] = (size >> 8) & 0xFF;
1722                         c->Request.CDB[13] = size & 0xFF;
1723                         c->Request.Timeout = 0;
1724                         c->Request.CDB[0] = cmd;
1725                         break;
1726                 case CCISS_CACHE_FLUSH:
1727                         c->Request.CDBLen = 12;
1728                         c->Request.Type.Attribute = ATTR_SIMPLE;
1729                         c->Request.Type.Direction = XFER_WRITE;
1730                         c->Request.Timeout = 0;
1731                         c->Request.CDB[0] = BMIC_WRITE;
1732                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
1733                         break;
1734                 default:
1735                         printk(KERN_WARNING
1736                                "cciss%d:  Unknown Command 0x%c\n", ctlr, cmd);
1737                         return IO_ERROR;
1738                 }
1739         } else if (cmd_type == TYPE_MSG) {
1740                 switch (cmd) {
1741                 case 0: /* ABORT message */
1742                         c->Request.CDBLen = 12;
1743                         c->Request.Type.Attribute = ATTR_SIMPLE;
1744                         c->Request.Type.Direction = XFER_WRITE;
1745                         c->Request.Timeout = 0;
1746                         c->Request.CDB[0] = cmd;        /* abort */
1747                         c->Request.CDB[1] = 0;  /* abort a command */
1748                         /* buff contains the tag of the command to abort */
1749                         memcpy(&c->Request.CDB[4], buff, 8);
1750                         break;
1751                 case 1: /* RESET message */
1752                         c->Request.CDBLen = 12;
1753                         c->Request.Type.Attribute = ATTR_SIMPLE;
1754                         c->Request.Type.Direction = XFER_WRITE;
1755                         c->Request.Timeout = 0;
1756                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
1757                         c->Request.CDB[0] = cmd;        /* reset */
1758                         c->Request.CDB[1] = 0x04;       /* reset a LUN */
1759                         break;
1760                 case 3: /* No-Op message */
1761                         c->Request.CDBLen = 1;
1762                         c->Request.Type.Attribute = ATTR_SIMPLE;
1763                         c->Request.Type.Direction = XFER_WRITE;
1764                         c->Request.Timeout = 0;
1765                         c->Request.CDB[0] = cmd;
1766                         break;
1767                 default:
1768                         printk(KERN_WARNING
1769                                "cciss%d: unknown message type %d\n", ctlr, cmd);
1770                         return IO_ERROR;
1771                 }
1772         } else {
1773                 printk(KERN_WARNING
1774                        "cciss%d: unknown command type %d\n", ctlr, cmd_type);
1775                 return IO_ERROR;
1776         }
1777         /* Fill in the scatter gather information */
1778         if (size > 0) {
1779                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
1780                                                              buff, size,
1781                                                              PCI_DMA_BIDIRECTIONAL);
1782                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
1783                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
1784                 c->SG[0].Len = size;
1785                 c->SG[0].Ext = 0;       /* we are not chaining */
1786         }
1787         return status;
1788 }
1789
1790 static int sendcmd_withirq(__u8 cmd,
1791                            int ctlr,
1792                            void *buff,
1793                            size_t size,
1794                            unsigned int use_unit_num,
1795                            unsigned int log_unit, __u8 page_code, int cmd_type)
1796 {
1797         ctlr_info_t *h = hba[ctlr];
1798         CommandList_struct *c;
1799         u64bit buff_dma_handle;
1800         unsigned long flags;
1801         int return_status;
1802         DECLARE_COMPLETION_ONSTACK(wait);
1803
1804         if ((c = cmd_alloc(h, 0)) == NULL)
1805                 return -ENOMEM;
1806         return_status = fill_cmd(c, cmd, ctlr, buff, size, use_unit_num,
1807                                  log_unit, page_code, NULL, cmd_type);
1808         if (return_status != IO_OK) {
1809                 cmd_free(h, c, 0);
1810                 return return_status;
1811         }
1812       resend_cmd2:
1813         c->waiting = &wait;
1814
1815         /* Put the request on the tail of the queue and send it */
1816         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1817         addQ(&h->reqQ, c);
1818         h->Qdepth++;
1819         start_io(h);
1820         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1821
1822         wait_for_completion(&wait);
1823
1824         if (c->err_info->CommandStatus != 0) {  /* an error has occurred */
1825                 switch (c->err_info->CommandStatus) {
1826                 case CMD_TARGET_STATUS:
1827                         printk(KERN_WARNING "cciss: cmd %p has "
1828                                " completed with errors\n", c);
1829                         if (c->err_info->ScsiStatus) {
1830                                 printk(KERN_WARNING "cciss: cmd %p "
1831                                        "has SCSI Status = %x\n",
1832                                        c, c->err_info->ScsiStatus);
1833                         }
1834
1835                         break;
1836                 case CMD_DATA_UNDERRUN:
1837                 case CMD_DATA_OVERRUN:
1838                         /* expected for inquire and report lun commands */
1839                         break;
1840                 case CMD_INVALID:
1841                         printk(KERN_WARNING "cciss: Cmd %p is "
1842                                "reported invalid\n", c);
1843                         return_status = IO_ERROR;
1844                         break;
1845                 case CMD_PROTOCOL_ERR:
1846                         printk(KERN_WARNING "cciss: cmd %p has "
1847                                "protocol error \n", c);
1848                         return_status = IO_ERROR;
1849                         break;
1850                 case CMD_HARDWARE_ERR:
1851                         printk(KERN_WARNING "cciss: cmd %p had "
1852                                " hardware error\n", c);
1853                         return_status = IO_ERROR;
1854                         break;
1855                 case CMD_CONNECTION_LOST:
1856                         printk(KERN_WARNING "cciss: cmd %p had "
1857                                "connection lost\n", c);
1858                         return_status = IO_ERROR;
1859                         break;
1860                 case CMD_ABORTED:
1861                         printk(KERN_WARNING "cciss: cmd %p was "
1862                                "aborted\n", c);
1863                         return_status = IO_ERROR;
1864                         break;
1865                 case CMD_ABORT_FAILED:
1866                         printk(KERN_WARNING "cciss: cmd %p reports "
1867                                "abort failed\n", c);
1868                         return_status = IO_ERROR;
1869                         break;
1870                 case CMD_UNSOLICITED_ABORT:
1871                         printk(KERN_WARNING
1872                                "cciss%d: unsolicited abort %p\n", ctlr, c);
1873                         if (c->retry_count < MAX_CMD_RETRIES) {
1874                                 printk(KERN_WARNING
1875                                        "cciss%d: retrying %p\n", ctlr, c);
1876                                 c->retry_count++;
1877                                 /* erase the old error information */
1878                                 memset(c->err_info, 0,
1879                                        sizeof(ErrorInfo_struct));
1880                                 return_status = IO_OK;
1881                                 INIT_COMPLETION(wait);
1882                                 goto resend_cmd2;
1883                         }
1884                         return_status = IO_ERROR;
1885                         break;
1886                 default:
1887                         printk(KERN_WARNING "cciss: cmd %p returned "
1888                                "unknown status %x\n", c,
1889                                c->err_info->CommandStatus);
1890                         return_status = IO_ERROR;
1891                 }
1892         }
1893         /* unlock the buffers from DMA */
1894         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
1895         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
1896         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
1897                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
1898         cmd_free(h, c, 0);
1899         return return_status;
1900 }
1901
1902 static void cciss_geometry_inquiry(int ctlr, int logvol,
1903                                    int withirq, sector_t total_size,
1904                                    unsigned int block_size,
1905                                    InquiryData_struct *inq_buff,
1906                                    drive_info_struct *drv)
1907 {
1908         int return_code;
1909         unsigned long t;
1910
1911         memset(inq_buff, 0, sizeof(InquiryData_struct));
1912         if (withirq)
1913                 return_code = sendcmd_withirq(CISS_INQUIRY, ctlr,
1914                                               inq_buff, sizeof(*inq_buff), 1,
1915                                               logvol, 0xC1, TYPE_CMD);
1916         else
1917                 return_code = sendcmd(CISS_INQUIRY, ctlr, inq_buff,
1918                                       sizeof(*inq_buff), 1, logvol, 0xC1, NULL,
1919                                       TYPE_CMD);
1920         if (return_code == IO_OK) {
1921                 if (inq_buff->data_byte[8] == 0xFF) {
1922                         printk(KERN_WARNING
1923                                "cciss: reading geometry failed, volume "
1924                                "does not support reading geometry\n");
1925                         drv->heads = 255;
1926                         drv->sectors = 32;      // Sectors per track
1927                         drv->cylinders = total_size + 1;
1928                         drv->raid_level = RAID_UNKNOWN;
1929                 } else {
1930                         drv->heads = inq_buff->data_byte[6];
1931                         drv->sectors = inq_buff->data_byte[7];
1932                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
1933                         drv->cylinders += inq_buff->data_byte[5];
1934                         drv->raid_level = inq_buff->data_byte[8];
1935                 }
1936                 drv->block_size = block_size;
1937                 drv->nr_blocks = total_size + 1;
1938                 t = drv->heads * drv->sectors;
1939                 if (t > 1) {
1940                         sector_t real_size = total_size + 1;
1941                         unsigned long rem = sector_div(real_size, t);
1942                         if (rem)
1943                                 real_size++;
1944                         drv->cylinders = real_size;
1945                 }
1946         } else {                /* Get geometry failed */
1947                 printk(KERN_WARNING "cciss: reading geometry failed\n");
1948         }
1949         printk(KERN_INFO "      heads=%d, sectors=%d, cylinders=%d\n\n",
1950                drv->heads, drv->sectors, drv->cylinders);
1951 }
1952
1953 static void
1954 cciss_read_capacity(int ctlr, int logvol, int withirq, sector_t *total_size,
1955                     unsigned int *block_size)
1956 {
1957         ReadCapdata_struct *buf;
1958         int return_code;
1959
1960         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
1961         if (!buf) {
1962                 printk(KERN_WARNING "cciss: out of memory\n");
1963                 return;
1964         }
1965
1966         if (withirq)
1967                 return_code = sendcmd_withirq(CCISS_READ_CAPACITY,
1968                                 ctlr, buf, sizeof(ReadCapdata_struct),
1969                                         1, logvol, 0, TYPE_CMD);
1970         else
1971                 return_code = sendcmd(CCISS_READ_CAPACITY,
1972                                 ctlr, buf, sizeof(ReadCapdata_struct),
1973                                         1, logvol, 0, NULL, TYPE_CMD);
1974         if (return_code == IO_OK) {
1975                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
1976                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
1977         } else {                /* read capacity command failed */
1978                 printk(KERN_WARNING "cciss: read capacity failed\n");
1979                 *total_size = 0;
1980                 *block_size = BLOCK_SIZE;
1981         }
1982         if (*total_size != 0)
1983                 printk(KERN_INFO "      blocks= %llu block_size= %d\n",
1984                 (unsigned long long)*total_size+1, *block_size);
1985         kfree(buf);
1986 }
1987
1988 static void
1989 cciss_read_capacity_16(int ctlr, int logvol, int withirq, sector_t *total_size,                                 unsigned int *block_size)
1990 {
1991         ReadCapdata_struct_16 *buf;
1992         int return_code;
1993
1994         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
1995         if (!buf) {
1996                 printk(KERN_WARNING "cciss: out of memory\n");
1997                 return;
1998         }
1999
2000         if (withirq) {
2001                 return_code = sendcmd_withirq(CCISS_READ_CAPACITY_16,
2002                         ctlr, buf, sizeof(ReadCapdata_struct_16),
2003                                 1, logvol, 0, TYPE_CMD);
2004         }
2005         else {
2006                 return_code = sendcmd(CCISS_READ_CAPACITY_16,
2007                         ctlr, buf, sizeof(ReadCapdata_struct_16),
2008                                 1, logvol, 0, NULL, TYPE_CMD);
2009         }
2010         if (return_code == IO_OK) {
2011                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2012                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2013         } else {                /* read capacity command failed */
2014                 printk(KERN_WARNING "cciss: read capacity failed\n");
2015                 *total_size = 0;
2016                 *block_size = BLOCK_SIZE;
2017         }
2018         printk(KERN_INFO "      blocks= %llu block_size= %d\n",
2019                (unsigned long long)*total_size+1, *block_size);
2020         kfree(buf);
2021 }
2022
2023 static int cciss_revalidate(struct gendisk *disk)
2024 {
2025         ctlr_info_t *h = get_host(disk);
2026         drive_info_struct *drv = get_drv(disk);
2027         int logvol;
2028         int FOUND = 0;
2029         unsigned int block_size;
2030         sector_t total_size;
2031         InquiryData_struct *inq_buff = NULL;
2032
2033         for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
2034                 if (h->drv[logvol].LunID == drv->LunID) {
2035                         FOUND = 1;
2036                         break;
2037                 }
2038         }
2039
2040         if (!FOUND)
2041                 return 1;
2042
2043         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2044         if (inq_buff == NULL) {
2045                 printk(KERN_WARNING "cciss: out of memory\n");
2046                 return 1;
2047         }
2048         if (h->cciss_read == CCISS_READ_10) {
2049                 cciss_read_capacity(h->ctlr, logvol, 1,
2050                                         &total_size, &block_size);
2051         } else {
2052                 cciss_read_capacity_16(h->ctlr, logvol, 1,
2053                                         &total_size, &block_size);
2054         }
2055         cciss_geometry_inquiry(h->ctlr, logvol, 1, total_size, block_size,
2056                                inq_buff, drv);
2057
2058         blk_queue_hardsect_size(drv->queue, drv->block_size);
2059         set_capacity(disk, drv->nr_blocks);
2060
2061         kfree(inq_buff);
2062         return 0;
2063 }
2064
2065 /*
2066  *   Wait polling for a command to complete.
2067  *   The memory mapped FIFO is polled for the completion.
2068  *   Used only at init time, interrupts from the HBA are disabled.
2069  */
2070 static unsigned long pollcomplete(int ctlr)
2071 {
2072         unsigned long done;
2073         int i;
2074
2075         /* Wait (up to 20 seconds) for a command to complete */
2076
2077         for (i = 20 * HZ; i > 0; i--) {
2078                 done = hba[ctlr]->access.command_completed(hba[ctlr]);
2079                 if (done == FIFO_EMPTY)
2080                         schedule_timeout_uninterruptible(1);
2081                 else
2082                         return done;
2083         }
2084         /* Invalid address to tell caller we ran out of time */
2085         return 1;
2086 }
2087
2088 static int add_sendcmd_reject(__u8 cmd, int ctlr, unsigned long complete)
2089 {
2090         /* We get in here if sendcmd() is polling for completions
2091            and gets some command back that it wasn't expecting --
2092            something other than that which it just sent down.
2093            Ordinarily, that shouldn't happen, but it can happen when
2094            the scsi tape stuff gets into error handling mode, and
2095            starts using sendcmd() to try to abort commands and
2096            reset tape drives.  In that case, sendcmd may pick up
2097            completions of commands that were sent to logical drives
2098            through the block i/o system, or cciss ioctls completing, etc.
2099            In that case, we need to save those completions for later
2100            processing by the interrupt handler.
2101          */
2102
2103 #ifdef CONFIG_CISS_SCSI_TAPE
2104         struct sendcmd_reject_list *srl = &hba[ctlr]->scsi_rejects;
2105
2106         /* If it's not the scsi tape stuff doing error handling, (abort */
2107         /* or reset) then we don't expect anything weird. */
2108         if (cmd != CCISS_RESET_MSG && cmd != CCISS_ABORT_MSG) {
2109 #endif
2110                 printk(KERN_WARNING "cciss cciss%d: SendCmd "
2111                        "Invalid command list address returned! (%lx)\n",
2112                        ctlr, complete);
2113                 /* not much we can do. */
2114 #ifdef CONFIG_CISS_SCSI_TAPE
2115                 return 1;
2116         }
2117
2118         /* We've sent down an abort or reset, but something else
2119            has completed */
2120         if (srl->ncompletions >= (hba[ctlr]->nr_cmds + 2)) {
2121                 /* Uh oh.  No room to save it for later... */
2122                 printk(KERN_WARNING "cciss%d: Sendcmd: Invalid command addr, "
2123                        "reject list overflow, command lost!\n", ctlr);
2124                 return 1;
2125         }
2126         /* Save it for later */
2127         srl->complete[srl->ncompletions] = complete;
2128         srl->ncompletions++;
2129 #endif
2130         return 0;
2131 }
2132
2133 /*
2134  * Send a command to the controller, and wait for it to complete.
2135  * Only used at init time.
2136  */
2137 static int sendcmd(__u8 cmd, int ctlr, void *buff, size_t size, unsigned int use_unit_num,      /* 0: address the controller,
2138                                                                                                    1: address logical volume log_unit,
2139                                                                                                    2: periph device address is scsi3addr */
2140                    unsigned int log_unit,
2141                    __u8 page_code, unsigned char *scsi3addr, int cmd_type)
2142 {
2143         CommandList_struct *c;
2144         int i;
2145         unsigned long complete;
2146         ctlr_info_t *info_p = hba[ctlr];
2147         u64bit buff_dma_handle;
2148         int status, done = 0;
2149
2150         if ((c = cmd_alloc(info_p, 1)) == NULL) {
2151                 printk(KERN_WARNING "cciss: unable to get memory");
2152                 return IO_ERROR;
2153         }
2154         status = fill_cmd(c, cmd, ctlr, buff, size, use_unit_num,
2155                           log_unit, page_code, scsi3addr, cmd_type);
2156         if (status != IO_OK) {
2157                 cmd_free(info_p, c, 1);
2158                 return status;
2159         }
2160       resend_cmd1:
2161         /*
2162          * Disable interrupt
2163          */
2164 #ifdef CCISS_DEBUG
2165         printk(KERN_DEBUG "cciss: turning intr off\n");
2166 #endif                          /* CCISS_DEBUG */
2167         info_p->access.set_intr_mask(info_p, CCISS_INTR_OFF);
2168
2169         /* Make sure there is room in the command FIFO */
2170         /* Actually it should be completely empty at this time */
2171         /* unless we are in here doing error handling for the scsi */
2172         /* tape side of the driver. */
2173         for (i = 200000; i > 0; i--) {
2174                 /* if fifo isn't full go */
2175                 if (!(info_p->access.fifo_full(info_p))) {
2176
2177                         break;
2178                 }
2179                 udelay(10);
2180                 printk(KERN_WARNING "cciss cciss%d: SendCmd FIFO full,"
2181                        " waiting!\n", ctlr);
2182         }
2183         /*
2184          * Send the cmd
2185          */
2186         info_p->access.submit_command(info_p, c);
2187         done = 0;
2188         do {
2189                 complete = pollcomplete(ctlr);
2190
2191 #ifdef CCISS_DEBUG
2192                 printk(KERN_DEBUG "cciss: command completed\n");
2193 #endif                          /* CCISS_DEBUG */
2194
2195                 if (complete == 1) {
2196                         printk(KERN_WARNING
2197                                "cciss cciss%d: SendCmd Timeout out, "
2198                                "No command list address returned!\n", ctlr);
2199                         status = IO_ERROR;
2200                         done = 1;
2201                         break;
2202                 }
2203
2204                 /* This will need to change for direct lookup completions */
2205                 if ((complete & CISS_ERROR_BIT)
2206                     && (complete & ~CISS_ERROR_BIT) == c->busaddr) {
2207                         /* if data overrun or underun on Report command
2208                            ignore it
2209                          */
2210                         if (((c->Request.CDB[0] == CISS_REPORT_LOG) ||
2211                              (c->Request.CDB[0] == CISS_REPORT_PHYS) ||
2212                              (c->Request.CDB[0] == CISS_INQUIRY)) &&
2213                             ((c->err_info->CommandStatus ==
2214                               CMD_DATA_OVERRUN) ||
2215                              (c->err_info->CommandStatus == CMD_DATA_UNDERRUN)
2216                             )) {
2217                                 complete = c->busaddr;
2218                         } else {
2219                                 if (c->err_info->CommandStatus ==
2220                                     CMD_UNSOLICITED_ABORT) {
2221                                         printk(KERN_WARNING "cciss%d: "
2222                                                "unsolicited abort %p\n",
2223                                                ctlr, c);
2224                                         if (c->retry_count < MAX_CMD_RETRIES) {
2225                                                 printk(KERN_WARNING
2226                                                        "cciss%d: retrying %p\n",
2227                                                        ctlr, c);
2228                                                 c->retry_count++;
2229                                                 /* erase the old error */
2230                                                 /* information */
2231                                                 memset(c->err_info, 0,
2232                                                        sizeof
2233                                                        (ErrorInfo_struct));
2234                                                 goto resend_cmd1;
2235                                         } else {
2236                                                 printk(KERN_WARNING
2237                                                        "cciss%d: retried %p too "
2238                                                        "many times\n", ctlr, c);
2239                                                 status = IO_ERROR;
2240                                                 goto cleanup1;
2241                                         }
2242                                 } else if (c->err_info->CommandStatus ==
2243                                            CMD_UNABORTABLE) {
2244                                         printk(KERN_WARNING
2245                                                "cciss%d: command could not be aborted.\n",
2246                                                ctlr);
2247                                         status = IO_ERROR;
2248                                         goto cleanup1;
2249                                 }
2250                                 printk(KERN_WARNING "ciss ciss%d: sendcmd"
2251                                        " Error %x \n", ctlr,
2252                                        c->err_info->CommandStatus);
2253                                 printk(KERN_WARNING "ciss ciss%d: sendcmd"
2254                                        " offensive info\n"
2255                                        "  size %x\n   num %x   value %x\n",
2256                                        ctlr,
2257                                        c->err_info->MoreErrInfo.Invalid_Cmd.
2258                                        offense_size,
2259                                        c->err_info->MoreErrInfo.Invalid_Cmd.
2260                                        offense_num,
2261                                        c->err_info->MoreErrInfo.Invalid_Cmd.
2262                                        offense_value);
2263                                 status = IO_ERROR;
2264                                 goto cleanup1;
2265                         }
2266                 }
2267                 /* This will need changing for direct lookup completions */
2268                 if (complete != c->busaddr) {
2269                         if (add_sendcmd_reject(cmd, ctlr, complete) != 0) {
2270                                 BUG();  /* we are pretty much hosed if we get here. */
2271                         }
2272                         continue;
2273                 } else
2274                         done = 1;
2275         } while (!done);
2276
2277       cleanup1:
2278         /* unlock the data buffer from DMA */
2279         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2280         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2281         pci_unmap_single(info_p->pdev, (dma_addr_t) buff_dma_handle.val,
2282                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2283 #ifdef CONFIG_CISS_SCSI_TAPE
2284         /* if we saved some commands for later, process them now. */
2285         if (info_p->scsi_rejects.ncompletions > 0)
2286                 do_cciss_intr(0, info_p);
2287 #endif
2288         cmd_free(info_p, c, 1);
2289         return status;
2290 }
2291
2292 /*
2293  * Map (physical) PCI mem into (virtual) kernel space
2294  */
2295 static void __iomem *remap_pci_mem(ulong base, ulong size)
2296 {
2297         ulong page_base = ((ulong) base) & PAGE_MASK;
2298         ulong page_offs = ((ulong) base) - page_base;
2299         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2300
2301         return page_remapped ? (page_remapped + page_offs) : NULL;
2302 }
2303
2304 /*
2305  * Takes jobs of the Q and sends them to the hardware, then puts it on
2306  * the Q to wait for completion.
2307  */
2308 static void start_io(ctlr_info_t *h)
2309 {
2310         CommandList_struct *c;
2311
2312         while ((c = h->reqQ) != NULL) {
2313                 /* can't do anything if fifo is full */
2314                 if ((h->access.fifo_full(h))) {
2315                         printk(KERN_WARNING "cciss: fifo full\n");
2316                         break;
2317                 }
2318
2319                 /* Get the first entry from the Request Q */
2320                 removeQ(&(h->reqQ), c);
2321                 h->Qdepth--;
2322
2323                 /* Tell the controller execute command */
2324                 h->access.submit_command(h, c);
2325
2326                 /* Put job onto the completed Q */
2327                 addQ(&(h->cmpQ), c);
2328         }
2329 }
2330
2331 /* Assumes that CCISS_LOCK(h->ctlr) is held. */
2332 /* Zeros out the error record and then resends the command back */
2333 /* to the controller */
2334 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2335 {
2336         /* erase the old error information */
2337         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2338
2339         /* add it to software queue and then send it to the controller */
2340         addQ(&(h->reqQ), c);
2341         h->Qdepth++;
2342         if (h->Qdepth > h->maxQsinceinit)
2343                 h->maxQsinceinit = h->Qdepth;
2344
2345         start_io(h);
2346 }
2347
2348 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
2349         unsigned int msg_byte, unsigned int host_byte,
2350         unsigned int driver_byte)
2351 {
2352         /* inverse of macros in scsi.h */
2353         return (scsi_status_byte & 0xff) |
2354                 ((msg_byte & 0xff) << 8) |
2355                 ((host_byte & 0xff) << 16) |
2356                 ((driver_byte & 0xff) << 24);
2357 }
2358
2359 static inline int evaluate_target_status(CommandList_struct *cmd)
2360 {
2361         unsigned char sense_key;
2362         unsigned char status_byte, msg_byte, host_byte, driver_byte;
2363         int error_value;
2364
2365         /* If we get in here, it means we got "target status", that is, scsi status */
2366         status_byte = cmd->err_info->ScsiStatus;
2367         driver_byte = DRIVER_OK;
2368         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
2369
2370         if (blk_pc_request(cmd->rq))
2371                 host_byte = DID_PASSTHROUGH;
2372         else
2373                 host_byte = DID_OK;
2374
2375         error_value = make_status_bytes(status_byte, msg_byte,
2376                 host_byte, driver_byte);
2377
2378         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
2379                 if (!blk_pc_request(cmd->rq))
2380                         printk(KERN_WARNING "cciss: cmd %p "
2381                                "has SCSI Status 0x%x\n",
2382                                cmd, cmd->err_info->ScsiStatus);
2383                 return error_value;
2384         }
2385
2386         /* check the sense key */
2387         sense_key = 0xf & cmd->err_info->SenseInfo[2];
2388         /* no status or recovered error */
2389         if (((sense_key == 0x0) || (sense_key == 0x1)) && !blk_pc_request(cmd->rq))
2390                 error_value = 0;
2391
2392         if (!blk_pc_request(cmd->rq)) { /* Not SG_IO or similar? */
2393                 if (error_value != 0)
2394                         printk(KERN_WARNING "cciss: cmd %p has CHECK CONDITION"
2395                                " sense key = 0x%x\n", cmd, sense_key);
2396                 return error_value;
2397         }
2398
2399         /* SG_IO or similar, copy sense data back */
2400         if (cmd->rq->sense) {
2401                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
2402                         cmd->rq->sense_len = cmd->err_info->SenseLen;
2403                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
2404                         cmd->rq->sense_len);
2405         } else
2406                 cmd->rq->sense_len = 0;
2407
2408         return error_value;
2409 }
2410
2411 /* checks the status of the job and calls complete buffers to mark all
2412  * buffers for the completed job. Note that this function does not need
2413  * to hold the hba/queue lock.
2414  */
2415 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
2416                                     int timeout)
2417 {
2418         int retry_cmd = 0;
2419         struct request *rq = cmd->rq;
2420
2421         rq->errors = 0;
2422
2423         if (timeout)
2424                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
2425
2426         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
2427                 goto after_error_processing;
2428
2429         switch (cmd->err_info->CommandStatus) {
2430         case CMD_TARGET_STATUS:
2431                 rq->errors = evaluate_target_status(cmd);
2432                 break;
2433         case CMD_DATA_UNDERRUN:
2434                 if (blk_fs_request(cmd->rq)) {
2435                         printk(KERN_WARNING "cciss: cmd %p has"
2436                                " completed with data underrun "
2437                                "reported\n", cmd);
2438                         cmd->rq->data_len = cmd->err_info->ResidualCnt;
2439                 }
2440                 break;
2441         case CMD_DATA_OVERRUN:
2442                 if (blk_fs_request(cmd->rq))
2443                         printk(KERN_WARNING "cciss: cmd %p has"
2444                                " completed with data overrun "
2445                                "reported\n", cmd);
2446                 break;
2447         case CMD_INVALID:
2448                 printk(KERN_WARNING "cciss: cmd %p is "
2449                        "reported invalid\n", cmd);
2450                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2451                         cmd->err_info->CommandStatus, DRIVER_OK,
2452                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2453                 break;
2454         case CMD_PROTOCOL_ERR:
2455                 printk(KERN_WARNING "cciss: cmd %p has "
2456                        "protocol error \n", cmd);
2457                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2458                         cmd->err_info->CommandStatus, DRIVER_OK,
2459                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2460                 break;
2461         case CMD_HARDWARE_ERR:
2462                 printk(KERN_WARNING "cciss: cmd %p had "
2463                        " hardware error\n", cmd);
2464                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2465                         cmd->err_info->CommandStatus, DRIVER_OK,
2466                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2467                 break;
2468         case CMD_CONNECTION_LOST:
2469                 printk(KERN_WARNING "cciss: cmd %p had "
2470                        "connection lost\n", cmd);
2471                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2472                         cmd->err_info->CommandStatus, DRIVER_OK,
2473                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2474                 break;
2475         case CMD_ABORTED:
2476                 printk(KERN_WARNING "cciss: cmd %p was "
2477                        "aborted\n", cmd);
2478                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2479                         cmd->err_info->CommandStatus, DRIVER_OK,
2480                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
2481                 break;
2482         case CMD_ABORT_FAILED:
2483                 printk(KERN_WARNING "cciss: cmd %p reports "
2484                        "abort failed\n", cmd);
2485                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2486                         cmd->err_info->CommandStatus, DRIVER_OK,
2487                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2488                 break;
2489         case CMD_UNSOLICITED_ABORT:
2490                 printk(KERN_WARNING "cciss%d: unsolicited "
2491                        "abort %p\n", h->ctlr, cmd);
2492                 if (cmd->retry_count < MAX_CMD_RETRIES) {
2493                         retry_cmd = 1;
2494                         printk(KERN_WARNING
2495                                "cciss%d: retrying %p\n", h->ctlr, cmd);
2496                         cmd->retry_count++;
2497                 } else
2498                         printk(KERN_WARNING
2499                                "cciss%d: %p retried too "
2500                                "many times\n", h->ctlr, cmd);
2501                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2502                         cmd->err_info->CommandStatus, DRIVER_OK,
2503                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
2504                 break;
2505         case CMD_TIMEOUT:
2506                 printk(KERN_WARNING "cciss: cmd %p timedout\n", cmd);
2507                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2508                         cmd->err_info->CommandStatus, DRIVER_OK,
2509                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2510                 break;
2511         default:
2512                 printk(KERN_WARNING "cciss: cmd %p returned "
2513                        "unknown status %x\n", cmd,
2514                        cmd->err_info->CommandStatus);
2515                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2516                         cmd->err_info->CommandStatus, DRIVER_OK,
2517                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2518         }
2519
2520 after_error_processing:
2521
2522         /* We need to return this command */
2523         if (retry_cmd) {
2524                 resend_cciss_cmd(h, cmd);
2525                 return;
2526         }
2527         cmd->rq->data_len = 0;
2528         cmd->rq->completion_data = cmd;
2529         blk_complete_request(cmd->rq);
2530 }
2531
2532 /*
2533  * Get a request and submit it to the controller.
2534  */
2535 static void do_cciss_request(struct request_queue *q)
2536 {
2537         ctlr_info_t *h = q->queuedata;
2538         CommandList_struct *c;
2539         sector_t start_blk;
2540         int seg;
2541         struct request *creq;
2542         u64bit temp64;
2543         struct scatterlist tmp_sg[MAXSGENTRIES];
2544         drive_info_struct *drv;
2545         int i, dir;
2546
2547         /* We call start_io here in case there is a command waiting on the
2548          * queue that has not been sent.
2549          */
2550         if (blk_queue_plugged(q))
2551                 goto startio;
2552
2553       queue:
2554         creq = elv_next_request(q);
2555         if (!creq)
2556                 goto startio;
2557
2558         BUG_ON(creq->nr_phys_segments > MAXSGENTRIES);
2559
2560         if ((c = cmd_alloc(h, 1)) == NULL)
2561                 goto full;
2562
2563         blkdev_dequeue_request(creq);
2564
2565         spin_unlock_irq(q->queue_lock);
2566
2567         c->cmd_type = CMD_RWREQ;
2568         c->rq = creq;
2569
2570         /* fill in the request */
2571         drv = creq->rq_disk->private_data;
2572         c->Header.ReplyQueue = 0;       // unused in simple mode
2573         /* got command from pool, so use the command block index instead */
2574         /* for direct lookups. */
2575         /* The first 2 bits are reserved for controller error reporting. */
2576         c->Header.Tag.lower = (c->cmdindex << 3);
2577         c->Header.Tag.lower |= 0x04;    /* flag for direct lookup. */
2578         c->Header.LUN.LogDev.VolId = drv->LunID;
2579         c->Header.LUN.LogDev.Mode = 1;
2580         c->Request.CDBLen = 10; // 12 byte commands not in FW yet;
2581         c->Request.Type.Type = TYPE_CMD;        // It is a command.
2582         c->Request.Type.Attribute = ATTR_SIMPLE;
2583         c->Request.Type.Direction =
2584             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
2585         c->Request.Timeout = 0; // Don't time out
2586         c->Request.CDB[0] =
2587             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
2588         start_blk = creq->sector;
2589 #ifdef CCISS_DEBUG
2590         printk(KERN_DEBUG "ciss: sector =%d nr_sectors=%d\n", (int)creq->sector,
2591                (int)creq->nr_sectors);
2592 #endif                          /* CCISS_DEBUG */
2593
2594         sg_init_table(tmp_sg, MAXSGENTRIES);
2595         seg = blk_rq_map_sg(q, creq, tmp_sg);
2596
2597         /* get the DMA records for the setup */
2598         if (c->Request.Type.Direction == XFER_READ)
2599                 dir = PCI_DMA_FROMDEVICE;
2600         else
2601                 dir = PCI_DMA_TODEVICE;
2602
2603         for (i = 0; i < seg; i++) {
2604                 c->SG[i].Len = tmp_sg[i].length;
2605                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
2606                                                   tmp_sg[i].offset,
2607                                                   tmp_sg[i].length, dir);
2608                 c->SG[i].Addr.lower = temp64.val32.lower;
2609                 c->SG[i].Addr.upper = temp64.val32.upper;
2610                 c->SG[i].Ext = 0;       // we are not chaining
2611         }
2612         /* track how many SG entries we are using */
2613         if (seg > h->maxSG)
2614                 h->maxSG = seg;
2615
2616 #ifdef CCISS_DEBUG
2617         printk(KERN_DEBUG "cciss: Submitting %d sectors in %d segments\n",
2618                creq->nr_sectors, seg);
2619 #endif                          /* CCISS_DEBUG */
2620
2621         c->Header.SGList = c->Header.SGTotal = seg;
2622         if (likely(blk_fs_request(creq))) {
2623                 if(h->cciss_read == CCISS_READ_10) {
2624                         c->Request.CDB[1] = 0;
2625                         c->Request.CDB[2] = (start_blk >> 24) & 0xff;   //MSB
2626                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
2627                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
2628                         c->Request.CDB[5] = start_blk & 0xff;
2629                         c->Request.CDB[6] = 0;  // (sect >> 24) & 0xff; MSB
2630                         c->Request.CDB[7] = (creq->nr_sectors >> 8) & 0xff;
2631                         c->Request.CDB[8] = creq->nr_sectors & 0xff;
2632                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
2633                 } else {
2634                         c->Request.CDBLen = 16;
2635                         c->Request.CDB[1]= 0;
2636                         c->Request.CDB[2]= (start_blk >> 56) & 0xff;    //MSB
2637                         c->Request.CDB[3]= (start_blk >> 48) & 0xff;
2638                         c->Request.CDB[4]= (start_blk >> 40) & 0xff;
2639                         c->Request.CDB[5]= (start_blk >> 32) & 0xff;
2640                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
2641                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
2642                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
2643                         c->Request.CDB[9]= start_blk & 0xff;
2644                         c->Request.CDB[10]= (creq->nr_sectors >>  24) & 0xff;
2645                         c->Request.CDB[11]= (creq->nr_sectors >>  16) & 0xff;
2646                         c->Request.CDB[12]= (creq->nr_sectors >>  8) & 0xff;
2647                         c->Request.CDB[13]= creq->nr_sectors & 0xff;
2648                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
2649                 }
2650         } else if (blk_pc_request(creq)) {
2651                 c->Request.CDBLen = creq->cmd_len;
2652                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
2653         } else {
2654                 printk(KERN_WARNING "cciss%d: bad request type %d\n", h->ctlr, creq->cmd_type);
2655                 BUG();
2656         }
2657
2658         spin_lock_irq(q->queue_lock);
2659
2660         addQ(&(h->reqQ), c);
2661         h->Qdepth++;
2662         if (h->Qdepth > h->maxQsinceinit)
2663                 h->maxQsinceinit = h->Qdepth;
2664
2665         goto queue;
2666 full:
2667         blk_stop_queue(q);
2668 startio:
2669         /* We will already have the driver lock here so not need
2670          * to lock it.
2671          */
2672         start_io(h);
2673 }
2674
2675 static inline unsigned long get_next_completion(ctlr_info_t *h)
2676 {
2677 #ifdef CONFIG_CISS_SCSI_TAPE
2678         /* Any rejects from sendcmd() lying around? Process them first */
2679         if (h->scsi_rejects.ncompletions == 0)
2680                 return h->access.command_completed(h);
2681         else {
2682                 struct sendcmd_reject_list *srl;
2683                 int n;
2684                 srl = &h->scsi_rejects;
2685                 n = --srl->ncompletions;
2686                 /* printk("cciss%d: processing saved reject\n", h->ctlr); */
2687                 printk("p");
2688                 return srl->complete[n];
2689         }
2690 #else
2691         return h->access.command_completed(h);
2692 #endif
2693 }
2694
2695 static inline int interrupt_pending(ctlr_info_t *h)
2696 {
2697 #ifdef CONFIG_CISS_SCSI_TAPE
2698         return (h->access.intr_pending(h)
2699                 || (h->scsi_rejects.ncompletions > 0));
2700 #else
2701         return h->access.intr_pending(h);
2702 #endif
2703 }
2704
2705 static inline long interrupt_not_for_us(ctlr_info_t *h)
2706 {
2707 #ifdef CONFIG_CISS_SCSI_TAPE
2708         return (((h->access.intr_pending(h) == 0) ||
2709                  (h->interrupts_enabled == 0))
2710                 && (h->scsi_rejects.ncompletions == 0));
2711 #else
2712         return (((h->access.intr_pending(h) == 0) ||
2713                  (h->interrupts_enabled == 0)));
2714 #endif
2715 }
2716
2717 static irqreturn_t do_cciss_intr(int irq, void *dev_id)
2718 {
2719         ctlr_info_t *h = dev_id;
2720         CommandList_struct *c;
2721         unsigned long flags;
2722         __u32 a, a1, a2;
2723
2724         if (interrupt_not_for_us(h))
2725                 return IRQ_NONE;
2726         /*
2727          * If there are completed commands in the completion queue,
2728          * we had better do something about it.
2729          */
2730         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2731         while (interrupt_pending(h)) {
2732                 while ((a = get_next_completion(h)) != FIFO_EMPTY) {
2733                         a1 = a;
2734                         if ((a & 0x04)) {
2735                                 a2 = (a >> 3);
2736                                 if (a2 >= h->nr_cmds) {
2737                                         printk(KERN_WARNING
2738                                                "cciss: controller cciss%d failed, stopping.\n",
2739                                                h->ctlr);
2740                                         fail_all_cmds(h->ctlr);
2741                                         return IRQ_HANDLED;
2742                                 }
2743
2744                                 c = h->cmd_pool + a2;
2745                                 a = c->busaddr;
2746
2747                         } else {
2748                                 a &= ~3;
2749                                 if ((c = h->cmpQ) == NULL) {
2750                                         printk(KERN_WARNING
2751                                                "cciss: Completion of %08x ignored\n",
2752                                                a1);
2753                                         continue;
2754                                 }
2755                                 while (c->busaddr != a) {
2756                                         c = c->next;
2757                                         if (c == h->cmpQ)
2758                                                 break;
2759                                 }
2760                         }
2761                         /*
2762                          * If we've found the command, take it off the
2763                          * completion Q and free it
2764                          */
2765                         if (c->busaddr == a) {
2766                                 removeQ(&h->cmpQ, c);
2767                                 if (c->cmd_type == CMD_RWREQ) {
2768                                         complete_command(h, c, 0);
2769                                 } else if (c->cmd_type == CMD_IOCTL_PEND) {
2770                                         complete(c->waiting);
2771                                 }
2772 #                               ifdef CONFIG_CISS_SCSI_TAPE
2773                                 else if (c->cmd_type == CMD_SCSI)
2774                                         complete_scsi_command(c, 0, a1);
2775 #                               endif
2776                                 continue;
2777                         }
2778                 }
2779         }
2780
2781         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2782         return IRQ_HANDLED;
2783 }
2784
2785 /*
2786  *  We cannot read the structure directly, for portability we must use
2787  *   the io functions.
2788  *   This is for debug only.
2789  */
2790 #ifdef CCISS_DEBUG
2791 static void print_cfg_table(CfgTable_struct *tb)
2792 {
2793         int i;
2794         char temp_name[17];
2795
2796         printk("Controller Configuration information\n");
2797         printk("------------------------------------\n");
2798         for (i = 0; i < 4; i++)
2799                 temp_name[i] = readb(&(tb->Signature[i]));
2800         temp_name[4] = '\0';
2801         printk("   Signature = %s\n", temp_name);
2802         printk("   Spec Number = %d\n", readl(&(tb->SpecValence)));
2803         printk("   Transport methods supported = 0x%x\n",
2804                readl(&(tb->TransportSupport)));
2805         printk("   Transport methods active = 0x%x\n",
2806                readl(&(tb->TransportActive)));
2807         printk("   Requested transport Method = 0x%x\n",
2808                readl(&(tb->HostWrite.TransportRequest)));
2809         printk("   Coalesce Interrupt Delay = 0x%x\n",
2810                readl(&(tb->HostWrite.CoalIntDelay)));
2811         printk("   Coalesce Interrupt Count = 0x%x\n",
2812                readl(&(tb->HostWrite.CoalIntCount)));
2813         printk("   Max outstanding commands = 0x%d\n",
2814                readl(&(tb->CmdsOutMax)));
2815         printk("   Bus Types = 0x%x\n", readl(&(tb->BusTypes)));
2816         for (i = 0; i < 16; i++)
2817                 temp_name[i] = readb(&(tb->ServerName[i]));
2818         temp_name[16] = '\0';
2819         printk("   Server Name = %s\n", temp_name);
2820         printk("   Heartbeat Counter = 0x%x\n\n\n", readl(&(tb->HeartBeat)));
2821 }
2822 #endif                          /* CCISS_DEBUG */
2823
2824 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
2825 {
2826         int i, offset, mem_type, bar_type;
2827         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
2828                 return 0;
2829         offset = 0;
2830         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
2831                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
2832                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
2833                         offset += 4;
2834                 else {
2835                         mem_type = pci_resource_flags(pdev, i) &
2836                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
2837                         switch (mem_type) {
2838                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
2839                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
2840                                 offset += 4;    /* 32 bit */
2841                                 break;
2842                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
2843                                 offset += 8;
2844                                 break;
2845                         default:        /* reserved in PCI 2.2 */
2846                                 printk(KERN_WARNING
2847                                        "Base address is invalid\n");
2848                                 return -1;
2849                                 break;
2850                         }
2851                 }
2852                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
2853                         return i + 1;
2854         }
2855         return -1;
2856 }
2857
2858 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
2859  * controllers that are capable. If not, we use IO-APIC mode.
2860  */
2861
2862 static void __devinit cciss_interrupt_mode(ctlr_info_t *c,
2863                                            struct pci_dev *pdev, __u32 board_id)
2864 {
2865 #ifdef CONFIG_PCI_MSI
2866         int err;
2867         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
2868         {0, 2}, {0, 3}
2869         };
2870
2871         /* Some boards advertise MSI but don't really support it */
2872         if ((board_id == 0x40700E11) ||
2873             (board_id == 0x40800E11) ||
2874             (board_id == 0x40820E11) || (board_id == 0x40830E11))
2875                 goto default_int_mode;
2876
2877         if (pci_find_capability(pdev, PCI_CAP_ID_MSIX)) {
2878                 err = pci_enable_msix(pdev, cciss_msix_entries, 4);
2879                 if (!err) {
2880                         c->intr[0] = cciss_msix_entries[0].vector;
2881                         c->intr[1] = cciss_msix_entries[1].vector;
2882                         c->intr[2] = cciss_msix_entries[2].vector;
2883                         c->intr[3] = cciss_msix_entries[3].vector;
2884                         c->msix_vector = 1;
2885                         return;
2886                 }
2887                 if (err > 0) {
2888                         printk(KERN_WARNING "cciss: only %d MSI-X vectors "
2889                                "available\n", err);
2890                         goto default_int_mode;
2891                 } else {
2892                         printk(KERN_WARNING "cciss: MSI-X init failed %d\n",
2893                                err);
2894                         goto default_int_mode;
2895                 }
2896         }
2897         if (pci_find_capability(pdev, PCI_CAP_ID_MSI)) {
2898                 if (!pci_enable_msi(pdev)) {
2899                         c->msi_vector = 1;
2900                 } else {
2901                         printk(KERN_WARNING "cciss: MSI init failed\n");
2902                 }
2903         }
2904 default_int_mode:
2905 #endif                          /* CONFIG_PCI_MSI */
2906         /* if we get here we're going to use the default interrupt mode */
2907         c->intr[SIMPLE_MODE_INT] = pdev->irq;
2908         return;
2909 }
2910
2911 static int __devinit cciss_pci_init(ctlr_info_t *c, struct pci_dev *pdev)
2912 {
2913         ushort subsystem_vendor_id, subsystem_device_id, command;
2914         __u32 board_id, scratchpad = 0;
2915         __u64 cfg_offset;
2916         __u32 cfg_base_addr;
2917         __u64 cfg_base_addr_index;
2918         int i, err;
2919
2920         /* check to see if controller has been disabled */
2921         /* BEFORE trying to enable it */
2922         (void)pci_read_config_word(pdev, PCI_COMMAND, &command);
2923         if (!(command & 0x02)) {
2924                 printk(KERN_WARNING
2925                        "cciss: controller appears to be disabled\n");
2926                 return -ENODEV;
2927         }
2928
2929         err = pci_enable_device(pdev);
2930         if (err) {
2931                 printk(KERN_ERR "cciss: Unable to Enable PCI device\n");
2932                 return err;
2933         }
2934
2935         err = pci_request_regions(pdev, "cciss");
2936         if (err) {
2937                 printk(KERN_ERR "cciss: Cannot obtain PCI resources, "
2938                        "aborting\n");
2939                 return err;
2940         }
2941
2942         subsystem_vendor_id = pdev->subsystem_vendor;
2943         subsystem_device_id = pdev->subsystem_device;
2944         board_id = (((__u32) (subsystem_device_id << 16) & 0xffff0000) |
2945                     subsystem_vendor_id);
2946
2947 #ifdef CCISS_DEBUG
2948         printk("command = %x\n", command);
2949         printk("irq = %x\n", pdev->irq);
2950         printk("board_id = %x\n", board_id);
2951 #endif                          /* CCISS_DEBUG */
2952
2953 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
2954  * else we use the IO-APIC interrupt assigned to us by system ROM.
2955  */
2956         cciss_interrupt_mode(c, pdev, board_id);
2957
2958         /*
2959          * Memory base addr is first addr , the second points to the config
2960          *   table
2961          */
2962
2963         c->paddr = pci_resource_start(pdev, 0); /* addressing mode bits already removed */
2964 #ifdef CCISS_DEBUG
2965         printk("address 0 = %x\n", c->paddr);
2966 #endif                          /* CCISS_DEBUG */
2967         c->vaddr = remap_pci_mem(c->paddr, 0x250);
2968
2969         /* Wait for the board to become ready.  (PCI hotplug needs this.)
2970          * We poll for up to 120 secs, once per 100ms. */
2971         for (i = 0; i < 1200; i++) {
2972                 scratchpad = readl(c->vaddr + SA5_SCRATCHPAD_OFFSET);
2973                 if (scratchpad == CCISS_FIRMWARE_READY)
2974                         break;
2975                 set_current_state(TASK_INTERRUPTIBLE);
2976                 schedule_timeout(HZ / 10);      /* wait 100ms */
2977         }
2978         if (scratchpad != CCISS_FIRMWARE_READY) {
2979                 printk(KERN_WARNING "cciss: Board not ready.  Timed out.\n");
2980                 err = -ENODEV;
2981                 goto err_out_free_res;
2982         }
2983
2984         /* get the address index number */
2985         cfg_base_addr = readl(c->vaddr + SA5_CTCFG_OFFSET);
2986         cfg_base_addr &= (__u32) 0x0000ffff;
2987 #ifdef CCISS_DEBUG
2988         printk("cfg base address = %x\n", cfg_base_addr);
2989 #endif                          /* CCISS_DEBUG */
2990         cfg_base_addr_index = find_PCI_BAR_index(pdev, cfg_base_addr);
2991 #ifdef CCISS_DEBUG
2992         printk("cfg base address index = %x\n", cfg_base_addr_index);
2993 #endif                          /* CCISS_DEBUG */
2994         if (cfg_base_addr_index == -1) {
2995                 printk(KERN_WARNING "cciss: Cannot find cfg_base_addr_index\n");
2996                 err = -ENODEV;
2997                 goto err_out_free_res;
2998         }
2999
3000         cfg_offset = readl(c->vaddr + SA5_CTMEM_OFFSET);
3001 #ifdef CCISS_DEBUG
3002         printk("cfg offset = %x\n", cfg_offset);
3003 #endif                          /* CCISS_DEBUG */
3004         c->cfgtable = remap_pci_mem(pci_resource_start(pdev,
3005                                                        cfg_base_addr_index) +
3006                                     cfg_offset, sizeof(CfgTable_struct));
3007         c->board_id = board_id;
3008
3009 #ifdef CCISS_DEBUG
3010         print_cfg_table(c->cfgtable);
3011 #endif                          /* CCISS_DEBUG */
3012
3013         for (i = 0; i < ARRAY_SIZE(products); i++) {
3014                 if (board_id == products[i].board_id) {
3015                         c->product_name = products[i].product_name;
3016                         c->access = *(products[i].access);
3017                         c->nr_cmds = products[i].nr_cmds;
3018                         break;
3019                 }
3020         }
3021         if ((readb(&c->cfgtable->Signature[0]) != 'C') ||
3022             (readb(&c->cfgtable->Signature[1]) != 'I') ||
3023             (readb(&c->cfgtable->Signature[2]) != 'S') ||
3024             (readb(&c->cfgtable->Signature[3]) != 'S')) {
3025                 printk("Does not appear to be a valid CISS config table\n");
3026                 err = -ENODEV;
3027                 goto err_out_free_res;
3028         }
3029         /* We didn't find the controller in our list. We know the
3030          * signature is valid. If it's an HP device let's try to
3031          * bind to the device and fire it up. Otherwise we bail.
3032          */
3033         if (i == ARRAY_SIZE(products)) {
3034                 if (subsystem_vendor_id == PCI_VENDOR_ID_HP) {
3035                         c->product_name = products[i-1].product_name;
3036                         c->access = *(products[i-1].access);
3037                         c->nr_cmds = products[i-1].nr_cmds;
3038                         printk(KERN_WARNING "cciss: This is an unknown "
3039                                 "Smart Array controller.\n"
3040                                 "cciss: Please update to the latest driver "
3041                                 "available from www.hp.com.\n");
3042                 } else {
3043                         printk(KERN_WARNING "cciss: Sorry, I don't know how"
3044                                 " to access the Smart Array controller %08lx\n"
3045                                         , (unsigned long)board_id);
3046                         err = -ENODEV;
3047                         goto err_out_free_res;
3048                 }
3049         }
3050 #ifdef CONFIG_X86
3051         {
3052                 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
3053                 __u32 prefetch;
3054                 prefetch = readl(&(c->cfgtable->SCSI_Prefetch));
3055                 prefetch |= 0x100;
3056                 writel(prefetch, &(c->cfgtable->SCSI_Prefetch));
3057         }
3058 #endif
3059
3060         /* Disabling DMA prefetch and refetch for the P600.
3061          * An ASIC bug may result in accesses to invalid memory addresses.
3062          * We've disabled prefetch for some time now. Testing with XEN
3063          * kernels revealed a bug in the refetch if dom0 resides on a P600.
3064          */
3065         if(board_id == 0x3225103C) {
3066                 __u32 dma_prefetch;
3067                 __u32 dma_refetch;
3068                 dma_prefetch = readl(c->vaddr + I2O_DMA1_CFG);
3069                 dma_prefetch |= 0x8000;
3070                 writel(dma_prefetch, c->vaddr + I2O_DMA1_CFG);
3071                 pci_read_config_dword(pdev, PCI_COMMAND_PARITY, &dma_refetch);
3072                 dma_refetch |= 0x1;
3073                 pci_write_config_dword(pdev, PCI_COMMAND_PARITY, dma_refetch);
3074         }
3075
3076 #ifdef CCISS_DEBUG
3077         printk("Trying to put board into Simple mode\n");
3078 #endif                          /* CCISS_DEBUG */
3079         c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
3080         /* Update the field, and then ring the doorbell */
3081         writel(CFGTBL_Trans_Simple, &(c->cfgtable->HostWrite.TransportRequest));
3082         writel(CFGTBL_ChangeReq, c->vaddr + SA5_DOORBELL);
3083
3084         /* under certain very rare conditions, this can take awhile.
3085          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3086          * as we enter this code.) */
3087         for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3088                 if (!(readl(c->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3089                         break;
3090                 /* delay and try again */
3091                 set_current_state(TASK_INTERRUPTIBLE);
3092                 schedule_timeout(10);
3093         }
3094
3095 #ifdef CCISS_DEBUG
3096         printk(KERN_DEBUG "I counter got to %d %x\n", i,
3097                readl(c->vaddr + SA5_DOORBELL));
3098 #endif                          /* CCISS_DEBUG */
3099 #ifdef CCISS_DEBUG
3100         print_cfg_table(c->cfgtable);
3101 #endif                          /* CCISS_DEBUG */
3102
3103         if (!(readl(&(c->cfgtable->TransportActive)) & CFGTBL_Trans_Simple)) {
3104                 printk(KERN_WARNING "cciss: unable to get board into"
3105                        " simple mode\n");
3106                 err = -ENODEV;
3107                 goto err_out_free_res;
3108         }
3109         return 0;
3110
3111 err_out_free_res:
3112         /*
3113          * Deliberately omit pci_disable_device(): it does something nasty to
3114          * Smart Array controllers that pci_enable_device does not undo
3115          */
3116         pci_release_regions(pdev);
3117         return err;
3118 }
3119
3120 /*
3121  * Gets information about the local volumes attached to the controller.
3122  */
3123 static void cciss_getgeometry(int cntl_num)
3124 {
3125         ReportLunData_struct *ld_buff;
3126         InquiryData_struct *inq_buff;
3127         int return_code;
3128         int i;
3129         int listlength = 0;
3130         __u32 lunid = 0;
3131         unsigned block_size;
3132         sector_t total_size;
3133
3134         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
3135         if (ld_buff == NULL) {
3136                 printk(KERN_ERR "cciss: out of memory\n");
3137                 return;
3138         }
3139         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
3140         if (inq_buff == NULL) {
3141                 printk(KERN_ERR "cciss: out of memory\n");
3142                 kfree(ld_buff);
3143                 return;
3144         }
3145         /* Get the firmware version */
3146         return_code = sendcmd(CISS_INQUIRY, cntl_num, inq_buff,
3147                               sizeof(InquiryData_struct), 0, 0, 0, NULL,
3148                               TYPE_CMD);
3149         if (return_code == IO_OK) {
3150                 hba[cntl_num]->firm_ver[0] = inq_buff->data_byte[32];
3151                 hba[cntl_num]->firm_ver[1] = inq_buff->data_byte[33];
3152                 hba[cntl_num]->firm_ver[2] = inq_buff->data_byte[34];
3153                 hba[cntl_num]->firm_ver[3] = inq_buff->data_byte[35];
3154         } else {                /* send command failed */
3155
3156                 printk(KERN_WARNING "cciss: unable to determine firmware"
3157                        " version of controller\n");
3158         }
3159         /* Get the number of logical volumes */
3160         return_code = sendcmd(CISS_REPORT_LOG, cntl_num, ld_buff,
3161                               sizeof(ReportLunData_struct), 0, 0, 0, NULL,
3162                               TYPE_CMD);
3163
3164         if (return_code == IO_OK) {
3165 #ifdef CCISS_DEBUG
3166                 printk("LUN Data\n--------------------------\n");
3167 #endif                          /* CCISS_DEBUG */
3168
3169                 listlength |=
3170                     (0xff & (unsigned int)(ld_buff->LUNListLength[0])) << 24;
3171                 listlength |=
3172                     (0xff & (unsigned int)(ld_buff->LUNListLength[1])) << 16;
3173                 listlength |=
3174                     (0xff & (unsigned int)(ld_buff->LUNListLength[2])) << 8;
3175                 listlength |= 0xff & (unsigned int)(ld_buff->LUNListLength[3]);
3176         } else {                /* reading number of logical volumes failed */
3177
3178                 printk(KERN_WARNING "cciss: report logical volume"
3179                        " command failed\n");
3180                 listlength = 0;
3181         }
3182         hba[cntl_num]->num_luns = listlength / 8;       // 8 bytes pre entry
3183         if (hba[cntl_num]->num_luns > CISS_MAX_LUN) {
3184                 printk(KERN_ERR
3185                        "ciss:  only %d number of logical volumes supported\n",
3186                        CISS_MAX_LUN);
3187                 hba[cntl_num]->num_luns = CISS_MAX_LUN;
3188         }
3189 #ifdef CCISS_DEBUG
3190         printk(KERN_DEBUG "Length = %x %x %x %x = %d\n",
3191                ld_buff->LUNListLength[0], ld_buff->LUNListLength[1],
3192                ld_buff->LUNListLength[2], ld_buff->LUNListLength[3],
3193                hba[cntl_num]->num_luns);
3194 #endif                          /* CCISS_DEBUG */
3195
3196         hba[cntl_num]->highest_lun = hba[cntl_num]->num_luns - 1;
3197         for (i = 0; i < CISS_MAX_LUN; i++) {
3198                 if (i < hba[cntl_num]->num_luns) {
3199                         lunid = (0xff & (unsigned int)(ld_buff->LUN[i][3]))
3200                             << 24;
3201                         lunid |= (0xff & (unsigned int)(ld_buff->LUN[i][2]))
3202                             << 16;
3203                         lunid |= (0xff & (unsigned int)(ld_buff->LUN[i][1]))
3204                             << 8;
3205                         lunid |= 0xff & (unsigned int)(ld_buff->LUN[i][0]);
3206
3207                         hba[cntl_num]->drv[i].LunID = lunid;
3208
3209 #ifdef CCISS_DEBUG
3210                         printk(KERN_DEBUG "LUN[%d]:  %x %x %x %x = %x\n", i,
3211                                ld_buff->LUN[i][0], ld_buff->LUN[i][1],
3212                                ld_buff->LUN[i][2], ld_buff->LUN[i][3],
3213                                hba[cntl_num]->drv[i].LunID);
3214 #endif                          /* CCISS_DEBUG */
3215
3216                 /* testing to see if 16-byte CDBs are already being used */
3217                 if(hba[cntl_num]->cciss_read == CCISS_READ_16) {
3218                         cciss_read_capacity_16(cntl_num, i, 0,
3219                                             &total_size, &block_size);
3220                         goto geo_inq;
3221                 }
3222                 cciss_read_capacity(cntl_num, i, 0, &total_size, &block_size);
3223
3224                 /* If read_capacity returns all F's the logical is >2TB */
3225                 /* so we switch to 16-byte CDBs for all read/write ops */
3226                 if(total_size == 0xFFFFFFFFULL) {
3227                         cciss_read_capacity_16(cntl_num, i, 0,
3228                         &total_size, &block_size);
3229                         hba[cntl_num]->cciss_read = CCISS_READ_16;
3230                         hba[cntl_num]->cciss_write = CCISS_WRITE_16;
3231                 } else {
3232                         hba[cntl_num]->cciss_read = CCISS_READ_10;
3233                         hba[cntl_num]->cciss_write = CCISS_WRITE_10;
3234                 }
3235 geo_inq:
3236                         cciss_geometry_inquiry(cntl_num, i, 0, total_size,
3237                                                block_size, inq_buff,
3238                                                &hba[cntl_num]->drv[i]);
3239                 } else {
3240                         /* initialize raid_level to indicate a free space */
3241                         hba[cntl_num]->drv[i].raid_level = -1;
3242                 }
3243         }
3244         kfree(ld_buff);
3245         kfree(inq_buff);
3246 }
3247
3248 /* Function to find the first free pointer into our hba[] array */
3249 /* Returns -1 if no free entries are left.  */
3250 static int alloc_cciss_hba(void)
3251 {
3252         int i;
3253
3254         for (i = 0; i < MAX_CTLR; i++) {
3255                 if (!hba[i]) {
3256                         ctlr_info_t *p;
3257
3258                         p = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
3259                         if (!p)
3260                                 goto Enomem;
3261                         p->gendisk[0] = alloc_disk(1 << NWD_SHIFT);
3262                         if (!p->gendisk[0]) {
3263                                 kfree(p);
3264                                 goto Enomem;
3265                         }
3266                         hba[i] = p;
3267                         return i;
3268                 }
3269         }
3270         printk(KERN_WARNING "cciss: This driver supports a maximum"
3271                " of %d controllers.\n", MAX_CTLR);
3272         return -1;
3273 Enomem:
3274         printk(KERN_ERR "cciss: out of memory.\n");
3275         return -1;
3276 }
3277
3278 static void free_hba(int i)
3279 {
3280         ctlr_info_t *p = hba[i];
3281         int n;
3282
3283         hba[i] = NULL;
3284         for (n = 0; n < CISS_MAX_LUN; n++)
3285                 put_disk(p->gendisk[n]);
3286         kfree(p);
3287 }
3288
3289 /*
3290  *  This is it.  Find all the controllers and register them.  I really hate
3291  *  stealing all these major device numbers.
3292  *  returns the number of block devices registered.
3293  */
3294 static int __devinit cciss_init_one(struct pci_dev *pdev,
3295                                     const struct pci_device_id *ent)
3296 {
3297         int i;
3298         int j = 0;
3299         int rc;
3300         int dac;
3301
3302         i = alloc_cciss_hba();
3303         if (i < 0)
3304                 return -1;
3305
3306         hba[i]->busy_initializing = 1;
3307
3308         if (cciss_pci_init(hba[i], pdev) != 0)
3309                 goto clean1;
3310
3311         sprintf(hba[i]->devname, "cciss%d", i);
3312         hba[i]->ctlr = i;
3313         hba[i]->pdev = pdev;
3314
3315         /* configure PCI DMA stuff */
3316         if (!pci_set_dma_mask(pdev, DMA_64BIT_MASK))
3317                 dac = 1;
3318         else if (!pci_set_dma_mask(pdev, DMA_32BIT_MASK))
3319                 dac = 0;
3320         else {
3321                 printk(KERN_ERR "cciss: no suitable DMA available\n");
3322                 goto clean1;
3323         }
3324
3325         /*
3326          * register with the major number, or get a dynamic major number
3327          * by passing 0 as argument.  This is done for greater than
3328          * 8 controller support.
3329          */
3330         if (i < MAX_CTLR_ORIG)
3331                 hba[i]->major = COMPAQ_CISS_MAJOR + i;
3332         rc = register_blkdev(hba[i]->major, hba[i]->devname);
3333         if (rc == -EBUSY || rc == -EINVAL) {
3334                 printk(KERN_ERR
3335                        "cciss:  Unable to get major number %d for %s "
3336                        "on hba %d\n", hba[i]->major, hba[i]->devname, i);
3337                 goto clean1;
3338         } else {
3339                 if (i >= MAX_CTLR_ORIG)
3340                         hba[i]->major = rc;
3341         }
3342
3343         /* make sure the board interrupts are off */
3344         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_OFF);
3345         if (request_irq(hba[i]->intr[SIMPLE_MODE_INT], do_cciss_intr,
3346                         IRQF_DISABLED | IRQF_SHARED, hba[i]->devname, hba[i])) {
3347                 printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
3348                        hba[i]->intr[SIMPLE_MODE_INT], hba[i]->devname);
3349                 goto clean2;
3350         }
3351
3352         printk(KERN_INFO "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
3353                hba[i]->devname, pdev->device, pci_name(pdev),
3354                hba[i]->intr[SIMPLE_MODE_INT], dac ? "" : " not");
3355
3356         hba[i]->cmd_pool_bits =
3357             kmalloc(((hba[i]->nr_cmds + BITS_PER_LONG -
3358                       1) / BITS_PER_LONG) * sizeof(unsigned long), GFP_KERNEL);
3359         hba[i]->cmd_pool = (CommandList_struct *)
3360             pci_alloc_consistent(hba[i]->pdev,
3361                     hba[i]->nr_cmds * sizeof(CommandList_struct),
3362                     &(hba[i]->cmd_pool_dhandle));
3363         hba[i]->errinfo_pool = (ErrorInfo_struct *)
3364             pci_alloc_consistent(hba[i]->pdev,
3365                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
3366                     &(hba[i]->errinfo_pool_dhandle));
3367         if ((hba[i]->cmd_pool_bits == NULL)
3368             || (hba[i]->cmd_pool == NULL)
3369             || (hba[i]->errinfo_pool == NULL)) {
3370                 printk(KERN_ERR "cciss: out of memory");
3371                 goto clean4;
3372         }
3373 #ifdef CONFIG_CISS_SCSI_TAPE
3374         hba[i]->scsi_rejects.complete =
3375             kmalloc(sizeof(hba[i]->scsi_rejects.complete[0]) *
3376                     (hba[i]->nr_cmds + 5), GFP_KERNEL);
3377         if (hba[i]->scsi_rejects.complete == NULL) {
3378                 printk(KERN_ERR "cciss: out of memory");
3379                 goto clean4;
3380         }
3381 #endif
3382         spin_lock_init(&hba[i]->lock);
3383
3384         /* Initialize the pdev driver private data.
3385            have it point to hba[i].  */
3386         pci_set_drvdata(pdev, hba[i]);
3387         /* command and error info recs zeroed out before
3388            they are used */
3389         memset(hba[i]->cmd_pool_bits, 0,
3390                ((hba[i]->nr_cmds + BITS_PER_LONG -
3391                  1) / BITS_PER_LONG) * sizeof(unsigned long));
3392
3393 #ifdef CCISS_DEBUG
3394         printk(KERN_DEBUG "Scanning for drives on controller cciss%d\n", i);
3395 #endif                          /* CCISS_DEBUG */
3396
3397         cciss_getgeometry(i);
3398
3399         cciss_scsi_setup(i);
3400
3401         /* Turn the interrupts on so we can service requests */
3402         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_ON);
3403
3404         cciss_procinit(i);
3405
3406         hba[i]->cciss_max_sectors = 2048;
3407
3408         hba[i]->busy_initializing = 0;
3409
3410         do {
3411                 drive_info_struct *drv = &(hba[i]->drv[j]);
3412                 struct gendisk *disk = hba[i]->gendisk[j];
3413                 struct request_queue *q;
3414
3415                 /* Check if the disk was allocated already */
3416                 if (!disk){
3417                         hba[i]->gendisk[j] = alloc_disk(1 << NWD_SHIFT);
3418                         disk = hba[i]->gendisk[j];
3419                 }
3420
3421                 /* Check that the disk was able to be allocated */
3422                 if (!disk) {
3423                         printk(KERN_ERR "cciss: unable to allocate memory for disk %d\n", j);
3424                         goto clean4;
3425                 }
3426
3427                 q = blk_init_queue(do_cciss_request, &hba[i]->lock);
3428                 if (!q) {
3429                         printk(KERN_ERR
3430                                "cciss:  unable to allocate queue for disk %d\n",
3431                                j);
3432                         goto clean4;
3433                 }
3434                 drv->queue = q;
3435
3436                 q->backing_dev_info.ra_pages = READ_AHEAD;
3437                 blk_queue_bounce_limit(q, hba[i]->pdev->dma_mask);
3438
3439                 /* This is a hardware imposed limit. */
3440                 blk_queue_max_hw_segments(q, MAXSGENTRIES);
3441
3442                 /* This is a limit in the driver and could be eliminated. */
3443                 blk_queue_max_phys_segments(q, MAXSGENTRIES);
3444
3445                 blk_queue_max_sectors(q, hba[i]->cciss_max_sectors);
3446
3447                 blk_queue_softirq_done(q, cciss_softirq_done);
3448
3449                 q->queuedata = hba[i];
3450                 sprintf(disk->disk_name, "cciss/c%dd%d", i, j);
3451                 disk->major = hba[i]->major;
3452                 disk->first_minor = j << NWD_SHIFT;
3453                 disk->fops = &cciss_fops;
3454                 disk->queue = q;
3455                 disk->private_data = drv;
3456                 disk->driverfs_dev = &pdev->dev;
3457                 /* we must register the controller even if no disks exist */
3458                 /* this is for the online array utilities */
3459                 if (!drv->heads && j)
3460                         continue;
3461                 blk_queue_hardsect_size(q, drv->block_size);
3462                 set_capacity(disk, drv->nr_blocks);
3463                 add_disk(disk);
3464                 j++;
3465         } while (j <= hba[i]->highest_lun);
3466
3467         return 1;
3468
3469       clean4:
3470 #ifdef CONFIG_CISS_SCSI_TAPE
3471         kfree(hba[i]->scsi_rejects.complete);
3472 #endif
3473         kfree(hba[i]->cmd_pool_bits);
3474         if (hba[i]->cmd_pool)
3475                 pci_free_consistent(hba[i]->pdev,
3476                                     hba[i]->nr_cmds * sizeof(CommandList_struct),
3477                                     hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
3478         if (hba[i]->errinfo_pool)
3479                 pci_free_consistent(hba[i]->pdev,
3480                                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
3481                                     hba[i]->errinfo_pool,
3482                                     hba[i]->errinfo_pool_dhandle);
3483         free_irq(hba[i]->intr[SIMPLE_MODE_INT], hba[i]);
3484       clean2:
3485         unregister_blkdev(hba[i]->major, hba[i]->devname);
3486       clean1:
3487         hba[i]->busy_initializing = 0;
3488         /* cleanup any queues that may have been initialized */
3489         for (j=0; j <= hba[i]->highest_lun; j++){
3490                 drive_info_struct *drv = &(hba[i]->drv[j]);
3491                 if (drv->queue)
3492                         blk_cleanup_queue(drv->queue);
3493         }
3494         /*
3495          * Deliberately omit pci_disable_device(): it does something nasty to
3496          * Smart Array controllers that pci_enable_device does not undo
3497          */
3498         pci_release_regions(pdev);
3499         pci_set_drvdata(pdev, NULL);
3500         free_hba(i);
3501         return -1;
3502 }
3503
3504 static void cciss_shutdown(struct pci_dev *pdev)
3505 {
3506         ctlr_info_t *tmp_ptr;
3507         int i;
3508         char flush_buf[4];
3509         int return_code;
3510
3511         tmp_ptr = pci_get_drvdata(pdev);
3512         if (tmp_ptr == NULL)
3513                 return;
3514         i = tmp_ptr->ctlr;
3515         if (hba[i] == NULL)
3516                 return;
3517
3518         /* Turn board interrupts off  and send the flush cache command */
3519         /* sendcmd will turn off interrupt, and send the flush...
3520          * To write all data in the battery backed cache to disks */
3521         memset(flush_buf, 0, 4);
3522         return_code = sendcmd(CCISS_CACHE_FLUSH, i, flush_buf, 4, 0, 0, 0, NULL,
3523                               TYPE_CMD);
3524         if (return_code == IO_OK) {
3525                 printk(KERN_INFO "Completed flushing cache on controller %d\n", i);
3526         } else {
3527                 printk(KERN_WARNING "Error flushing cache on controller %d\n", i);
3528         }
3529         free_irq(hba[i]->intr[2], hba[i]);
3530 }
3531
3532 static void __devexit cciss_remove_one(struct pci_dev *pdev)
3533 {
3534         ctlr_info_t *tmp_ptr;
3535         int i, j;
3536
3537         if (pci_get_drvdata(pdev) == NULL) {
3538                 printk(KERN_ERR "cciss: Unable to remove device \n");
3539                 return;
3540         }
3541         tmp_ptr = pci_get_drvdata(pdev);
3542         i = tmp_ptr->ctlr;
3543         if (hba[i] == NULL) {
3544                 printk(KERN_ERR "cciss: device appears to "
3545                        "already be removed \n");
3546                 return;
3547         }
3548
3549         remove_proc_entry(hba[i]->devname, proc_cciss);
3550         unregister_blkdev(hba[i]->major, hba[i]->devname);
3551
3552         /* remove it from the disk list */
3553         for (j = 0; j < CISS_MAX_LUN; j++) {
3554                 struct gendisk *disk = hba[i]->gendisk[j];
3555                 if (disk) {
3556                         struct request_queue *q = disk->queue;
3557
3558                         if (disk->flags & GENHD_FL_UP)
3559                                 del_gendisk(disk);
3560                         if (q)
3561                                 blk_cleanup_queue(q);
3562                 }
3563         }
3564
3565         cciss_unregister_scsi(i);       /* unhook from SCSI subsystem */
3566
3567         cciss_shutdown(pdev);
3568
3569 #ifdef CONFIG_PCI_MSI
3570         if (hba[i]->msix_vector)
3571                 pci_disable_msix(hba[i]->pdev);
3572         else if (hba[i]->msi_vector)
3573                 pci_disable_msi(hba[i]->pdev);
3574 #endif                          /* CONFIG_PCI_MSI */
3575
3576         iounmap(hba[i]->vaddr);
3577
3578         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(CommandList_struct),
3579                             hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
3580         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
3581                             hba[i]->errinfo_pool, hba[i]->errinfo_pool_dhandle);
3582         kfree(hba[i]->cmd_pool_bits);
3583 #ifdef CONFIG_CISS_SCSI_TAPE
3584         kfree(hba[i]->scsi_rejects.complete);
3585 #endif
3586         /*
3587          * Deliberately omit pci_disable_device(): it does something nasty to
3588          * Smart Array controllers that pci_enable_device does not undo
3589          */
3590         pci_release_regions(pdev);
3591         pci_set_drvdata(pdev, NULL);
3592         free_hba(i);
3593 }
3594
3595 static struct pci_driver cciss_pci_driver = {
3596         .name = "cciss",
3597         .probe = cciss_init_one,
3598         .remove = __devexit_p(cciss_remove_one),
3599         .id_table = cciss_pci_device_id,        /* id_table */
3600         .shutdown = cciss_shutdown,
3601 };
3602
3603 /*
3604  *  This is it.  Register the PCI driver information for the cards we control
3605  *  the OS will call our registered routines when it finds one of our cards.
3606  */
3607 static int __init cciss_init(void)
3608 {
3609         printk(KERN_INFO DRIVER_NAME "\n");
3610
3611         /* Register for our PCI devices */
3612         return pci_register_driver(&cciss_pci_driver);
3613 }
3614
3615 static void __exit cciss_cleanup(void)
3616 {
3617         int i;
3618
3619         pci_unregister_driver(&cciss_pci_driver);
3620         /* double check that all controller entrys have been removed */
3621         for (i = 0; i < MAX_CTLR; i++) {
3622                 if (hba[i] != NULL) {
3623                         printk(KERN_WARNING "cciss: had to remove"
3624                                " controller %d\n", i);
3625                         cciss_remove_one(hba[i]->pdev);
3626                 }
3627         }
3628         remove_proc_entry("cciss", proc_root_driver);
3629 }
3630
3631 static void fail_all_cmds(unsigned long ctlr)
3632 {
3633         /* If we get here, the board is apparently dead. */
3634         ctlr_info_t *h = hba[ctlr];
3635         CommandList_struct *c;
3636         unsigned long flags;
3637
3638         printk(KERN_WARNING "cciss%d: controller not responding.\n", h->ctlr);
3639         h->alive = 0;           /* the controller apparently died... */
3640
3641         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
3642
3643         pci_disable_device(h->pdev);    /* Make sure it is really dead. */
3644
3645         /* move everything off the request queue onto the completed queue */
3646         while ((c = h->reqQ) != NULL) {
3647                 removeQ(&(h->reqQ), c);
3648                 h->Qdepth--;
3649                 addQ(&(h->cmpQ), c);
3650         }
3651
3652         /* Now, fail everything on the completed queue with a HW error */
3653         while ((c = h->cmpQ) != NULL) {
3654                 removeQ(&h->cmpQ, c);
3655                 c->err_info->CommandStatus = CMD_HARDWARE_ERR;
3656                 if (c->cmd_type == CMD_RWREQ) {
3657                         complete_command(h, c, 0);
3658                 } else if (c->cmd_type == CMD_IOCTL_PEND)
3659                         complete(c->waiting);
3660 #ifdef CONFIG_CISS_SCSI_TAPE
3661                 else if (c->cmd_type == CMD_SCSI)
3662                         complete_scsi_command(c, 0, 0);
3663 #endif
3664         }
3665         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
3666         return;
3667 }
3668
3669 module_init(cciss_init);
3670 module_exit(cciss_cleanup);