?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
sfc: Always close net device at the end of a disabling reset
[linux-2.6/cjktty.git] / drivers / block / cciss.c
blobeb5ff0531cfb33cf41df0a089bf8eb7a0119099f
1 /*
2 * Disk Array driver for HP Smart Array controllers.
3 * (C) Copyright 2000, 2007 Hewlett-Packard Development Company, L.P.
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; version 2 of the License.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12 * General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
17 * 02111-1307, USA.
18 *
19 * Questions/Comments/Bugfixes to iss_storagedev@hp.com
20 *
21 */
22
23 #include <linux/module.h>
24 #include <linux/interrupt.h>
25 #include <linux/types.h>
26 #include <linux/pci.h>
27 #include <linux/kernel.h>
28 #include <linux/slab.h>
29 #include <linux/smp_lock.h>
30 #include <linux/delay.h>
31 #include <linux/major.h>
32 #include <linux/fs.h>
33 #include <linux/bio.h>
34 #include <linux/blkpg.h>
35 #include <linux/timer.h>
36 #include <linux/proc_fs.h>
37 #include <linux/seq_file.h>
38 #include <linux/init.h>
39 #include <linux/jiffies.h>
40 #include <linux/hdreg.h>
41 #include <linux/spinlock.h>
42 #include <linux/compat.h>
43 #include <linux/mutex.h>
44 #include <asm/uaccess.h>
45 #include <asm/io.h>
46
47 #include <linux/dma-mapping.h>
48 #include <linux/blkdev.h>
49 #include <linux/genhd.h>
50 #include <linux/completion.h>
51 #include <scsi/scsi.h>
52 #include <scsi/sg.h>
53 #include <scsi/scsi_ioctl.h>
54 #include <linux/cdrom.h>
55 #include <linux/scatterlist.h>
56 #include <linux/kthread.h>
57
58 #define CCISS_DRIVER_VERSION(maj,min,submin) ((maj<<16)|(min<<8)|(submin))
59 #define DRIVER_NAME "HP CISS Driver (v 3.6.20)"
60 #define DRIVER_VERSION CCISS_DRIVER_VERSION(3, 6, 20)
61
62 /* Embedded module documentation macros - see modules.h */
63 MODULE_AUTHOR("Hewlett-Packard Company");
64 MODULE_DESCRIPTION("Driver for HP Smart Array Controllers");
65 MODULE_SUPPORTED_DEVICE("HP SA5i SA5i+ SA532 SA5300 SA5312 SA641 SA642 SA6400"
66 " SA6i P600 P800 P400 P400i E200 E200i E500 P700m"
67 " Smart Array G2 Series SAS/SATA Controllers");
68 MODULE_VERSION("3.6.20");
69 MODULE_LICENSE("GPL");
70
71 static int cciss_allow_hpsa;
72 module_param(cciss_allow_hpsa, int, S_IRUGO|S_IWUSR);
73 MODULE_PARM_DESC(cciss_allow_hpsa,
74 "Prevent cciss driver from accessing hardware known to be "
75 " supported by the hpsa driver");
76
77 #include "cciss_cmd.h"
78 #include "cciss.h"
79 #include <linux/cciss_ioctl.h>
80
81 /* define the PCI info for the cards we can control */
82 static const struct pci_device_id cciss_pci_device_id[] = {
83 {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISS, 0x0E11, 0x4070},
84 {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4080},
85 {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4082},
86 {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4083},
87 {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x4091},
88 {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409A},
89 {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409B},
90 {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409C},
91 {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409D},
92 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSA, 0x103C, 0x3225},
93 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSC, 0x103C, 0x3223},
94 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSC, 0x103C, 0x3234},
95 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSC, 0x103C, 0x3235},
96 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSD, 0x103C, 0x3211},
97 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSD, 0x103C, 0x3212},
98 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSD, 0x103C, 0x3213},
99 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSD, 0x103C, 0x3214},
100 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSD, 0x103C, 0x3215},
101 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSC, 0x103C, 0x3237},
102 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSC, 0x103C, 0x323D},
103 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSE, 0x103C, 0x3241},
104 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSE, 0x103C, 0x3243},
105 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSE, 0x103C, 0x3245},
106 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSE, 0x103C, 0x3247},
107 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSE, 0x103C, 0x3249},
108 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSE, 0x103C, 0x324A},
109 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSE, 0x103C, 0x324B},
110 {0,}
111 };
112
113 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
114
115 /* board_id = Subsystem Device ID & Vendor ID
116 * product = Marketing Name for the board
117 * access = Address of the struct of function pointers
118 */
119 static struct board_type products[] = {
120 {0x40700E11, "Smart Array 5300", &SA5_access},
121 {0x40800E11, "Smart Array 5i", &SA5B_access},
122 {0x40820E11, "Smart Array 532", &SA5B_access},
123 {0x40830E11, "Smart Array 5312", &SA5B_access},
124 {0x409A0E11, "Smart Array 641", &SA5_access},
125 {0x409B0E11, "Smart Array 642", &SA5_access},
126 {0x409C0E11, "Smart Array 6400", &SA5_access},
127 {0x409D0E11, "Smart Array 6400 EM", &SA5_access},
128 {0x40910E11, "Smart Array 6i", &SA5_access},
129 {0x3225103C, "Smart Array P600", &SA5_access},
130 {0x3235103C, "Smart Array P400i", &SA5_access},
131 {0x3211103C, "Smart Array E200i", &SA5_access},
132 {0x3212103C, "Smart Array E200", &SA5_access},
133 {0x3213103C, "Smart Array E200i", &SA5_access},
134 {0x3214103C, "Smart Array E200i", &SA5_access},
135 {0x3215103C, "Smart Array E200i", &SA5_access},
136 {0x3237103C, "Smart Array E500", &SA5_access},
137 /* controllers below this line are also supported by the hpsa driver. */
138 #define HPSA_BOUNDARY 0x3223103C
139 {0x3223103C, "Smart Array P800", &SA5_access},
140 {0x3234103C, "Smart Array P400", &SA5_access},
141 {0x323D103C, "Smart Array P700m", &SA5_access},
142 {0x3241103C, "Smart Array P212", &SA5_access},
143 {0x3243103C, "Smart Array P410", &SA5_access},
144 {0x3245103C, "Smart Array P410i", &SA5_access},
145 {0x3247103C, "Smart Array P411", &SA5_access},
146 {0x3249103C, "Smart Array P812", &SA5_access},
147 {0x324A103C, "Smart Array P712m", &SA5_access},
148 {0x324B103C, "Smart Array P711m", &SA5_access},
149 };
150
151 /* How long to wait (in milliseconds) for board to go into simple mode */
152 #define MAX_CONFIG_WAIT 30000
153 #define MAX_IOCTL_CONFIG_WAIT 1000
154
155 /*define how many times we will try a command because of bus resets */
156 #define MAX_CMD_RETRIES 3
157
158 #define MAX_CTLR 32
159
160 /* Originally cciss driver only supports 8 major numbers */
161 #define MAX_CTLR_ORIG 8
162
163 static ctlr_info_t *hba[MAX_CTLR];
164
165 static struct task_struct *cciss_scan_thread;
166 static DEFINE_MUTEX(scan_mutex);
167 static LIST_HEAD(scan_q);
168
169 static void do_cciss_request(struct request_queue *q);
170 static irqreturn_t do_cciss_intr(int irq, void *dev_id);
171 static int cciss_open(struct block_device *bdev, fmode_t mode);
172 static int cciss_release(struct gendisk *disk, fmode_t mode);
173 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
174 unsigned int cmd, unsigned long arg);
175 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
176
177 static int cciss_revalidate(struct gendisk *disk);
178 static int rebuild_lun_table(ctlr_info_t *h, int first_time, int via_ioctl);
179 static int deregister_disk(ctlr_info_t *h, int drv_index,
180 int clear_all, int via_ioctl);
181
182 static void cciss_read_capacity(int ctlr, int logvol,
183 sector_t *total_size, unsigned int *block_size);
184 static void cciss_read_capacity_16(int ctlr, int logvol,
185 sector_t *total_size, unsigned int *block_size);
186 static void cciss_geometry_inquiry(int ctlr, int logvol,
187 sector_t total_size,
188 unsigned int block_size, InquiryData_struct *inq_buff,
189 drive_info_struct *drv);
190 static void __devinit cciss_interrupt_mode(ctlr_info_t *, struct pci_dev *,
191 __u32);
192 static void start_io(ctlr_info_t *h);
193 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
194 __u8 page_code, unsigned char scsi3addr[],
195 int cmd_type);
196 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
197 int attempt_retry);
198 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c);
199
200 static void fail_all_cmds(unsigned long ctlr);
201 static int add_to_scan_list(struct ctlr_info *h);
202 static int scan_thread(void *data);
203 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c);
204 static void cciss_hba_release(struct device *dev);
205 static void cciss_device_release(struct device *dev);
206 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index);
207 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index);
208
209 #ifdef CONFIG_PROC_FS
210 static void cciss_procinit(int i);
211 #else
212 static void cciss_procinit(int i)
213 {
214 }
215 #endif /* CONFIG_PROC_FS */
216
217 #ifdef CONFIG_COMPAT
218 static int cciss_compat_ioctl(struct block_device *, fmode_t,
219 unsigned, unsigned long);
220 #endif
221
222 static const struct block_device_operations cciss_fops = {
223 .owner = THIS_MODULE,
224 .open = cciss_open,
225 .release = cciss_release,
226 .locked_ioctl = cciss_ioctl,
227 .getgeo = cciss_getgeo,
228 #ifdef CONFIG_COMPAT
229 .compat_ioctl = cciss_compat_ioctl,
230 #endif
231 .revalidate_disk = cciss_revalidate,
232 };
233
234 /*
235 * Enqueuing and dequeuing functions for cmdlists.
236 */
237 static inline void addQ(struct hlist_head *list, CommandList_struct *c)
238 {
239 hlist_add_head(&c->list, list);
240 }
241
242 static inline void removeQ(CommandList_struct *c)
243 {
244 /*
245 * After kexec/dump some commands might still
246 * be in flight, which the firmware will try
247 * to complete. Resetting the firmware doesn't work
248 * with old fw revisions, so we have to mark
249 * them off as 'stale' to prevent the driver from
250 * falling over.
251 */
252 if (WARN_ON(hlist_unhashed(&c->list))) {
253 c->cmd_type = CMD_MSG_STALE;
254 return;
255 }
256
257 hlist_del_init(&c->list);
258 }
259
260 static void cciss_free_sg_chain_blocks(SGDescriptor_struct **cmd_sg_list,
261 int nr_cmds)
262 {
263 int i;
264
265 if (!cmd_sg_list)
266 return;
267 for (i = 0; i < nr_cmds; i++) {
268 kfree(cmd_sg_list[i]);
269 cmd_sg_list[i] = NULL;
270 }
271 kfree(cmd_sg_list);
272 }
273
274 static SGDescriptor_struct **cciss_allocate_sg_chain_blocks(
275 ctlr_info_t *h, int chainsize, int nr_cmds)
276 {
277 int j;
278 SGDescriptor_struct **cmd_sg_list;
279
280 if (chainsize <= 0)
281 return NULL;
282
283 cmd_sg_list = kmalloc(sizeof(*cmd_sg_list) * nr_cmds, GFP_KERNEL);
284 if (!cmd_sg_list)
285 return NULL;
286
287 /* Build up chain blocks for each command */
288 for (j = 0; j < nr_cmds; j++) {
289 /* Need a block of chainsized s/g elements. */
290 cmd_sg_list[j] = kmalloc((chainsize *
291 sizeof(*cmd_sg_list[j])), GFP_KERNEL);
292 if (!cmd_sg_list[j]) {
293 dev_err(&h->pdev->dev, "Cannot get memory "
294 "for s/g chains.\n");
295 goto clean;
296 }
297 }
298 return cmd_sg_list;
299 clean:
300 cciss_free_sg_chain_blocks(cmd_sg_list, nr_cmds);
301 return NULL;
302 }
303
304 static void cciss_unmap_sg_chain_block(ctlr_info_t *h, CommandList_struct *c)
305 {
306 SGDescriptor_struct *chain_sg;
307 u64bit temp64;
308
309 if (c->Header.SGTotal <= h->max_cmd_sgentries)
310 return;
311
312 chain_sg = &c->SG[h->max_cmd_sgentries - 1];
313 temp64.val32.lower = chain_sg->Addr.lower;
314 temp64.val32.upper = chain_sg->Addr.upper;
315 pci_unmap_single(h->pdev, temp64.val, chain_sg->Len, PCI_DMA_TODEVICE);
316 }
317
318 static void cciss_map_sg_chain_block(ctlr_info_t *h, CommandList_struct *c,
319 SGDescriptor_struct *chain_block, int len)
320 {
321 SGDescriptor_struct *chain_sg;
322 u64bit temp64;
323
324 chain_sg = &c->SG[h->max_cmd_sgentries - 1];
325 chain_sg->Ext = CCISS_SG_CHAIN;
326 chain_sg->Len = len;
327 temp64.val = pci_map_single(h->pdev, chain_block, len,
328 PCI_DMA_TODEVICE);
329 chain_sg->Addr.lower = temp64.val32.lower;
330 chain_sg->Addr.upper = temp64.val32.upper;
331 }
332
333 #include "cciss_scsi.c" /* For SCSI tape support */
334
335 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
336 "UNKNOWN"
337 };
338 #define RAID_UNKNOWN (sizeof(raid_label) / sizeof(raid_label[0])-1)
339
340 #ifdef CONFIG_PROC_FS
341
342 /*
343 * Report information about this controller.
344 */
345 #define ENG_GIG 1000000000
346 #define ENG_GIG_FACTOR (ENG_GIG/512)
347 #define ENGAGE_SCSI "engage scsi"
348
349 static struct proc_dir_entry *proc_cciss;
350
351 static void cciss_seq_show_header(struct seq_file *seq)
352 {
353 ctlr_info_t *h = seq->private;
354
355 seq_printf(seq, "%s: HP %s Controller\n"
356 "Board ID: 0x%08lx\n"
357 "Firmware Version: %c%c%c%c\n"
358 "IRQ: %d\n"
359 "Logical drives: %d\n"
360 "Current Q depth: %d\n"
361 "Current # commands on controller: %d\n"
362 "Max Q depth since init: %d\n"
363 "Max # commands on controller since init: %d\n"
364 "Max SG entries since init: %d\n",
365 h->devname,
366 h->product_name,
367 (unsigned long)h->board_id,
368 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
369 h->firm_ver[3], (unsigned int)h->intr[SIMPLE_MODE_INT],
370 h->num_luns,
371 h->Qdepth, h->commands_outstanding,
372 h->maxQsinceinit, h->max_outstanding, h->maxSG);
373
374 #ifdef CONFIG_CISS_SCSI_TAPE
375 cciss_seq_tape_report(seq, h->ctlr);
376 #endif /* CONFIG_CISS_SCSI_TAPE */
377 }
378
379 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
380 {
381 ctlr_info_t *h = seq->private;
382 unsigned ctlr = h->ctlr;
383 unsigned long flags;
384
385 /* prevent displaying bogus info during configuration
386 * or deconfiguration of a logical volume
387 */
388 spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
389 if (h->busy_configuring) {
390 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
391 return ERR_PTR(-EBUSY);
392 }
393 h->busy_configuring = 1;
394 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
395
396 if (*pos == 0)
397 cciss_seq_show_header(seq);
398
399 return pos;
400 }
401
402 static int cciss_seq_show(struct seq_file *seq, void *v)
403 {
404 sector_t vol_sz, vol_sz_frac;
405 ctlr_info_t *h = seq->private;
406 unsigned ctlr = h->ctlr;
407 loff_t *pos = v;
408 drive_info_struct *drv = h->drv[*pos];
409
410 if (*pos > h->highest_lun)
411 return 0;
412
413 if (drv == NULL) /* it's possible for h->drv[] to have holes. */
414 return 0;
415
416 if (drv->heads == 0)
417 return 0;
418
419 vol_sz = drv->nr_blocks;
420 vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
421 vol_sz_frac *= 100;
422 sector_div(vol_sz_frac, ENG_GIG_FACTOR);
423
424 if (drv->raid_level < 0 || drv->raid_level > RAID_UNKNOWN)
425 drv->raid_level = RAID_UNKNOWN;
426 seq_printf(seq, "cciss/c%dd%d:"
427 "\t%4u.%02uGB\tRAID %s\n",
428 ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
429 raid_label[drv->raid_level]);
430 return 0;
431 }
432
433 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
434 {
435 ctlr_info_t *h = seq->private;
436
437 if (*pos > h->highest_lun)
438 return NULL;
439 *pos += 1;
440
441 return pos;
442 }
443
444 static void cciss_seq_stop(struct seq_file *seq, void *v)
445 {
446 ctlr_info_t *h = seq->private;
447
448 /* Only reset h->busy_configuring if we succeeded in setting
449 * it during cciss_seq_start. */
450 if (v == ERR_PTR(-EBUSY))
451 return;
452
453 h->busy_configuring = 0;
454 }
455
456 static const struct seq_operations cciss_seq_ops = {
457 .start = cciss_seq_start,
458 .show = cciss_seq_show,
459 .next = cciss_seq_next,
460 .stop = cciss_seq_stop,
461 };
462
463 static int cciss_seq_open(struct inode *inode, struct file *file)
464 {
465 int ret = seq_open(file, &cciss_seq_ops);
466 struct seq_file *seq = file->private_data;
467
468 if (!ret)
469 seq->private = PDE(inode)->data;
470
471 return ret;
472 }
473
474 static ssize_t
475 cciss_proc_write(struct file *file, const char __user *buf,
476 size_t length, loff_t *ppos)
477 {
478 int err;
479 char *buffer;
480
481 #ifndef CONFIG_CISS_SCSI_TAPE
482 return -EINVAL;
483 #endif
484
485 if (!buf || length > PAGE_SIZE - 1)
486 return -EINVAL;
487
488 buffer = (char *)__get_free_page(GFP_KERNEL);
489 if (!buffer)
490 return -ENOMEM;
491
492 err = -EFAULT;
493 if (copy_from_user(buffer, buf, length))
494 goto out;
495 buffer[length] = '\0';
496
497 #ifdef CONFIG_CISS_SCSI_TAPE
498 if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
499 struct seq_file *seq = file->private_data;
500 ctlr_info_t *h = seq->private;
501
502 err = cciss_engage_scsi(h->ctlr);
503 if (err == 0)
504 err = length;
505 } else
506 #endif /* CONFIG_CISS_SCSI_TAPE */
507 err = -EINVAL;
508 /* might be nice to have "disengage" too, but it's not
509 safely possible. (only 1 module use count, lock issues.) */
510
511 out:
512 free_page((unsigned long)buffer);
513 return err;
514 }
515
516 static const struct file_operations cciss_proc_fops = {
517 .owner = THIS_MODULE,
518 .open = cciss_seq_open,
519 .read = seq_read,
520 .llseek = seq_lseek,
521 .release = seq_release,
522 .write = cciss_proc_write,
523 };
524
525 static void __devinit cciss_procinit(int i)
526 {
527 struct proc_dir_entry *pde;
528
529 if (proc_cciss == NULL)
530 proc_cciss = proc_mkdir("driver/cciss", NULL);
531 if (!proc_cciss)
532 return;
533 pde = proc_create_data(hba[i]->devname, S_IWUSR | S_IRUSR | S_IRGRP |
534 S_IROTH, proc_cciss,
535 &cciss_proc_fops, hba[i]);
536 }
537 #endif /* CONFIG_PROC_FS */
538
539 #define MAX_PRODUCT_NAME_LEN 19
540
541 #define to_hba(n) container_of(n, struct ctlr_info, dev)
542 #define to_drv(n) container_of(n, drive_info_struct, dev)
543
544 static ssize_t host_store_rescan(struct device *dev,
545 struct device_attribute *attr,
546 const char *buf, size_t count)
547 {
548 struct ctlr_info *h = to_hba(dev);
549
550 add_to_scan_list(h);
551 wake_up_process(cciss_scan_thread);
552 wait_for_completion_interruptible(&h->scan_wait);
553
554 return count;
555 }
556 static DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
557
558 static ssize_t dev_show_unique_id(struct device *dev,
559 struct device_attribute *attr,
560 char *buf)
561 {
562 drive_info_struct *drv = to_drv(dev);
563 struct ctlr_info *h = to_hba(drv->dev.parent);
564 __u8 sn[16];
565 unsigned long flags;
566 int ret = 0;
567
568 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
569 if (h->busy_configuring)
570 ret = -EBUSY;
571 else
572 memcpy(sn, drv->serial_no, sizeof(sn));
573 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
574
575 if (ret)
576 return ret;
577 else
578 return snprintf(buf, 16 * 2 + 2,
579 "%02X%02X%02X%02X%02X%02X%02X%02X"
580 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
581 sn[0], sn[1], sn[2], sn[3],
582 sn[4], sn[5], sn[6], sn[7],
583 sn[8], sn[9], sn[10], sn[11],
584 sn[12], sn[13], sn[14], sn[15]);
585 }
586 static DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
587
588 static ssize_t dev_show_vendor(struct device *dev,
589 struct device_attribute *attr,
590 char *buf)
591 {
592 drive_info_struct *drv = to_drv(dev);
593 struct ctlr_info *h = to_hba(drv->dev.parent);
594 char vendor[VENDOR_LEN + 1];
595 unsigned long flags;
596 int ret = 0;
597
598 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
599 if (h->busy_configuring)
600 ret = -EBUSY;
601 else
602 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
603 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
604
605 if (ret)
606 return ret;
607 else
608 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
609 }
610 static DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
611
612 static ssize_t dev_show_model(struct device *dev,
613 struct device_attribute *attr,
614 char *buf)
615 {
616 drive_info_struct *drv = to_drv(dev);
617 struct ctlr_info *h = to_hba(drv->dev.parent);
618 char model[MODEL_LEN + 1];
619 unsigned long flags;
620 int ret = 0;
621
622 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
623 if (h->busy_configuring)
624 ret = -EBUSY;
625 else
626 memcpy(model, drv->model, MODEL_LEN + 1);
627 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
628
629 if (ret)
630 return ret;
631 else
632 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
633 }
634 static DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
635
636 static ssize_t dev_show_rev(struct device *dev,
637 struct device_attribute *attr,
638 char *buf)
639 {
640 drive_info_struct *drv = to_drv(dev);
641 struct ctlr_info *h = to_hba(drv->dev.parent);
642 char rev[REV_LEN + 1];
643 unsigned long flags;
644 int ret = 0;
645
646 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
647 if (h->busy_configuring)
648 ret = -EBUSY;
649 else
650 memcpy(rev, drv->rev, REV_LEN + 1);
651 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
652
653 if (ret)
654 return ret;
655 else
656 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
657 }
658 static DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
659
660 static ssize_t cciss_show_lunid(struct device *dev,
661 struct device_attribute *attr, char *buf)
662 {
663 drive_info_struct *drv = to_drv(dev);
664 struct ctlr_info *h = to_hba(drv->dev.parent);
665 unsigned long flags;
666 unsigned char lunid[8];
667
668 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
669 if (h->busy_configuring) {
670 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
671 return -EBUSY;
672 }
673 if (!drv->heads) {
674 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
675 return -ENOTTY;
676 }
677 memcpy(lunid, drv->LunID, sizeof(lunid));
678 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
679 return snprintf(buf, 20, "0x%02x%02x%02x%02x%02x%02x%02x%02x\n",
680 lunid[0], lunid[1], lunid[2], lunid[3],
681 lunid[4], lunid[5], lunid[6], lunid[7]);
682 }
683 static DEVICE_ATTR(lunid, S_IRUGO, cciss_show_lunid, NULL);
684
685 static ssize_t cciss_show_raid_level(struct device *dev,
686 struct device_attribute *attr, char *buf)
687 {
688 drive_info_struct *drv = to_drv(dev);
689 struct ctlr_info *h = to_hba(drv->dev.parent);
690 int raid;
691 unsigned long flags;
692
693 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
694 if (h->busy_configuring) {
695 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
696 return -EBUSY;
697 }
698 raid = drv->raid_level;
699 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
700 if (raid < 0 || raid > RAID_UNKNOWN)
701 raid = RAID_UNKNOWN;
702
703 return snprintf(buf, strlen(raid_label[raid]) + 7, "RAID %s\n",
704 raid_label[raid]);
705 }
706 static DEVICE_ATTR(raid_level, S_IRUGO, cciss_show_raid_level, NULL);
707
708 static ssize_t cciss_show_usage_count(struct device *dev,
709 struct device_attribute *attr, char *buf)
710 {
711 drive_info_struct *drv = to_drv(dev);
712 struct ctlr_info *h = to_hba(drv->dev.parent);
713 unsigned long flags;
714 int count;
715
716 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
717 if (h->busy_configuring) {
718 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
719 return -EBUSY;
720 }
721 count = drv->usage_count;
722 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
723 return snprintf(buf, 20, "%d\n", count);
724 }
725 static DEVICE_ATTR(usage_count, S_IRUGO, cciss_show_usage_count, NULL);
726
727 static struct attribute *cciss_host_attrs[] = {
728 &dev_attr_rescan.attr,
729 NULL
730 };
731
732 static struct attribute_group cciss_host_attr_group = {
733 .attrs = cciss_host_attrs,
734 };
735
736 static const struct attribute_group *cciss_host_attr_groups[] = {
737 &cciss_host_attr_group,
738 NULL
739 };
740
741 static struct device_type cciss_host_type = {
742 .name = "cciss_host",
743 .groups = cciss_host_attr_groups,
744 .release = cciss_hba_release,
745 };
746
747 static struct attribute *cciss_dev_attrs[] = {
748 &dev_attr_unique_id.attr,
749 &dev_attr_model.attr,
750 &dev_attr_vendor.attr,
751 &dev_attr_rev.attr,
752 &dev_attr_lunid.attr,
753 &dev_attr_raid_level.attr,
754 &dev_attr_usage_count.attr,
755 NULL
756 };
757
758 static struct attribute_group cciss_dev_attr_group = {
759 .attrs = cciss_dev_attrs,
760 };
761
762 static const struct attribute_group *cciss_dev_attr_groups[] = {
763 &cciss_dev_attr_group,
764 NULL
765 };
766
767 static struct device_type cciss_dev_type = {
768 .name = "cciss_device",
769 .groups = cciss_dev_attr_groups,
770 .release = cciss_device_release,
771 };
772
773 static struct bus_type cciss_bus_type = {
774 .name = "cciss",
775 };
776
777 /*
778 * cciss_hba_release is called when the reference count
779 * of h->dev goes to zero.
780 */
781 static void cciss_hba_release(struct device *dev)
782 {
783 /*
784 * nothing to do, but need this to avoid a warning
785 * about not having a release handler from lib/kref.c.
786 */
787 }
788
789 /*
790 * Initialize sysfs entry for each controller. This sets up and registers
791 * the 'cciss#' directory for each individual controller under
792 * /sys/bus/pci/devices/<dev>/.
793 */
794 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
795 {
796 device_initialize(&h->dev);
797 h->dev.type = &cciss_host_type;
798 h->dev.bus = &cciss_bus_type;
799 dev_set_name(&h->dev, "%s", h->devname);
800 h->dev.parent = &h->pdev->dev;
801
802 return device_add(&h->dev);
803 }
804
805 /*
806 * Remove sysfs entries for an hba.
807 */
808 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
809 {
810 device_del(&h->dev);
811 put_device(&h->dev); /* final put. */
812 }
813
814 /* cciss_device_release is called when the reference count
815 * of h->drv[x]dev goes to zero.
816 */
817 static void cciss_device_release(struct device *dev)
818 {
819 drive_info_struct *drv = to_drv(dev);
820 kfree(drv);
821 }
822
823 /*
824 * Initialize sysfs for each logical drive. This sets up and registers
825 * the 'c#d#' directory for each individual logical drive under
826 * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
827 * /sys/block/cciss!c#d# to this entry.
828 */
829 static long cciss_create_ld_sysfs_entry(struct ctlr_info *h,
830 int drv_index)
831 {
832 struct device *dev;
833
834 if (h->drv[drv_index]->device_initialized)
835 return 0;
836
837 dev = &h->drv[drv_index]->dev;
838 device_initialize(dev);
839 dev->type = &cciss_dev_type;
840 dev->bus = &cciss_bus_type;
841 dev_set_name(dev, "c%dd%d", h->ctlr, drv_index);
842 dev->parent = &h->dev;
843 h->drv[drv_index]->device_initialized = 1;
844 return device_add(dev);
845 }
846
847 /*
848 * Remove sysfs entries for a logical drive.
849 */
850 static void cciss_destroy_ld_sysfs_entry(struct ctlr_info *h, int drv_index,
851 int ctlr_exiting)
852 {
853 struct device *dev = &h->drv[drv_index]->dev;
854
855 /* special case for c*d0, we only destroy it on controller exit */
856 if (drv_index == 0 && !ctlr_exiting)
857 return;
858
859 device_del(dev);
860 put_device(dev); /* the "final" put. */
861 h->drv[drv_index] = NULL;
862 }
863
864 /*
865 * For operations that cannot sleep, a command block is allocated at init,
866 * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
867 * which ones are free or in use. For operations that can wait for kmalloc
868 * to possible sleep, this routine can be called with get_from_pool set to 0.
869 * cmd_free() MUST be called with a got_from_pool set to 0 if cmd_alloc was.
870 */
871 static CommandList_struct *cmd_alloc(ctlr_info_t *h, int get_from_pool)
872 {
873 CommandList_struct *c;
874 int i;
875 u64bit temp64;
876 dma_addr_t cmd_dma_handle, err_dma_handle;
877
878 if (!get_from_pool) {
879 c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
880 sizeof(CommandList_struct), &cmd_dma_handle);
881 if (c == NULL)
882 return NULL;
883 memset(c, 0, sizeof(CommandList_struct));
884
885 c->cmdindex = -1;
886
887 c->err_info = (ErrorInfo_struct *)
888 pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
889 &err_dma_handle);
890
891 if (c->err_info == NULL) {
892 pci_free_consistent(h->pdev,
893 sizeof(CommandList_struct), c, cmd_dma_handle);
894 return NULL;
895 }
896 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
897 } else { /* get it out of the controllers pool */
898
899 do {
900 i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
901 if (i == h->nr_cmds)
902 return NULL;
903 } while (test_and_set_bit
904 (i & (BITS_PER_LONG - 1),
905 h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
906 #ifdef CCISS_DEBUG
907 printk(KERN_DEBUG "cciss: using command buffer %d\n", i);
908 #endif
909 c = h->cmd_pool + i;
910 memset(c, 0, sizeof(CommandList_struct));
911 cmd_dma_handle = h->cmd_pool_dhandle
912 + i * sizeof(CommandList_struct);
913 c->err_info = h->errinfo_pool + i;
914 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
915 err_dma_handle = h->errinfo_pool_dhandle
916 + i * sizeof(ErrorInfo_struct);
917 h->nr_allocs++;
918
919 c->cmdindex = i;
920 }
921
922 INIT_HLIST_NODE(&c->list);
923 c->busaddr = (__u32) cmd_dma_handle;
924 temp64.val = (__u64) err_dma_handle;
925 c->ErrDesc.Addr.lower = temp64.val32.lower;
926 c->ErrDesc.Addr.upper = temp64.val32.upper;
927 c->ErrDesc.Len = sizeof(ErrorInfo_struct);
928
929 c->ctlr = h->ctlr;
930 return c;
931 }
932
933 /*
934 * Frees a command block that was previously allocated with cmd_alloc().
935 */
936 static void cmd_free(ctlr_info_t *h, CommandList_struct *c, int got_from_pool)
937 {
938 int i;
939 u64bit temp64;
940
941 if (!got_from_pool) {
942 temp64.val32.lower = c->ErrDesc.Addr.lower;
943 temp64.val32.upper = c->ErrDesc.Addr.upper;
944 pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
945 c->err_info, (dma_addr_t) temp64.val);
946 pci_free_consistent(h->pdev, sizeof(CommandList_struct),
947 c, (dma_addr_t) c->busaddr);
948 } else {
949 i = c - h->cmd_pool;
950 clear_bit(i & (BITS_PER_LONG - 1),
951 h->cmd_pool_bits + (i / BITS_PER_LONG));
952 h->nr_frees++;
953 }
954 }
955
956 static inline ctlr_info_t *get_host(struct gendisk *disk)
957 {
958 return disk->queue->queuedata;
959 }
960
961 static inline drive_info_struct *get_drv(struct gendisk *disk)
962 {
963 return disk->private_data;
964 }
965
966 /*
967 * Open. Make sure the device is really there.
968 */
969 static int cciss_open(struct block_device *bdev, fmode_t mode)
970 {
971 ctlr_info_t *host = get_host(bdev->bd_disk);
972 drive_info_struct *drv = get_drv(bdev->bd_disk);
973
974 #ifdef CCISS_DEBUG
975 printk(KERN_DEBUG "cciss_open %s\n", bdev->bd_disk->disk_name);
976 #endif /* CCISS_DEBUG */
977
978 if (drv->busy_configuring)
979 return -EBUSY;
980 /*
981 * Root is allowed to open raw volume zero even if it's not configured
982 * so array config can still work. Root is also allowed to open any
983 * volume that has a LUN ID, so it can issue IOCTL to reread the
984 * disk information. I don't think I really like this
985 * but I'm already using way to many device nodes to claim another one
986 * for "raw controller".
987 */
988 if (drv->heads == 0) {
989 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
990 /* if not node 0 make sure it is a partition = 0 */
991 if (MINOR(bdev->bd_dev) & 0x0f) {
992 return -ENXIO;
993 /* if it is, make sure we have a LUN ID */
994 } else if (memcmp(drv->LunID, CTLR_LUNID,
995 sizeof(drv->LunID))) {
996 return -ENXIO;
997 }
998 }
999 if (!capable(CAP_SYS_ADMIN))
1000 return -EPERM;
1001 }
1002 drv->usage_count++;
1003 host->usage_count++;
1004 return 0;
1005 }
1006
1007 /*
1008 * Close. Sync first.
1009 */
1010 static int cciss_release(struct gendisk *disk, fmode_t mode)
1011 {
1012 ctlr_info_t *host = get_host(disk);
1013 drive_info_struct *drv = get_drv(disk);
1014
1015 #ifdef CCISS_DEBUG
1016 printk(KERN_DEBUG "cciss_release %s\n", disk->disk_name);
1017 #endif /* CCISS_DEBUG */
1018
1019 drv->usage_count--;
1020 host->usage_count--;
1021 return 0;
1022 }
1023
1024 #ifdef CONFIG_COMPAT
1025
1026 static int do_ioctl(struct block_device *bdev, fmode_t mode,
1027 unsigned cmd, unsigned long arg)
1028 {
1029 int ret;
1030 lock_kernel();
1031 ret = cciss_ioctl(bdev, mode, cmd, arg);
1032 unlock_kernel();
1033 return ret;
1034 }
1035
1036 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1037 unsigned cmd, unsigned long arg);
1038 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1039 unsigned cmd, unsigned long arg);
1040
1041 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
1042 unsigned cmd, unsigned long arg)
1043 {
1044 switch (cmd) {
1045 case CCISS_GETPCIINFO:
1046 case CCISS_GETINTINFO:
1047 case CCISS_SETINTINFO:
1048 case CCISS_GETNODENAME:
1049 case CCISS_SETNODENAME:
1050 case CCISS_GETHEARTBEAT:
1051 case CCISS_GETBUSTYPES:
1052 case CCISS_GETFIRMVER:
1053 case CCISS_GETDRIVVER:
1054 case CCISS_REVALIDVOLS:
1055 case CCISS_DEREGDISK:
1056 case CCISS_REGNEWDISK:
1057 case CCISS_REGNEWD:
1058 case CCISS_RESCANDISK:
1059 case CCISS_GETLUNINFO:
1060 return do_ioctl(bdev, mode, cmd, arg);
1061
1062 case CCISS_PASSTHRU32:
1063 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
1064 case CCISS_BIG_PASSTHRU32:
1065 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
1066
1067 default:
1068 return -ENOIOCTLCMD;
1069 }
1070 }
1071
1072 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1073 unsigned cmd, unsigned long arg)
1074 {
1075 IOCTL32_Command_struct __user *arg32 =
1076 (IOCTL32_Command_struct __user *) arg;
1077 IOCTL_Command_struct arg64;
1078 IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
1079 int err;
1080 u32 cp;
1081
1082 err = 0;
1083 err |=
1084 copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1085 sizeof(arg64.LUN_info));
1086 err |=
1087 copy_from_user(&arg64.Request, &arg32->Request,
1088 sizeof(arg64.Request));
1089 err |=
1090 copy_from_user(&arg64.error_info, &arg32->error_info,
1091 sizeof(arg64.error_info));
1092 err |= get_user(arg64.buf_size, &arg32->buf_size);
1093 err |= get_user(cp, &arg32->buf);
1094 arg64.buf = compat_ptr(cp);
1095 err |= copy_to_user(p, &arg64, sizeof(arg64));
1096
1097 if (err)
1098 return -EFAULT;
1099
1100 err = do_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
1101 if (err)
1102 return err;
1103 err |=
1104 copy_in_user(&arg32->error_info, &p->error_info,
1105 sizeof(arg32->error_info));
1106 if (err)
1107 return -EFAULT;
1108 return err;
1109 }
1110
1111 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1112 unsigned cmd, unsigned long arg)
1113 {
1114 BIG_IOCTL32_Command_struct __user *arg32 =
1115 (BIG_IOCTL32_Command_struct __user *) arg;
1116 BIG_IOCTL_Command_struct arg64;
1117 BIG_IOCTL_Command_struct __user *p =
1118 compat_alloc_user_space(sizeof(arg64));
1119 int err;
1120 u32 cp;
1121
1122 err = 0;
1123 err |=
1124 copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1125 sizeof(arg64.LUN_info));
1126 err |=
1127 copy_from_user(&arg64.Request, &arg32->Request,
1128 sizeof(arg64.Request));
1129 err |=
1130 copy_from_user(&arg64.error_info, &arg32->error_info,
1131 sizeof(arg64.error_info));
1132 err |= get_user(arg64.buf_size, &arg32->buf_size);
1133 err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1134 err |= get_user(cp, &arg32->buf);
1135 arg64.buf = compat_ptr(cp);
1136 err |= copy_to_user(p, &arg64, sizeof(arg64));
1137
1138 if (err)
1139 return -EFAULT;
1140
1141 err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1142 if (err)
1143 return err;
1144 err |=
1145 copy_in_user(&arg32->error_info, &p->error_info,
1146 sizeof(arg32->error_info));
1147 if (err)
1148 return -EFAULT;
1149 return err;
1150 }
1151 #endif
1152
1153 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1154 {
1155 drive_info_struct *drv = get_drv(bdev->bd_disk);
1156
1157 if (!drv->cylinders)
1158 return -ENXIO;
1159
1160 geo->heads = drv->heads;
1161 geo->sectors = drv->sectors;
1162 geo->cylinders = drv->cylinders;
1163 return 0;
1164 }
1165
1166 static void check_ioctl_unit_attention(ctlr_info_t *host, CommandList_struct *c)
1167 {
1168 if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1169 c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1170 (void)check_for_unit_attention(host, c);
1171 }
1172 /*
1173 * ioctl
1174 */
1175 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1176 unsigned int cmd, unsigned long arg)
1177 {
1178 struct gendisk *disk = bdev->bd_disk;
1179 ctlr_info_t *host = get_host(disk);
1180 drive_info_struct *drv = get_drv(disk);
1181 int ctlr = host->ctlr;
1182 void __user *argp = (void __user *)arg;
1183
1184 #ifdef CCISS_DEBUG
1185 printk(KERN_DEBUG "cciss_ioctl: Called with cmd=%x %lx\n", cmd, arg);
1186 #endif /* CCISS_DEBUG */
1187
1188 switch (cmd) {
1189 case CCISS_GETPCIINFO:
1190 {
1191 cciss_pci_info_struct pciinfo;
1192
1193 if (!arg)
1194 return -EINVAL;
1195 pciinfo.domain = pci_domain_nr(host->pdev->bus);
1196 pciinfo.bus = host->pdev->bus->number;
1197 pciinfo.dev_fn = host->pdev->devfn;
1198 pciinfo.board_id = host->board_id;
1199 if (copy_to_user
1200 (argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1201 return -EFAULT;
1202 return 0;
1203 }
1204 case CCISS_GETINTINFO:
1205 {
1206 cciss_coalint_struct intinfo;
1207 if (!arg)
1208 return -EINVAL;
1209 intinfo.delay =
1210 readl(&host->cfgtable->HostWrite.CoalIntDelay);
1211 intinfo.count =
1212 readl(&host->cfgtable->HostWrite.CoalIntCount);
1213 if (copy_to_user
1214 (argp, &intinfo, sizeof(cciss_coalint_struct)))
1215 return -EFAULT;
1216 return 0;
1217 }
1218 case CCISS_SETINTINFO:
1219 {
1220 cciss_coalint_struct intinfo;
1221 unsigned long flags;
1222 int i;
1223
1224 if (!arg)
1225 return -EINVAL;
1226 if (!capable(CAP_SYS_ADMIN))
1227 return -EPERM;
1228 if (copy_from_user
1229 (&intinfo, argp, sizeof(cciss_coalint_struct)))
1230 return -EFAULT;
1231 if ((intinfo.delay == 0) && (intinfo.count == 0))
1232 {
1233 // printk("cciss_ioctl: delay and count cannot be 0\n");
1234 return -EINVAL;
1235 }
1236 spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1237 /* Update the field, and then ring the doorbell */
1238 writel(intinfo.delay,
1239 &(host->cfgtable->HostWrite.CoalIntDelay));
1240 writel(intinfo.count,
1241 &(host->cfgtable->HostWrite.CoalIntCount));
1242 writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1243
1244 for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1245 if (!(readl(host->vaddr + SA5_DOORBELL)
1246 & CFGTBL_ChangeReq))
1247 break;
1248 /* delay and try again */
1249 udelay(1000);
1250 }
1251 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1252 if (i >= MAX_IOCTL_CONFIG_WAIT)
1253 return -EAGAIN;
1254 return 0;
1255 }
1256 case CCISS_GETNODENAME:
1257 {
1258 NodeName_type NodeName;
1259 int i;
1260
1261 if (!arg)
1262 return -EINVAL;
1263 for (i = 0; i < 16; i++)
1264 NodeName[i] =
1265 readb(&host->cfgtable->ServerName[i]);
1266 if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1267 return -EFAULT;
1268 return 0;
1269 }
1270 case CCISS_SETNODENAME:
1271 {
1272 NodeName_type NodeName;
1273 unsigned long flags;
1274 int i;
1275
1276 if (!arg)
1277 return -EINVAL;
1278 if (!capable(CAP_SYS_ADMIN))
1279 return -EPERM;
1280
1281 if (copy_from_user
1282 (NodeName, argp, sizeof(NodeName_type)))
1283 return -EFAULT;
1284
1285 spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1286
1287 /* Update the field, and then ring the doorbell */
1288 for (i = 0; i < 16; i++)
1289 writeb(NodeName[i],
1290 &host->cfgtable->ServerName[i]);
1291
1292 writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1293
1294 for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1295 if (!(readl(host->vaddr + SA5_DOORBELL)
1296 & CFGTBL_ChangeReq))
1297 break;
1298 /* delay and try again */
1299 udelay(1000);
1300 }
1301 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1302 if (i >= MAX_IOCTL_CONFIG_WAIT)
1303 return -EAGAIN;
1304 return 0;
1305 }
1306
1307 case CCISS_GETHEARTBEAT:
1308 {
1309 Heartbeat_type heartbeat;
1310
1311 if (!arg)
1312 return -EINVAL;
1313 heartbeat = readl(&host->cfgtable->HeartBeat);
1314 if (copy_to_user
1315 (argp, &heartbeat, sizeof(Heartbeat_type)))
1316 return -EFAULT;
1317 return 0;
1318 }
1319 case CCISS_GETBUSTYPES:
1320 {
1321 BusTypes_type BusTypes;
1322
1323 if (!arg)
1324 return -EINVAL;
1325 BusTypes = readl(&host->cfgtable->BusTypes);
1326 if (copy_to_user
1327 (argp, &BusTypes, sizeof(BusTypes_type)))
1328 return -EFAULT;
1329 return 0;
1330 }
1331 case CCISS_GETFIRMVER:
1332 {
1333 FirmwareVer_type firmware;
1334
1335 if (!arg)
1336 return -EINVAL;
1337 memcpy(firmware, host->firm_ver, 4);
1338
1339 if (copy_to_user
1340 (argp, firmware, sizeof(FirmwareVer_type)))
1341 return -EFAULT;
1342 return 0;
1343 }
1344 case CCISS_GETDRIVVER:
1345 {
1346 DriverVer_type DriverVer = DRIVER_VERSION;
1347
1348 if (!arg)
1349 return -EINVAL;
1350
1351 if (copy_to_user
1352 (argp, &DriverVer, sizeof(DriverVer_type)))
1353 return -EFAULT;
1354 return 0;
1355 }
1356
1357 case CCISS_DEREGDISK:
1358 case CCISS_REGNEWD:
1359 case CCISS_REVALIDVOLS:
1360 return rebuild_lun_table(host, 0, 1);
1361
1362 case CCISS_GETLUNINFO:{
1363 LogvolInfo_struct luninfo;
1364
1365 memcpy(&luninfo.LunID, drv->LunID,
1366 sizeof(luninfo.LunID));
1367 luninfo.num_opens = drv->usage_count;
1368 luninfo.num_parts = 0;
1369 if (copy_to_user(argp, &luninfo,
1370 sizeof(LogvolInfo_struct)))
1371 return -EFAULT;
1372 return 0;
1373 }
1374 case CCISS_PASSTHRU:
1375 {
1376 IOCTL_Command_struct iocommand;
1377 CommandList_struct *c;
1378 char *buff = NULL;
1379 u64bit temp64;
1380 unsigned long flags;
1381 DECLARE_COMPLETION_ONSTACK(wait);
1382
1383 if (!arg)
1384 return -EINVAL;
1385
1386 if (!capable(CAP_SYS_RAWIO))
1387 return -EPERM;
1388
1389 if (copy_from_user
1390 (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1391 return -EFAULT;
1392 if ((iocommand.buf_size < 1) &&
1393 (iocommand.Request.Type.Direction != XFER_NONE)) {
1394 return -EINVAL;
1395 }
1396 #if 0 /* 'buf_size' member is 16-bits, and always smaller than kmalloc limit */
1397 /* Check kmalloc limits */
1398 if (iocommand.buf_size > 128000)
1399 return -EINVAL;
1400 #endif
1401 if (iocommand.buf_size > 0) {
1402 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1403 if (buff == NULL)
1404 return -EFAULT;
1405 }
1406 if (iocommand.Request.Type.Direction == XFER_WRITE) {
1407 /* Copy the data into the buffer we created */
1408 if (copy_from_user
1409 (buff, iocommand.buf, iocommand.buf_size)) {
1410 kfree(buff);
1411 return -EFAULT;
1412 }
1413 } else {
1414 memset(buff, 0, iocommand.buf_size);
1415 }
1416 if ((c = cmd_alloc(host, 0)) == NULL) {
1417 kfree(buff);
1418 return -ENOMEM;
1419 }
1420 /* Fill in the command type */
1421 c->cmd_type = CMD_IOCTL_PEND;
1422 /* Fill in Command Header */
1423 c->Header.ReplyQueue = 0; /* unused in simple mode */
1424 if (iocommand.buf_size > 0) /* buffer to fill */
1425 {
1426 c->Header.SGList = 1;
1427 c->Header.SGTotal = 1;
1428 } else /* no buffers to fill */
1429 {
1430 c->Header.SGList = 0;
1431 c->Header.SGTotal = 0;
1432 }
1433 c->Header.LUN = iocommand.LUN_info;
1434 /* use the kernel address the cmd block for tag */
1435 c->Header.Tag.lower = c->busaddr;
1436
1437 /* Fill in Request block */
1438 c->Request = iocommand.Request;
1439
1440 /* Fill in the scatter gather information */
1441 if (iocommand.buf_size > 0) {
1442 temp64.val = pci_map_single(host->pdev, buff,
1443 iocommand.buf_size,
1444 PCI_DMA_BIDIRECTIONAL);
1445 c->SG[0].Addr.lower = temp64.val32.lower;
1446 c->SG[0].Addr.upper = temp64.val32.upper;
1447 c->SG[0].Len = iocommand.buf_size;
1448 c->SG[0].Ext = 0; /* we are not chaining */
1449 }
1450 c->waiting = &wait;
1451
1452 /* Put the request on the tail of the request queue */
1453 spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1454 addQ(&host->reqQ, c);
1455 host->Qdepth++;
1456 start_io(host);
1457 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1458
1459 wait_for_completion(&wait);
1460
1461 /* unlock the buffers from DMA */
1462 temp64.val32.lower = c->SG[0].Addr.lower;
1463 temp64.val32.upper = c->SG[0].Addr.upper;
1464 pci_unmap_single(host->pdev, (dma_addr_t) temp64.val,
1465 iocommand.buf_size,
1466 PCI_DMA_BIDIRECTIONAL);
1467
1468 check_ioctl_unit_attention(host, c);
1469
1470 /* Copy the error information out */
1471 iocommand.error_info = *(c->err_info);
1472 if (copy_to_user
1473 (argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1474 kfree(buff);
1475 cmd_free(host, c, 0);
1476 return -EFAULT;
1477 }
1478
1479 if (iocommand.Request.Type.Direction == XFER_READ) {
1480 /* Copy the data out of the buffer we created */
1481 if (copy_to_user
1482 (iocommand.buf, buff, iocommand.buf_size)) {
1483 kfree(buff);
1484 cmd_free(host, c, 0);
1485 return -EFAULT;
1486 }
1487 }
1488 kfree(buff);
1489 cmd_free(host, c, 0);
1490 return 0;
1491 }
1492 case CCISS_BIG_PASSTHRU:{
1493 BIG_IOCTL_Command_struct *ioc;
1494 CommandList_struct *c;
1495 unsigned char **buff = NULL;
1496 int *buff_size = NULL;
1497 u64bit temp64;
1498 unsigned long flags;
1499 BYTE sg_used = 0;
1500 int status = 0;
1501 int i;
1502 DECLARE_COMPLETION_ONSTACK(wait);
1503 __u32 left;
1504 __u32 sz;
1505 BYTE __user *data_ptr;
1506
1507 if (!arg)
1508 return -EINVAL;
1509 if (!capable(CAP_SYS_RAWIO))
1510 return -EPERM;
1511 ioc = (BIG_IOCTL_Command_struct *)
1512 kmalloc(sizeof(*ioc), GFP_KERNEL);
1513 if (!ioc) {
1514 status = -ENOMEM;
1515 goto cleanup1;
1516 }
1517 if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1518 status = -EFAULT;
1519 goto cleanup1;
1520 }
1521 if ((ioc->buf_size < 1) &&
1522 (ioc->Request.Type.Direction != XFER_NONE)) {
1523 status = -EINVAL;
1524 goto cleanup1;
1525 }
1526 /* Check kmalloc limits using all SGs */
1527 if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1528 status = -EINVAL;
1529 goto cleanup1;
1530 }
1531 if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1532 status = -EINVAL;
1533 goto cleanup1;
1534 }
1535 buff =
1536 kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1537 if (!buff) {
1538 status = -ENOMEM;
1539 goto cleanup1;
1540 }
1541 buff_size = kmalloc(MAXSGENTRIES * sizeof(int),
1542 GFP_KERNEL);
1543 if (!buff_size) {
1544 status = -ENOMEM;
1545 goto cleanup1;
1546 }
1547 left = ioc->buf_size;
1548 data_ptr = ioc->buf;
1549 while (left) {
1550 sz = (left >
1551 ioc->malloc_size) ? ioc->
1552 malloc_size : left;
1553 buff_size[sg_used] = sz;
1554 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1555 if (buff[sg_used] == NULL) {
1556 status = -ENOMEM;
1557 goto cleanup1;
1558 }
1559 if (ioc->Request.Type.Direction == XFER_WRITE) {
1560 if (copy_from_user
1561 (buff[sg_used], data_ptr, sz)) {
1562 status = -EFAULT;
1563 goto cleanup1;
1564 }
1565 } else {
1566 memset(buff[sg_used], 0, sz);
1567 }
1568 left -= sz;
1569 data_ptr += sz;
1570 sg_used++;
1571 }
1572 if ((c = cmd_alloc(host, 0)) == NULL) {
1573 status = -ENOMEM;
1574 goto cleanup1;
1575 }
1576 c->cmd_type = CMD_IOCTL_PEND;
1577 c->Header.ReplyQueue = 0;
1578
1579 if (ioc->buf_size > 0) {
1580 c->Header.SGList = sg_used;
1581 c->Header.SGTotal = sg_used;
1582 } else {
1583 c->Header.SGList = 0;
1584 c->Header.SGTotal = 0;
1585 }
1586 c->Header.LUN = ioc->LUN_info;
1587 c->Header.Tag.lower = c->busaddr;
1588
1589 c->Request = ioc->Request;
1590 if (ioc->buf_size > 0) {
1591 int i;
1592 for (i = 0; i < sg_used; i++) {
1593 temp64.val =
1594 pci_map_single(host->pdev, buff[i],
1595 buff_size[i],
1596 PCI_DMA_BIDIRECTIONAL);
1597 c->SG[i].Addr.lower =
1598 temp64.val32.lower;
1599 c->SG[i].Addr.upper =
1600 temp64.val32.upper;
1601 c->SG[i].Len = buff_size[i];
1602 c->SG[i].Ext = 0; /* we are not chaining */
1603 }
1604 }
1605 c->waiting = &wait;
1606 /* Put the request on the tail of the request queue */
1607 spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1608 addQ(&host->reqQ, c);
1609 host->Qdepth++;
1610 start_io(host);
1611 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1612 wait_for_completion(&wait);
1613 /* unlock the buffers from DMA */
1614 for (i = 0; i < sg_used; i++) {
1615 temp64.val32.lower = c->SG[i].Addr.lower;
1616 temp64.val32.upper = c->SG[i].Addr.upper;
1617 pci_unmap_single(host->pdev,
1618 (dma_addr_t) temp64.val, buff_size[i],
1619 PCI_DMA_BIDIRECTIONAL);
1620 }
1621 check_ioctl_unit_attention(host, c);
1622 /* Copy the error information out */
1623 ioc->error_info = *(c->err_info);
1624 if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1625 cmd_free(host, c, 0);
1626 status = -EFAULT;
1627 goto cleanup1;
1628 }
1629 if (ioc->Request.Type.Direction == XFER_READ) {
1630 /* Copy the data out of the buffer we created */
1631 BYTE __user *ptr = ioc->buf;
1632 for (i = 0; i < sg_used; i++) {
1633 if (copy_to_user
1634 (ptr, buff[i], buff_size[i])) {
1635 cmd_free(host, c, 0);
1636 status = -EFAULT;
1637 goto cleanup1;
1638 }
1639 ptr += buff_size[i];
1640 }
1641 }
1642 cmd_free(host, c, 0);
1643 status = 0;
1644 cleanup1:
1645 if (buff) {
1646 for (i = 0; i < sg_used; i++)
1647 kfree(buff[i]);
1648 kfree(buff);
1649 }
1650 kfree(buff_size);
1651 kfree(ioc);
1652 return status;
1653 }
1654
1655 /* scsi_cmd_ioctl handles these, below, though some are not */
1656 /* very meaningful for cciss. SG_IO is the main one people want. */
1657
1658 case SG_GET_VERSION_NUM:
1659 case SG_SET_TIMEOUT:
1660 case SG_GET_TIMEOUT:
1661 case SG_GET_RESERVED_SIZE:
1662 case SG_SET_RESERVED_SIZE:
1663 case SG_EMULATED_HOST:
1664 case SG_IO:
1665 case SCSI_IOCTL_SEND_COMMAND:
1666 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
1667
1668 /* scsi_cmd_ioctl would normally handle these, below, but */
1669 /* they aren't a good fit for cciss, as CD-ROMs are */
1670 /* not supported, and we don't have any bus/target/lun */
1671 /* which we present to the kernel. */
1672
1673 case CDROM_SEND_PACKET:
1674 case CDROMCLOSETRAY:
1675 case CDROMEJECT:
1676 case SCSI_IOCTL_GET_IDLUN:
1677 case SCSI_IOCTL_GET_BUS_NUMBER:
1678 default:
1679 return -ENOTTY;
1680 }
1681 }
1682
1683 static void cciss_check_queues(ctlr_info_t *h)
1684 {
1685 int start_queue = h->next_to_run;
1686 int i;
1687
1688 /* check to see if we have maxed out the number of commands that can
1689 * be placed on the queue. If so then exit. We do this check here
1690 * in case the interrupt we serviced was from an ioctl and did not
1691 * free any new commands.
1692 */
1693 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1694 return;
1695
1696 /* We have room on the queue for more commands. Now we need to queue
1697 * them up. We will also keep track of the next queue to run so
1698 * that every queue gets a chance to be started first.
1699 */
1700 for (i = 0; i < h->highest_lun + 1; i++) {
1701 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1702 /* make sure the disk has been added and the drive is real
1703 * because this can be called from the middle of init_one.
1704 */
1705 if (!h->drv[curr_queue])
1706 continue;
1707 if (!(h->drv[curr_queue]->queue) ||
1708 !(h->drv[curr_queue]->heads))
1709 continue;
1710 blk_start_queue(h->gendisk[curr_queue]->queue);
1711
1712 /* check to see if we have maxed out the number of commands
1713 * that can be placed on the queue.
1714 */
1715 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1716 if (curr_queue == start_queue) {
1717 h->next_to_run =
1718 (start_queue + 1) % (h->highest_lun + 1);
1719 break;
1720 } else {
1721 h->next_to_run = curr_queue;
1722 break;
1723 }
1724 }
1725 }
1726 }
1727
1728 static void cciss_softirq_done(struct request *rq)
1729 {
1730 CommandList_struct *cmd = rq->completion_data;
1731 ctlr_info_t *h = hba[cmd->ctlr];
1732 SGDescriptor_struct *curr_sg = cmd->SG;
1733 unsigned long flags;
1734 u64bit temp64;
1735 int i, ddir;
1736 int sg_index = 0;
1737
1738 if (cmd->Request.Type.Direction == XFER_READ)
1739 ddir = PCI_DMA_FROMDEVICE;
1740 else
1741 ddir = PCI_DMA_TODEVICE;
1742
1743 /* command did not need to be retried */
1744 /* unmap the DMA mapping for all the scatter gather elements */
1745 for (i = 0; i < cmd->Header.SGList; i++) {
1746 if (curr_sg[sg_index].Ext == CCISS_SG_CHAIN) {
1747 cciss_unmap_sg_chain_block(h, cmd);
1748 /* Point to the next block */
1749 curr_sg = h->cmd_sg_list[cmd->cmdindex];
1750 sg_index = 0;
1751 }
1752 temp64.val32.lower = curr_sg[sg_index].Addr.lower;
1753 temp64.val32.upper = curr_sg[sg_index].Addr.upper;
1754 pci_unmap_page(h->pdev, temp64.val, curr_sg[sg_index].Len,
1755 ddir);
1756 ++sg_index;
1757 }
1758
1759 #ifdef CCISS_DEBUG
1760 printk("Done with %p\n", rq);
1761 #endif /* CCISS_DEBUG */
1762
1763 /* set the residual count for pc requests */
1764 if (blk_pc_request(rq))
1765 rq->resid_len = cmd->err_info->ResidualCnt;
1766
1767 blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1768
1769 spin_lock_irqsave(&h->lock, flags);
1770 cmd_free(h, cmd, 1);
1771 cciss_check_queues(h);
1772 spin_unlock_irqrestore(&h->lock, flags);
1773 }
1774
1775 static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1776 unsigned char scsi3addr[], uint32_t log_unit)
1777 {
1778 memcpy(scsi3addr, h->drv[log_unit]->LunID,
1779 sizeof(h->drv[log_unit]->LunID));
1780 }
1781
1782 /* This function gets the SCSI vendor, model, and revision of a logical drive
1783 * via the inquiry page 0. Model, vendor, and rev are set to empty strings if
1784 * they cannot be read.
1785 */
1786 static void cciss_get_device_descr(int ctlr, int logvol,
1787 char *vendor, char *model, char *rev)
1788 {
1789 int rc;
1790 InquiryData_struct *inq_buf;
1791 unsigned char scsi3addr[8];
1792
1793 *vendor = '\0';
1794 *model = '\0';
1795 *rev = '\0';
1796
1797 inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1798 if (!inq_buf)
1799 return;
1800
1801 log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1802 rc = sendcmd_withirq(CISS_INQUIRY, ctlr, inq_buf, sizeof(*inq_buf), 0,
1803 scsi3addr, TYPE_CMD);
1804 if (rc == IO_OK) {
1805 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1806 vendor[VENDOR_LEN] = '\0';
1807 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1808 model[MODEL_LEN] = '\0';
1809 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1810 rev[REV_LEN] = '\0';
1811 }
1812
1813 kfree(inq_buf);
1814 return;
1815 }
1816
1817 /* This function gets the serial number of a logical drive via
1818 * inquiry page 0x83. Serial no. is 16 bytes. If the serial
1819 * number cannot be had, for whatever reason, 16 bytes of 0xff
1820 * are returned instead.
1821 */
1822 static void cciss_get_serial_no(int ctlr, int logvol,
1823 unsigned char *serial_no, int buflen)
1824 {
1825 #define PAGE_83_INQ_BYTES 64
1826 int rc;
1827 unsigned char *buf;
1828 unsigned char scsi3addr[8];
1829
1830 if (buflen > 16)
1831 buflen = 16;
1832 memset(serial_no, 0xff, buflen);
1833 buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1834 if (!buf)
1835 return;
1836 memset(serial_no, 0, buflen);
1837 log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1838 rc = sendcmd_withirq(CISS_INQUIRY, ctlr, buf,
1839 PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1840 if (rc == IO_OK)
1841 memcpy(serial_no, &buf[8], buflen);
1842 kfree(buf);
1843 return;
1844 }
1845
1846 /*
1847 * cciss_add_disk sets up the block device queue for a logical drive
1848 */
1849 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1850 int drv_index)
1851 {
1852 disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1853 if (!disk->queue)
1854 goto init_queue_failure;
1855 sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1856 disk->major = h->major;
1857 disk->first_minor = drv_index << NWD_SHIFT;
1858 disk->fops = &cciss_fops;
1859 if (cciss_create_ld_sysfs_entry(h, drv_index))
1860 goto cleanup_queue;
1861 disk->private_data = h->drv[drv_index];
1862 disk->driverfs_dev = &h->drv[drv_index]->dev;
1863
1864 /* Set up queue information */
1865 blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1866
1867 /* This is a hardware imposed limit. */
1868 blk_queue_max_segments(disk->queue, h->maxsgentries);
1869
1870 blk_queue_max_hw_sectors(disk->queue, h->cciss_max_sectors);
1871
1872 blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1873
1874 disk->queue->queuedata = h;
1875
1876 blk_queue_logical_block_size(disk->queue,
1877 h->drv[drv_index]->block_size);
1878
1879 /* Make sure all queue data is written out before */
1880 /* setting h->drv[drv_index]->queue, as setting this */
1881 /* allows the interrupt handler to start the queue */
1882 wmb();
1883 h->drv[drv_index]->queue = disk->queue;
1884 add_disk(disk);
1885 return 0;
1886
1887 cleanup_queue:
1888 blk_cleanup_queue(disk->queue);
1889 disk->queue = NULL;
1890 init_queue_failure:
1891 return -1;
1892 }
1893
1894 /* This function will check the usage_count of the drive to be updated/added.
1895 * If the usage_count is zero and it is a heretofore unknown drive, or,
1896 * the drive's capacity, geometry, or serial number has changed,
1897 * then the drive information will be updated and the disk will be
1898 * re-registered with the kernel. If these conditions don't hold,
1899 * then it will be left alone for the next reboot. The exception to this
1900 * is disk 0 which will always be left registered with the kernel since it
1901 * is also the controller node. Any changes to disk 0 will show up on
1902 * the next reboot.
1903 */
1904 static void cciss_update_drive_info(int ctlr, int drv_index, int first_time,
1905 int via_ioctl)
1906 {
1907 ctlr_info_t *h = hba[ctlr];
1908 struct gendisk *disk;
1909 InquiryData_struct *inq_buff = NULL;
1910 unsigned int block_size;
1911 sector_t total_size;
1912 unsigned long flags = 0;
1913 int ret = 0;
1914 drive_info_struct *drvinfo;
1915
1916 /* Get information about the disk and modify the driver structure */
1917 inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1918 drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
1919 if (inq_buff == NULL || drvinfo == NULL)
1920 goto mem_msg;
1921
1922 /* testing to see if 16-byte CDBs are already being used */
1923 if (h->cciss_read == CCISS_READ_16) {
1924 cciss_read_capacity_16(h->ctlr, drv_index,
1925 &total_size, &block_size);
1926
1927 } else {
1928 cciss_read_capacity(ctlr, drv_index, &total_size, &block_size);
1929 /* if read_capacity returns all F's this volume is >2TB */
1930 /* in size so we switch to 16-byte CDB's for all */
1931 /* read/write ops */
1932 if (total_size == 0xFFFFFFFFULL) {
1933 cciss_read_capacity_16(ctlr, drv_index,
1934 &total_size, &block_size);
1935 h->cciss_read = CCISS_READ_16;
1936 h->cciss_write = CCISS_WRITE_16;
1937 } else {
1938 h->cciss_read = CCISS_READ_10;
1939 h->cciss_write = CCISS_WRITE_10;
1940 }
1941 }
1942
1943 cciss_geometry_inquiry(ctlr, drv_index, total_size, block_size,
1944 inq_buff, drvinfo);
1945 drvinfo->block_size = block_size;
1946 drvinfo->nr_blocks = total_size + 1;
1947
1948 cciss_get_device_descr(ctlr, drv_index, drvinfo->vendor,
1949 drvinfo->model, drvinfo->rev);
1950 cciss_get_serial_no(ctlr, drv_index, drvinfo->serial_no,
1951 sizeof(drvinfo->serial_no));
1952 /* Save the lunid in case we deregister the disk, below. */
1953 memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
1954 sizeof(drvinfo->LunID));
1955
1956 /* Is it the same disk we already know, and nothing's changed? */
1957 if (h->drv[drv_index]->raid_level != -1 &&
1958 ((memcmp(drvinfo->serial_no,
1959 h->drv[drv_index]->serial_no, 16) == 0) &&
1960 drvinfo->block_size == h->drv[drv_index]->block_size &&
1961 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
1962 drvinfo->heads == h->drv[drv_index]->heads &&
1963 drvinfo->sectors == h->drv[drv_index]->sectors &&
1964 drvinfo->cylinders == h->drv[drv_index]->cylinders))
1965 /* The disk is unchanged, nothing to update */
1966 goto freeret;
1967
1968 /* If we get here it's not the same disk, or something's changed,
1969 * so we need to * deregister it, and re-register it, if it's not
1970 * in use.
1971 * If the disk already exists then deregister it before proceeding
1972 * (unless it's the first disk (for the controller node).
1973 */
1974 if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
1975 printk(KERN_WARNING "disk %d has changed.\n", drv_index);
1976 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1977 h->drv[drv_index]->busy_configuring = 1;
1978 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1979
1980 /* deregister_disk sets h->drv[drv_index]->queue = NULL
1981 * which keeps the interrupt handler from starting
1982 * the queue.
1983 */
1984 ret = deregister_disk(h, drv_index, 0, via_ioctl);
1985 }
1986
1987 /* If the disk is in use return */
1988 if (ret)
1989 goto freeret;
1990
1991 /* Save the new information from cciss_geometry_inquiry
1992 * and serial number inquiry. If the disk was deregistered
1993 * above, then h->drv[drv_index] will be NULL.
1994 */
1995 if (h->drv[drv_index] == NULL) {
1996 drvinfo->device_initialized = 0;
1997 h->drv[drv_index] = drvinfo;
1998 drvinfo = NULL; /* so it won't be freed below. */
1999 } else {
2000 /* special case for cxd0 */
2001 h->drv[drv_index]->block_size = drvinfo->block_size;
2002 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
2003 h->drv[drv_index]->heads = drvinfo->heads;
2004 h->drv[drv_index]->sectors = drvinfo->sectors;
2005 h->drv[drv_index]->cylinders = drvinfo->cylinders;
2006 h->drv[drv_index]->raid_level = drvinfo->raid_level;
2007 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
2008 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
2009 VENDOR_LEN + 1);
2010 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
2011 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
2012 }
2013
2014 ++h->num_luns;
2015 disk = h->gendisk[drv_index];
2016 set_capacity(disk, h->drv[drv_index]->nr_blocks);
2017
2018 /* If it's not disk 0 (drv_index != 0)
2019 * or if it was disk 0, but there was previously
2020 * no actual corresponding configured logical drive
2021 * (raid_leve == -1) then we want to update the
2022 * logical drive's information.
2023 */
2024 if (drv_index || first_time) {
2025 if (cciss_add_disk(h, disk, drv_index) != 0) {
2026 cciss_free_gendisk(h, drv_index);
2027 cciss_free_drive_info(h, drv_index);
2028 printk(KERN_WARNING "cciss:%d could not update "
2029 "disk %d\n", h->ctlr, drv_index);
2030 --h->num_luns;
2031 }
2032 }
2033
2034 freeret:
2035 kfree(inq_buff);
2036 kfree(drvinfo);
2037 return;
2038 mem_msg:
2039 printk(KERN_ERR "cciss: out of memory\n");
2040 goto freeret;
2041 }
2042
2043 /* This function will find the first index of the controllers drive array
2044 * that has a null drv pointer and allocate the drive info struct and
2045 * will return that index This is where new drives will be added.
2046 * If the index to be returned is greater than the highest_lun index for
2047 * the controller then highest_lun is set * to this new index.
2048 * If there are no available indexes or if tha allocation fails, then -1
2049 * is returned. * "controller_node" is used to know if this is a real
2050 * logical drive, or just the controller node, which determines if this
2051 * counts towards highest_lun.
2052 */
2053 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
2054 {
2055 int i;
2056 drive_info_struct *drv;
2057
2058 /* Search for an empty slot for our drive info */
2059 for (i = 0; i < CISS_MAX_LUN; i++) {
2060
2061 /* if not cxd0 case, and it's occupied, skip it. */
2062 if (h->drv[i] && i != 0)
2063 continue;
2064 /*
2065 * If it's cxd0 case, and drv is alloc'ed already, and a
2066 * disk is configured there, skip it.
2067 */
2068 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2069 continue;
2070
2071 /*
2072 * We've found an empty slot. Update highest_lun
2073 * provided this isn't just the fake cxd0 controller node.
2074 */
2075 if (i > h->highest_lun && !controller_node)
2076 h->highest_lun = i;
2077
2078 /* If adding a real disk at cxd0, and it's already alloc'ed */
2079 if (i == 0 && h->drv[i] != NULL)
2080 return i;
2081
2082 /*
2083 * Found an empty slot, not already alloc'ed. Allocate it.
2084 * Mark it with raid_level == -1, so we know it's new later on.
2085 */
2086 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2087 if (!drv)
2088 return -1;
2089 drv->raid_level = -1; /* so we know it's new */
2090 h->drv[i] = drv;
2091 return i;
2092 }
2093 return -1;
2094 }
2095
2096 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2097 {
2098 kfree(h->drv[drv_index]);
2099 h->drv[drv_index] = NULL;
2100 }
2101
2102 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2103 {
2104 put_disk(h->gendisk[drv_index]);
2105 h->gendisk[drv_index] = NULL;
2106 }
2107
2108 /* cciss_add_gendisk finds a free hba[]->drv structure
2109 * and allocates a gendisk if needed, and sets the lunid
2110 * in the drvinfo structure. It returns the index into
2111 * the ->drv[] array, or -1 if none are free.
2112 * is_controller_node indicates whether highest_lun should
2113 * count this disk, or if it's only being added to provide
2114 * a means to talk to the controller in case no logical
2115 * drives have yet been configured.
2116 */
2117 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2118 int controller_node)
2119 {
2120 int drv_index;
2121
2122 drv_index = cciss_alloc_drive_info(h, controller_node);
2123 if (drv_index == -1)
2124 return -1;
2125
2126 /*Check if the gendisk needs to be allocated */
2127 if (!h->gendisk[drv_index]) {
2128 h->gendisk[drv_index] =
2129 alloc_disk(1 << NWD_SHIFT);
2130 if (!h->gendisk[drv_index]) {
2131 printk(KERN_ERR "cciss%d: could not "
2132 "allocate a new disk %d\n",
2133 h->ctlr, drv_index);
2134 goto err_free_drive_info;
2135 }
2136 }
2137 memcpy(h->drv[drv_index]->LunID, lunid,
2138 sizeof(h->drv[drv_index]->LunID));
2139 if (cciss_create_ld_sysfs_entry(h, drv_index))
2140 goto err_free_disk;
2141 /* Don't need to mark this busy because nobody */
2142 /* else knows about this disk yet to contend */
2143 /* for access to it. */
2144 h->drv[drv_index]->busy_configuring = 0;
2145 wmb();
2146 return drv_index;
2147
2148 err_free_disk:
2149 cciss_free_gendisk(h, drv_index);
2150 err_free_drive_info:
2151 cciss_free_drive_info(h, drv_index);
2152 return -1;
2153 }
2154
2155 /* This is for the special case of a controller which
2156 * has no logical drives. In this case, we still need
2157 * to register a disk so the controller can be accessed
2158 * by the Array Config Utility.
2159 */
2160 static void cciss_add_controller_node(ctlr_info_t *h)
2161 {
2162 struct gendisk *disk;
2163 int drv_index;
2164
2165 if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2166 return;
2167
2168 drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2169 if (drv_index == -1)
2170 goto error;
2171 h->drv[drv_index]->block_size = 512;
2172 h->drv[drv_index]->nr_blocks = 0;
2173 h->drv[drv_index]->heads = 0;
2174 h->drv[drv_index]->sectors = 0;
2175 h->drv[drv_index]->cylinders = 0;
2176 h->drv[drv_index]->raid_level = -1;
2177 memset(h->drv[drv_index]->serial_no, 0, 16);
2178 disk = h->gendisk[drv_index];
2179 if (cciss_add_disk(h, disk, drv_index) == 0)
2180 return;
2181 cciss_free_gendisk(h, drv_index);
2182 cciss_free_drive_info(h, drv_index);
2183 error:
2184 printk(KERN_WARNING "cciss%d: could not "
2185 "add disk 0.\n", h->ctlr);
2186 return;
2187 }
2188
2189 /* This function will add and remove logical drives from the Logical
2190 * drive array of the controller and maintain persistency of ordering
2191 * so that mount points are preserved until the next reboot. This allows
2192 * for the removal of logical drives in the middle of the drive array
2193 * without a re-ordering of those drives.
2194 * INPUT
2195 * h = The controller to perform the operations on
2196 */
2197 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2198 int via_ioctl)
2199 {
2200 int ctlr = h->ctlr;
2201 int num_luns;
2202 ReportLunData_struct *ld_buff = NULL;
2203 int return_code;
2204 int listlength = 0;
2205 int i;
2206 int drv_found;
2207 int drv_index = 0;
2208 unsigned char lunid[8] = CTLR_LUNID;
2209 unsigned long flags;
2210
2211 if (!capable(CAP_SYS_RAWIO))
2212 return -EPERM;
2213
2214 /* Set busy_configuring flag for this operation */
2215 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2216 if (h->busy_configuring) {
2217 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2218 return -EBUSY;
2219 }
2220 h->busy_configuring = 1;
2221 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2222
2223 ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2224 if (ld_buff == NULL)
2225 goto mem_msg;
2226
2227 return_code = sendcmd_withirq(CISS_REPORT_LOG, ctlr, ld_buff,
2228 sizeof(ReportLunData_struct),
2229 0, CTLR_LUNID, TYPE_CMD);
2230
2231 if (return_code == IO_OK)
2232 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2233 else { /* reading number of logical volumes failed */
2234 printk(KERN_WARNING "cciss: report logical volume"
2235 " command failed\n");
2236 listlength = 0;
2237 goto freeret;
2238 }
2239
2240 num_luns = listlength / 8; /* 8 bytes per entry */
2241 if (num_luns > CISS_MAX_LUN) {
2242 num_luns = CISS_MAX_LUN;
2243 printk(KERN_WARNING "cciss: more luns configured"
2244 " on controller than can be handled by"
2245 " this driver.\n");
2246 }
2247
2248 if (num_luns == 0)
2249 cciss_add_controller_node(h);
2250
2251 /* Compare controller drive array to driver's drive array
2252 * to see if any drives are missing on the controller due
2253 * to action of Array Config Utility (user deletes drive)
2254 * and deregister logical drives which have disappeared.
2255 */
2256 for (i = 0; i <= h->highest_lun; i++) {
2257 int j;
2258 drv_found = 0;
2259
2260 /* skip holes in the array from already deleted drives */
2261 if (h->drv[i] == NULL)
2262 continue;
2263
2264 for (j = 0; j < num_luns; j++) {
2265 memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2266 if (memcmp(h->drv[i]->LunID, lunid,
2267 sizeof(lunid)) == 0) {
2268 drv_found = 1;
2269 break;
2270 }
2271 }
2272 if (!drv_found) {
2273 /* Deregister it from the OS, it's gone. */
2274 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2275 h->drv[i]->busy_configuring = 1;
2276 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2277 return_code = deregister_disk(h, i, 1, via_ioctl);
2278 if (h->drv[i] != NULL)
2279 h->drv[i]->busy_configuring = 0;
2280 }
2281 }
2282
2283 /* Compare controller drive array to driver's drive array.
2284 * Check for updates in the drive information and any new drives
2285 * on the controller due to ACU adding logical drives, or changing
2286 * a logical drive's size, etc. Reregister any new/changed drives
2287 */
2288 for (i = 0; i < num_luns; i++) {
2289 int j;
2290
2291 drv_found = 0;
2292
2293 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2294 /* Find if the LUN is already in the drive array
2295 * of the driver. If so then update its info
2296 * if not in use. If it does not exist then find
2297 * the first free index and add it.
2298 */
2299 for (j = 0; j <= h->highest_lun; j++) {
2300 if (h->drv[j] != NULL &&
2301 memcmp(h->drv[j]->LunID, lunid,
2302 sizeof(h->drv[j]->LunID)) == 0) {
2303 drv_index = j;
2304 drv_found = 1;
2305 break;
2306 }
2307 }
2308
2309 /* check if the drive was found already in the array */
2310 if (!drv_found) {
2311 drv_index = cciss_add_gendisk(h, lunid, 0);
2312 if (drv_index == -1)
2313 goto freeret;
2314 }
2315 cciss_update_drive_info(ctlr, drv_index, first_time,
2316 via_ioctl);
2317 } /* end for */
2318
2319 freeret:
2320 kfree(ld_buff);
2321 h->busy_configuring = 0;
2322 /* We return -1 here to tell the ACU that we have registered/updated
2323 * all of the drives that we can and to keep it from calling us
2324 * additional times.
2325 */
2326 return -1;
2327 mem_msg:
2328 printk(KERN_ERR "cciss: out of memory\n");
2329 h->busy_configuring = 0;
2330 goto freeret;
2331 }
2332
2333 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2334 {
2335 /* zero out the disk size info */
2336 drive_info->nr_blocks = 0;
2337 drive_info->block_size = 0;
2338 drive_info->heads = 0;
2339 drive_info->sectors = 0;
2340 drive_info->cylinders = 0;
2341 drive_info->raid_level = -1;
2342 memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2343 memset(drive_info->model, 0, sizeof(drive_info->model));
2344 memset(drive_info->rev, 0, sizeof(drive_info->rev));
2345 memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2346 /*
2347 * don't clear the LUNID though, we need to remember which
2348 * one this one is.
2349 */
2350 }
2351
2352 /* This function will deregister the disk and it's queue from the
2353 * kernel. It must be called with the controller lock held and the
2354 * drv structures busy_configuring flag set. It's parameters are:
2355 *
2356 * disk = This is the disk to be deregistered
2357 * drv = This is the drive_info_struct associated with the disk to be
2358 * deregistered. It contains information about the disk used
2359 * by the driver.
2360 * clear_all = This flag determines whether or not the disk information
2361 * is going to be completely cleared out and the highest_lun
2362 * reset. Sometimes we want to clear out information about
2363 * the disk in preparation for re-adding it. In this case
2364 * the highest_lun should be left unchanged and the LunID
2365 * should not be cleared.
2366 * via_ioctl
2367 * This indicates whether we've reached this path via ioctl.
2368 * This affects the maximum usage count allowed for c0d0 to be messed with.
2369 * If this path is reached via ioctl(), then the max_usage_count will
2370 * be 1, as the process calling ioctl() has got to have the device open.
2371 * If we get here via sysfs, then the max usage count will be zero.
2372 */
2373 static int deregister_disk(ctlr_info_t *h, int drv_index,
2374 int clear_all, int via_ioctl)
2375 {
2376 int i;
2377 struct gendisk *disk;
2378 drive_info_struct *drv;
2379 int recalculate_highest_lun;
2380
2381 if (!capable(CAP_SYS_RAWIO))
2382 return -EPERM;
2383
2384 drv = h->drv[drv_index];
2385 disk = h->gendisk[drv_index];
2386
2387 /* make sure logical volume is NOT is use */
2388 if (clear_all || (h->gendisk[0] == disk)) {
2389 if (drv->usage_count > via_ioctl)
2390 return -EBUSY;
2391 } else if (drv->usage_count > 0)
2392 return -EBUSY;
2393
2394 recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2395
2396 /* invalidate the devices and deregister the disk. If it is disk
2397 * zero do not deregister it but just zero out it's values. This
2398 * allows us to delete disk zero but keep the controller registered.
2399 */
2400 if (h->gendisk[0] != disk) {
2401 struct request_queue *q = disk->queue;
2402 if (disk->flags & GENHD_FL_UP) {
2403 cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2404 del_gendisk(disk);
2405 }
2406 if (q)
2407 blk_cleanup_queue(q);
2408 /* If clear_all is set then we are deleting the logical
2409 * drive, not just refreshing its info. For drives
2410 * other than disk 0 we will call put_disk. We do not
2411 * do this for disk 0 as we need it to be able to
2412 * configure the controller.
2413 */
2414 if (clear_all){
2415 /* This isn't pretty, but we need to find the
2416 * disk in our array and NULL our the pointer.
2417 * This is so that we will call alloc_disk if
2418 * this index is used again later.
2419 */
2420 for (i=0; i < CISS_MAX_LUN; i++){
2421 if (h->gendisk[i] == disk) {
2422 h->gendisk[i] = NULL;
2423 break;
2424 }
2425 }
2426 put_disk(disk);
2427 }
2428 } else {
2429 set_capacity(disk, 0);
2430 cciss_clear_drive_info(drv);
2431 }
2432
2433 --h->num_luns;
2434
2435 /* if it was the last disk, find the new hightest lun */
2436 if (clear_all && recalculate_highest_lun) {
2437 int i, newhighest = -1;
2438 for (i = 0; i <= h->highest_lun; i++) {
2439 /* if the disk has size > 0, it is available */
2440 if (h->drv[i] && h->drv[i]->heads)
2441 newhighest = i;
2442 }
2443 h->highest_lun = newhighest;
2444 }
2445 return 0;
2446 }
2447
2448 static int fill_cmd(CommandList_struct *c, __u8 cmd, int ctlr, void *buff,
2449 size_t size, __u8 page_code, unsigned char *scsi3addr,
2450 int cmd_type)
2451 {
2452 ctlr_info_t *h = hba[ctlr];
2453 u64bit buff_dma_handle;
2454 int status = IO_OK;
2455
2456 c->cmd_type = CMD_IOCTL_PEND;
2457 c->Header.ReplyQueue = 0;
2458 if (buff != NULL) {
2459 c->Header.SGList = 1;
2460 c->Header.SGTotal = 1;
2461 } else {
2462 c->Header.SGList = 0;
2463 c->Header.SGTotal = 0;
2464 }
2465 c->Header.Tag.lower = c->busaddr;
2466 memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2467
2468 c->Request.Type.Type = cmd_type;
2469 if (cmd_type == TYPE_CMD) {
2470 switch (cmd) {
2471 case CISS_INQUIRY:
2472 /* are we trying to read a vital product page */
2473 if (page_code != 0) {
2474 c->Request.CDB[1] = 0x01;
2475 c->Request.CDB[2] = page_code;
2476 }
2477 c->Request.CDBLen = 6;
2478 c->Request.Type.Attribute = ATTR_SIMPLE;
2479 c->Request.Type.Direction = XFER_READ;
2480 c->Request.Timeout = 0;
2481 c->Request.CDB[0] = CISS_INQUIRY;
2482 c->Request.CDB[4] = size & 0xFF;
2483 break;
2484 case CISS_REPORT_LOG:
2485 case CISS_REPORT_PHYS:
2486 /* Talking to controller so It's a physical command
2487 mode = 00 target = 0. Nothing to write.
2488 */
2489 c->Request.CDBLen = 12;
2490 c->Request.Type.Attribute = ATTR_SIMPLE;
2491 c->Request.Type.Direction = XFER_READ;
2492 c->Request.Timeout = 0;
2493 c->Request.CDB[0] = cmd;
2494 c->Request.CDB[6] = (size >> 24) & 0xFF; /* MSB */
2495 c->Request.CDB[7] = (size >> 16) & 0xFF;
2496 c->Request.CDB[8] = (size >> 8) & 0xFF;
2497 c->Request.CDB[9] = size & 0xFF;
2498 break;
2499
2500 case CCISS_READ_CAPACITY:
2501 c->Request.CDBLen = 10;
2502 c->Request.Type.Attribute = ATTR_SIMPLE;
2503 c->Request.Type.Direction = XFER_READ;
2504 c->Request.Timeout = 0;
2505 c->Request.CDB[0] = cmd;
2506 break;
2507 case CCISS_READ_CAPACITY_16:
2508 c->Request.CDBLen = 16;
2509 c->Request.Type.Attribute = ATTR_SIMPLE;
2510 c->Request.Type.Direction = XFER_READ;
2511 c->Request.Timeout = 0;
2512 c->Request.CDB[0] = cmd;
2513 c->Request.CDB[1] = 0x10;
2514 c->Request.CDB[10] = (size >> 24) & 0xFF;
2515 c->Request.CDB[11] = (size >> 16) & 0xFF;
2516 c->Request.CDB[12] = (size >> 8) & 0xFF;
2517 c->Request.CDB[13] = size & 0xFF;
2518 c->Request.Timeout = 0;
2519 c->Request.CDB[0] = cmd;
2520 break;
2521 case CCISS_CACHE_FLUSH:
2522 c->Request.CDBLen = 12;
2523 c->Request.Type.Attribute = ATTR_SIMPLE;
2524 c->Request.Type.Direction = XFER_WRITE;
2525 c->Request.Timeout = 0;
2526 c->Request.CDB[0] = BMIC_WRITE;
2527 c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2528 break;
2529 case TEST_UNIT_READY:
2530 c->Request.CDBLen = 6;
2531 c->Request.Type.Attribute = ATTR_SIMPLE;
2532 c->Request.Type.Direction = XFER_NONE;
2533 c->Request.Timeout = 0;
2534 break;
2535 default:
2536 printk(KERN_WARNING
2537 "cciss%d: Unknown Command 0x%c\n", ctlr, cmd);
2538 return IO_ERROR;
2539 }
2540 } else if (cmd_type == TYPE_MSG) {
2541 switch (cmd) {
2542 case 0: /* ABORT message */
2543 c->Request.CDBLen = 12;
2544 c->Request.Type.Attribute = ATTR_SIMPLE;
2545 c->Request.Type.Direction = XFER_WRITE;
2546 c->Request.Timeout = 0;
2547 c->Request.CDB[0] = cmd; /* abort */
2548 c->Request.CDB[1] = 0; /* abort a command */
2549 /* buff contains the tag of the command to abort */
2550 memcpy(&c->Request.CDB[4], buff, 8);
2551 break;
2552 case 1: /* RESET message */
2553 c->Request.CDBLen = 16;
2554 c->Request.Type.Attribute = ATTR_SIMPLE;
2555 c->Request.Type.Direction = XFER_NONE;
2556 c->Request.Timeout = 0;
2557 memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2558 c->Request.CDB[0] = cmd; /* reset */
2559 c->Request.CDB[1] = 0x03; /* reset a target */
2560 break;
2561 case 3: /* No-Op message */
2562 c->Request.CDBLen = 1;
2563 c->Request.Type.Attribute = ATTR_SIMPLE;
2564 c->Request.Type.Direction = XFER_WRITE;
2565 c->Request.Timeout = 0;
2566 c->Request.CDB[0] = cmd;
2567 break;
2568 default:
2569 printk(KERN_WARNING
2570 "cciss%d: unknown message type %d\n", ctlr, cmd);
2571 return IO_ERROR;
2572 }
2573 } else {
2574 printk(KERN_WARNING
2575 "cciss%d: unknown command type %d\n", ctlr, cmd_type);
2576 return IO_ERROR;
2577 }
2578 /* Fill in the scatter gather information */
2579 if (size > 0) {
2580 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2581 buff, size,
2582 PCI_DMA_BIDIRECTIONAL);
2583 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2584 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2585 c->SG[0].Len = size;
2586 c->SG[0].Ext = 0; /* we are not chaining */
2587 }
2588 return status;
2589 }
2590
2591 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2592 {
2593 switch (c->err_info->ScsiStatus) {
2594 case SAM_STAT_GOOD:
2595 return IO_OK;
2596 case SAM_STAT_CHECK_CONDITION:
2597 switch (0xf & c->err_info->SenseInfo[2]) {
2598 case 0: return IO_OK; /* no sense */
2599 case 1: return IO_OK; /* recovered error */
2600 default:
2601 if (check_for_unit_attention(h, c))
2602 return IO_NEEDS_RETRY;
2603 printk(KERN_WARNING "cciss%d: cmd 0x%02x "
2604 "check condition, sense key = 0x%02x\n",
2605 h->ctlr, c->Request.CDB[0],
2606 c->err_info->SenseInfo[2]);
2607 }
2608 break;
2609 default:
2610 printk(KERN_WARNING "cciss%d: cmd 0x%02x"
2611 "scsi status = 0x%02x\n", h->ctlr,
2612 c->Request.CDB[0], c->err_info->ScsiStatus);
2613 break;
2614 }
2615 return IO_ERROR;
2616 }
2617
2618 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2619 {
2620 int return_status = IO_OK;
2621
2622 if (c->err_info->CommandStatus == CMD_SUCCESS)
2623 return IO_OK;
2624
2625 switch (c->err_info->CommandStatus) {
2626 case CMD_TARGET_STATUS:
2627 return_status = check_target_status(h, c);
2628 break;
2629 case CMD_DATA_UNDERRUN:
2630 case CMD_DATA_OVERRUN:
2631 /* expected for inquiry and report lun commands */
2632 break;
2633 case CMD_INVALID:
2634 printk(KERN_WARNING "cciss: cmd 0x%02x is "
2635 "reported invalid\n", c->Request.CDB[0]);
2636 return_status = IO_ERROR;
2637 break;
2638 case CMD_PROTOCOL_ERR:
2639 printk(KERN_WARNING "cciss: cmd 0x%02x has "
2640 "protocol error \n", c->Request.CDB[0]);
2641 return_status = IO_ERROR;
2642 break;
2643 case CMD_HARDWARE_ERR:
2644 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2645 " hardware error\n", c->Request.CDB[0]);
2646 return_status = IO_ERROR;
2647 break;
2648 case CMD_CONNECTION_LOST:
2649 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2650 "connection lost\n", c->Request.CDB[0]);
2651 return_status = IO_ERROR;
2652 break;
2653 case CMD_ABORTED:
2654 printk(KERN_WARNING "cciss: cmd 0x%02x was "
2655 "aborted\n", c->Request.CDB[0]);
2656 return_status = IO_ERROR;
2657 break;
2658 case CMD_ABORT_FAILED:
2659 printk(KERN_WARNING "cciss: cmd 0x%02x reports "
2660 "abort failed\n", c->Request.CDB[0]);
2661 return_status = IO_ERROR;
2662 break;
2663 case CMD_UNSOLICITED_ABORT:
2664 printk(KERN_WARNING
2665 "cciss%d: unsolicited abort 0x%02x\n", h->ctlr,
2666 c->Request.CDB[0]);
2667 return_status = IO_NEEDS_RETRY;
2668 break;
2669 default:
2670 printk(KERN_WARNING "cciss: cmd 0x%02x returned "
2671 "unknown status %x\n", c->Request.CDB[0],
2672 c->err_info->CommandStatus);
2673 return_status = IO_ERROR;
2674 }
2675 return return_status;
2676 }
2677
2678 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2679 int attempt_retry)
2680 {
2681 DECLARE_COMPLETION_ONSTACK(wait);
2682 u64bit buff_dma_handle;
2683 unsigned long flags;
2684 int return_status = IO_OK;
2685
2686 resend_cmd2:
2687 c->waiting = &wait;
2688 /* Put the request on the tail of the queue and send it */
2689 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2690 addQ(&h->reqQ, c);
2691 h->Qdepth++;
2692 start_io(h);
2693 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2694
2695 wait_for_completion(&wait);
2696
2697 if (c->err_info->CommandStatus == 0 || !attempt_retry)
2698 goto command_done;
2699
2700 return_status = process_sendcmd_error(h, c);
2701
2702 if (return_status == IO_NEEDS_RETRY &&
2703 c->retry_count < MAX_CMD_RETRIES) {
2704 printk(KERN_WARNING "cciss%d: retrying 0x%02x\n", h->ctlr,
2705 c->Request.CDB[0]);
2706 c->retry_count++;
2707 /* erase the old error information */
2708 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2709 return_status = IO_OK;
2710 INIT_COMPLETION(wait);
2711 goto resend_cmd2;
2712 }
2713
2714 command_done:
2715 /* unlock the buffers from DMA */
2716 buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2717 buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2718 pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2719 c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2720 return return_status;
2721 }
2722
2723 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
2724 __u8 page_code, unsigned char scsi3addr[],
2725 int cmd_type)
2726 {
2727 ctlr_info_t *h = hba[ctlr];
2728 CommandList_struct *c;
2729 int return_status;
2730
2731 c = cmd_alloc(h, 0);
2732 if (!c)
2733 return -ENOMEM;
2734 return_status = fill_cmd(c, cmd, ctlr, buff, size, page_code,
2735 scsi3addr, cmd_type);
2736 if (return_status == IO_OK)
2737 return_status = sendcmd_withirq_core(h, c, 1);
2738
2739 cmd_free(h, c, 0);
2740 return return_status;
2741 }
2742
2743 static void cciss_geometry_inquiry(int ctlr, int logvol,
2744 sector_t total_size,
2745 unsigned int block_size,
2746 InquiryData_struct *inq_buff,
2747 drive_info_struct *drv)
2748 {
2749 int return_code;
2750 unsigned long t;
2751 unsigned char scsi3addr[8];
2752
2753 memset(inq_buff, 0, sizeof(InquiryData_struct));
2754 log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2755 return_code = sendcmd_withirq(CISS_INQUIRY, ctlr, inq_buff,
2756 sizeof(*inq_buff), 0xC1, scsi3addr, TYPE_CMD);
2757 if (return_code == IO_OK) {
2758 if (inq_buff->data_byte[8] == 0xFF) {
2759 printk(KERN_WARNING
2760 "cciss: reading geometry failed, volume "
2761 "does not support reading geometry\n");
2762 drv->heads = 255;
2763 drv->sectors = 32; /* Sectors per track */
2764 drv->cylinders = total_size + 1;
2765 drv->raid_level = RAID_UNKNOWN;
2766 } else {
2767 drv->heads = inq_buff->data_byte[6];
2768 drv->sectors = inq_buff->data_byte[7];
2769 drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2770 drv->cylinders += inq_buff->data_byte[5];
2771 drv->raid_level = inq_buff->data_byte[8];
2772 }
2773 drv->block_size = block_size;
2774 drv->nr_blocks = total_size + 1;
2775 t = drv->heads * drv->sectors;
2776 if (t > 1) {
2777 sector_t real_size = total_size + 1;
2778 unsigned long rem = sector_div(real_size, t);
2779 if (rem)
2780 real_size++;
2781 drv->cylinders = real_size;
2782 }
2783 } else { /* Get geometry failed */
2784 printk(KERN_WARNING "cciss: reading geometry failed\n");
2785 }
2786 }
2787
2788 static void
2789 cciss_read_capacity(int ctlr, int logvol, sector_t *total_size,
2790 unsigned int *block_size)
2791 {
2792 ReadCapdata_struct *buf;
2793 int return_code;
2794 unsigned char scsi3addr[8];
2795
2796 buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2797 if (!buf) {
2798 printk(KERN_WARNING "cciss: out of memory\n");
2799 return;
2800 }
2801
2802 log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2803 return_code = sendcmd_withirq(CCISS_READ_CAPACITY, ctlr, buf,
2804 sizeof(ReadCapdata_struct), 0, scsi3addr, TYPE_CMD);
2805 if (return_code == IO_OK) {
2806 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2807 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2808 } else { /* read capacity command failed */
2809 printk(KERN_WARNING "cciss: read capacity failed\n");
2810 *total_size = 0;
2811 *block_size = BLOCK_SIZE;
2812 }
2813 kfree(buf);
2814 }
2815
2816 static void cciss_read_capacity_16(int ctlr, int logvol,
2817 sector_t *total_size, unsigned int *block_size)
2818 {
2819 ReadCapdata_struct_16 *buf;
2820 int return_code;
2821 unsigned char scsi3addr[8];
2822
2823 buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2824 if (!buf) {
2825 printk(KERN_WARNING "cciss: out of memory\n");
2826 return;
2827 }
2828
2829 log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2830 return_code = sendcmd_withirq(CCISS_READ_CAPACITY_16,
2831 ctlr, buf, sizeof(ReadCapdata_struct_16),
2832 0, scsi3addr, TYPE_CMD);
2833 if (return_code == IO_OK) {
2834 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2835 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2836 } else { /* read capacity command failed */
2837 printk(KERN_WARNING "cciss: read capacity failed\n");
2838 *total_size = 0;
2839 *block_size = BLOCK_SIZE;
2840 }
2841 printk(KERN_INFO " blocks= %llu block_size= %d\n",
2842 (unsigned long long)*total_size+1, *block_size);
2843 kfree(buf);
2844 }
2845
2846 static int cciss_revalidate(struct gendisk *disk)
2847 {
2848 ctlr_info_t *h = get_host(disk);
2849 drive_info_struct *drv = get_drv(disk);
2850 int logvol;
2851 int FOUND = 0;
2852 unsigned int block_size;
2853 sector_t total_size;
2854 InquiryData_struct *inq_buff = NULL;
2855
2856 for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
2857 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2858 sizeof(drv->LunID)) == 0) {
2859 FOUND = 1;
2860 break;
2861 }
2862 }
2863
2864 if (!FOUND)
2865 return 1;
2866
2867 inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2868 if (inq_buff == NULL) {
2869 printk(KERN_WARNING "cciss: out of memory\n");
2870 return 1;
2871 }
2872 if (h->cciss_read == CCISS_READ_10) {
2873 cciss_read_capacity(h->ctlr, logvol,
2874 &total_size, &block_size);
2875 } else {
2876 cciss_read_capacity_16(h->ctlr, logvol,
2877 &total_size, &block_size);
2878 }
2879 cciss_geometry_inquiry(h->ctlr, logvol, total_size, block_size,
2880 inq_buff, drv);
2881
2882 blk_queue_logical_block_size(drv->queue, drv->block_size);
2883 set_capacity(disk, drv->nr_blocks);
2884
2885 kfree(inq_buff);
2886 return 0;
2887 }
2888
2889 /*
2890 * Map (physical) PCI mem into (virtual) kernel space
2891 */
2892 static void __iomem *remap_pci_mem(ulong base, ulong size)
2893 {
2894 ulong page_base = ((ulong) base) & PAGE_MASK;
2895 ulong page_offs = ((ulong) base) - page_base;
2896 void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2897
2898 return page_remapped ? (page_remapped + page_offs) : NULL;
2899 }
2900
2901 /*
2902 * Takes jobs of the Q and sends them to the hardware, then puts it on
2903 * the Q to wait for completion.
2904 */
2905 static void start_io(ctlr_info_t *h)
2906 {
2907 CommandList_struct *c;
2908
2909 while (!hlist_empty(&h->reqQ)) {
2910 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
2911 /* can't do anything if fifo is full */
2912 if ((h->access.fifo_full(h))) {
2913 printk(KERN_WARNING "cciss: fifo full\n");
2914 break;
2915 }
2916
2917 /* Get the first entry from the Request Q */
2918 removeQ(c);
2919 h->Qdepth--;
2920
2921 /* Tell the controller execute command */
2922 h->access.submit_command(h, c);
2923
2924 /* Put job onto the completed Q */
2925 addQ(&h->cmpQ, c);
2926 }
2927 }
2928
2929 /* Assumes that CCISS_LOCK(h->ctlr) is held. */
2930 /* Zeros out the error record and then resends the command back */
2931 /* to the controller */
2932 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2933 {
2934 /* erase the old error information */
2935 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2936
2937 /* add it to software queue and then send it to the controller */
2938 addQ(&h->reqQ, c);
2939 h->Qdepth++;
2940 if (h->Qdepth > h->maxQsinceinit)
2941 h->maxQsinceinit = h->Qdepth;
2942
2943 start_io(h);
2944 }
2945
2946 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
2947 unsigned int msg_byte, unsigned int host_byte,
2948 unsigned int driver_byte)
2949 {
2950 /* inverse of macros in scsi.h */
2951 return (scsi_status_byte & 0xff) |
2952 ((msg_byte & 0xff) << 8) |
2953 ((host_byte & 0xff) << 16) |
2954 ((driver_byte & 0xff) << 24);
2955 }
2956
2957 static inline int evaluate_target_status(ctlr_info_t *h,
2958 CommandList_struct *cmd, int *retry_cmd)
2959 {
2960 unsigned char sense_key;
2961 unsigned char status_byte, msg_byte, host_byte, driver_byte;
2962 int error_value;
2963
2964 *retry_cmd = 0;
2965 /* If we get in here, it means we got "target status", that is, scsi status */
2966 status_byte = cmd->err_info->ScsiStatus;
2967 driver_byte = DRIVER_OK;
2968 msg_byte = cmd->err_info->CommandStatus; /* correct? seems too device specific */
2969
2970 if (blk_pc_request(cmd->rq))
2971 host_byte = DID_PASSTHROUGH;
2972 else
2973 host_byte = DID_OK;
2974
2975 error_value = make_status_bytes(status_byte, msg_byte,
2976 host_byte, driver_byte);
2977
2978 if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
2979 if (!blk_pc_request(cmd->rq))
2980 printk(KERN_WARNING "cciss: cmd %p "
2981 "has SCSI Status 0x%x\n",
2982 cmd, cmd->err_info->ScsiStatus);
2983 return error_value;
2984 }
2985
2986 /* check the sense key */
2987 sense_key = 0xf & cmd->err_info->SenseInfo[2];
2988 /* no status or recovered error */
2989 if (((sense_key == 0x0) || (sense_key == 0x1)) && !blk_pc_request(cmd->rq))
2990 error_value = 0;
2991
2992 if (check_for_unit_attention(h, cmd)) {
2993 *retry_cmd = !blk_pc_request(cmd->rq);
2994 return 0;
2995 }
2996
2997 if (!blk_pc_request(cmd->rq)) { /* Not SG_IO or similar? */
2998 if (error_value != 0)
2999 printk(KERN_WARNING "cciss: cmd %p has CHECK CONDITION"
3000 " sense key = 0x%x\n", cmd, sense_key);
3001 return error_value;
3002 }
3003
3004 /* SG_IO or similar, copy sense data back */
3005 if (cmd->rq->sense) {
3006 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
3007 cmd->rq->sense_len = cmd->err_info->SenseLen;
3008 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
3009 cmd->rq->sense_len);
3010 } else
3011 cmd->rq->sense_len = 0;
3012
3013 return error_value;
3014 }
3015
3016 /* checks the status of the job and calls complete buffers to mark all
3017 * buffers for the completed job. Note that this function does not need
3018 * to hold the hba/queue lock.
3019 */
3020 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
3021 int timeout)
3022 {
3023 int retry_cmd = 0;
3024 struct request *rq = cmd->rq;
3025
3026 rq->errors = 0;
3027
3028 if (timeout)
3029 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
3030
3031 if (cmd->err_info->CommandStatus == 0) /* no error has occurred */
3032 goto after_error_processing;
3033
3034 switch (cmd->err_info->CommandStatus) {
3035 case CMD_TARGET_STATUS:
3036 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
3037 break;
3038 case CMD_DATA_UNDERRUN:
3039 if (blk_fs_request(cmd->rq)) {
3040 printk(KERN_WARNING "cciss: cmd %p has"
3041 " completed with data underrun "
3042 "reported\n", cmd);
3043 cmd->rq->resid_len = cmd->err_info->ResidualCnt;
3044 }
3045 break;
3046 case CMD_DATA_OVERRUN:
3047 if (blk_fs_request(cmd->rq))
3048 printk(KERN_WARNING "cciss: cmd %p has"
3049 " completed with data overrun "
3050 "reported\n", cmd);
3051 break;
3052 case CMD_INVALID:
3053 printk(KERN_WARNING "cciss: cmd %p is "
3054 "reported invalid\n", cmd);
3055 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3056 cmd->err_info->CommandStatus, DRIVER_OK,
3057 blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3058 break;
3059 case CMD_PROTOCOL_ERR:
3060 printk(KERN_WARNING "cciss: cmd %p has "
3061 "protocol error \n", cmd);
3062 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3063 cmd->err_info->CommandStatus, DRIVER_OK,
3064 blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3065 break;
3066 case CMD_HARDWARE_ERR:
3067 printk(KERN_WARNING "cciss: cmd %p had "
3068 " hardware error\n", cmd);
3069 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3070 cmd->err_info->CommandStatus, DRIVER_OK,
3071 blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3072 break;
3073 case CMD_CONNECTION_LOST:
3074 printk(KERN_WARNING "cciss: cmd %p had "
3075 "connection lost\n", cmd);
3076 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3077 cmd->err_info->CommandStatus, DRIVER_OK,
3078 blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3079 break;
3080 case CMD_ABORTED:
3081 printk(KERN_WARNING "cciss: cmd %p was "
3082 "aborted\n", cmd);
3083 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3084 cmd->err_info->CommandStatus, DRIVER_OK,
3085 blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
3086 break;
3087 case CMD_ABORT_FAILED:
3088 printk(KERN_WARNING "cciss: cmd %p reports "
3089 "abort failed\n", cmd);
3090 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3091 cmd->err_info->CommandStatus, DRIVER_OK,
3092 blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3093 break;
3094 case CMD_UNSOLICITED_ABORT:
3095 printk(KERN_WARNING "cciss%d: unsolicited "
3096 "abort %p\n", h->ctlr, cmd);
3097 if (cmd->retry_count < MAX_CMD_RETRIES) {
3098 retry_cmd = 1;
3099 printk(KERN_WARNING
3100 "cciss%d: retrying %p\n", h->ctlr, cmd);
3101 cmd->retry_count++;
3102 } else
3103 printk(KERN_WARNING
3104 "cciss%d: %p retried too "
3105 "many times\n", h->ctlr, cmd);
3106 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3107 cmd->err_info->CommandStatus, DRIVER_OK,
3108 blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
3109 break;
3110 case CMD_TIMEOUT:
3111 printk(KERN_WARNING "cciss: cmd %p timedout\n", cmd);
3112 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3113 cmd->err_info->CommandStatus, DRIVER_OK,
3114 blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3115 break;
3116 default:
3117 printk(KERN_WARNING "cciss: cmd %p returned "
3118 "unknown status %x\n", cmd,
3119 cmd->err_info->CommandStatus);
3120 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3121 cmd->err_info->CommandStatus, DRIVER_OK,
3122 blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3123 }
3124
3125 after_error_processing:
3126
3127 /* We need to return this command */
3128 if (retry_cmd) {
3129 resend_cciss_cmd(h, cmd);
3130 return;
3131 }
3132 cmd->rq->completion_data = cmd;
3133 blk_complete_request(cmd->rq);
3134 }
3135
3136 /*
3137 * Get a request and submit it to the controller.
3138 */
3139 static void do_cciss_request(struct request_queue *q)
3140 {
3141 ctlr_info_t *h = q->queuedata;
3142 CommandList_struct *c;
3143 sector_t start_blk;
3144 int seg;
3145 struct request *creq;
3146 u64bit temp64;
3147 struct scatterlist *tmp_sg;
3148 SGDescriptor_struct *curr_sg;
3149 drive_info_struct *drv;
3150 int i, dir;
3151 int sg_index = 0;
3152 int chained = 0;
3153
3154 /* We call start_io here in case there is a command waiting on the
3155 * queue that has not been sent.
3156 */
3157 if (blk_queue_plugged(q))
3158 goto startio;
3159
3160 queue:
3161 creq = blk_peek_request(q);
3162 if (!creq)
3163 goto startio;
3164
3165 BUG_ON(creq->nr_phys_segments > h->maxsgentries);
3166
3167 if ((c = cmd_alloc(h, 1)) == NULL)
3168 goto full;
3169
3170 blk_start_request(creq);
3171
3172 tmp_sg = h->scatter_list[c->cmdindex];
3173 spin_unlock_irq(q->queue_lock);
3174
3175 c->cmd_type = CMD_RWREQ;
3176 c->rq = creq;
3177
3178 /* fill in the request */
3179 drv = creq->rq_disk->private_data;
3180 c->Header.ReplyQueue = 0; /* unused in simple mode */
3181 /* got command from pool, so use the command block index instead */
3182 /* for direct lookups. */
3183 /* The first 2 bits are reserved for controller error reporting. */
3184 c->Header.Tag.lower = (c->cmdindex << 3);
3185 c->Header.Tag.lower |= 0x04; /* flag for direct lookup. */
3186 memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3187 c->Request.CDBLen = 10; /* 12 byte commands not in FW yet; */
3188 c->Request.Type.Type = TYPE_CMD; /* It is a command. */
3189 c->Request.Type.Attribute = ATTR_SIMPLE;
3190 c->Request.Type.Direction =
3191 (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3192 c->Request.Timeout = 0; /* Don't time out */
3193 c->Request.CDB[0] =
3194 (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3195 start_blk = blk_rq_pos(creq);
3196 #ifdef CCISS_DEBUG
3197 printk(KERN_DEBUG "ciss: sector =%d nr_sectors=%d\n",
3198 (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3199 #endif /* CCISS_DEBUG */
3200
3201 sg_init_table(tmp_sg, h->maxsgentries);
3202 seg = blk_rq_map_sg(q, creq, tmp_sg);
3203
3204 /* get the DMA records for the setup */
3205 if (c->Request.Type.Direction == XFER_READ)
3206 dir = PCI_DMA_FROMDEVICE;
3207 else
3208 dir = PCI_DMA_TODEVICE;
3209
3210 curr_sg = c->SG;
3211 sg_index = 0;
3212 chained = 0;
3213
3214 for (i = 0; i < seg; i++) {
3215 if (((sg_index+1) == (h->max_cmd_sgentries)) &&
3216 !chained && ((seg - i) > 1)) {
3217 /* Point to next chain block. */
3218 curr_sg = h->cmd_sg_list[c->cmdindex];
3219 sg_index = 0;
3220 chained = 1;
3221 }
3222 curr_sg[sg_index].Len = tmp_sg[i].length;
3223 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3224 tmp_sg[i].offset,
3225 tmp_sg[i].length, dir);
3226 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3227 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3228 curr_sg[sg_index].Ext = 0; /* we are not chaining */
3229 ++sg_index;
3230 }
3231 if (chained)
3232 cciss_map_sg_chain_block(h, c, h->cmd_sg_list[c->cmdindex],
3233 (seg - (h->max_cmd_sgentries - 1)) *
3234 sizeof(SGDescriptor_struct));
3235
3236 /* track how many SG entries we are using */
3237 if (seg > h->maxSG)
3238 h->maxSG = seg;
3239
3240 #ifdef CCISS_DEBUG
3241 printk(KERN_DEBUG "cciss: Submitting %ld sectors in %d segments "
3242 "chained[%d]\n",
3243 blk_rq_sectors(creq), seg, chained);
3244 #endif /* CCISS_DEBUG */
3245
3246 c->Header.SGList = c->Header.SGTotal = seg + chained;
3247 if (seg > h->max_cmd_sgentries)
3248 c->Header.SGList = h->max_cmd_sgentries;
3249
3250 if (likely(blk_fs_request(creq))) {
3251 if(h->cciss_read == CCISS_READ_10) {
3252 c->Request.CDB[1] = 0;
3253 c->Request.CDB[2] = (start_blk >> 24) & 0xff; /* MSB */
3254 c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3255 c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3256 c->Request.CDB[5] = start_blk & 0xff;
3257 c->Request.CDB[6] = 0; /* (sect >> 24) & 0xff; MSB */
3258 c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3259 c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3260 c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3261 } else {
3262 u32 upper32 = upper_32_bits(start_blk);
3263
3264 c->Request.CDBLen = 16;
3265 c->Request.CDB[1]= 0;
3266 c->Request.CDB[2]= (upper32 >> 24) & 0xff; /* MSB */
3267 c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3268 c->Request.CDB[4]= (upper32 >> 8) & 0xff;
3269 c->Request.CDB[5]= upper32 & 0xff;
3270 c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3271 c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3272 c->Request.CDB[8]= (start_blk >> 8) & 0xff;
3273 c->Request.CDB[9]= start_blk & 0xff;
3274 c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3275 c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3276 c->Request.CDB[12]= (blk_rq_sectors(creq) >> 8) & 0xff;
3277 c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3278 c->Request.CDB[14] = c->Request.CDB[15] = 0;
3279 }
3280 } else if (blk_pc_request(creq)) {
3281 c->Request.CDBLen = creq->cmd_len;
3282 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3283 } else {
3284 printk(KERN_WARNING "cciss%d: bad request type %d\n", h->ctlr, creq->cmd_type);
3285 BUG();
3286 }
3287
3288 spin_lock_irq(q->queue_lock);
3289
3290 addQ(&h->reqQ, c);
3291 h->Qdepth++;
3292 if (h->Qdepth > h->maxQsinceinit)
3293 h->maxQsinceinit = h->Qdepth;
3294
3295 goto queue;
3296 full:
3297 blk_stop_queue(q);
3298 startio:
3299 /* We will already have the driver lock here so not need
3300 * to lock it.
3301 */
3302 start_io(h);
3303 }
3304
3305 static inline unsigned long get_next_completion(ctlr_info_t *h)
3306 {
3307 return h->access.command_completed(h);
3308 }
3309
3310 static inline int interrupt_pending(ctlr_info_t *h)
3311 {
3312 return h->access.intr_pending(h);
3313 }
3314
3315 static inline long interrupt_not_for_us(ctlr_info_t *h)
3316 {
3317 return (((h->access.intr_pending(h) == 0) ||
3318 (h->interrupts_enabled == 0)));
3319 }
3320
3321 static irqreturn_t do_cciss_intr(int irq, void *dev_id)
3322 {
3323 ctlr_info_t *h = dev_id;
3324 CommandList_struct *c;
3325 unsigned long flags;
3326 __u32 a, a1, a2;
3327
3328 if (interrupt_not_for_us(h))
3329 return IRQ_NONE;
3330 /*
3331 * If there are completed commands in the completion queue,
3332 * we had better do something about it.
3333 */
3334 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
3335 while (interrupt_pending(h)) {
3336 while ((a = get_next_completion(h)) != FIFO_EMPTY) {
3337 a1 = a;
3338 if ((a & 0x04)) {
3339 a2 = (a >> 3);
3340 if (a2 >= h->nr_cmds) {
3341 printk(KERN_WARNING
3342 "cciss: controller cciss%d failed, stopping.\n",
3343 h->ctlr);
3344 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
3345 fail_all_cmds(h->ctlr);
3346 return IRQ_HANDLED;
3347 }
3348
3349 c = h->cmd_pool + a2;
3350 a = c->busaddr;
3351
3352 } else {
3353 struct hlist_node *tmp;
3354
3355 a &= ~3;
3356 c = NULL;
3357 hlist_for_each_entry(c, tmp, &h->cmpQ, list) {
3358 if (c->busaddr == a)
3359 break;
3360 }
3361 }
3362 /*
3363 * If we've found the command, take it off the
3364 * completion Q and free it
3365 */
3366 if (c && c->busaddr == a) {
3367 removeQ(c);
3368 if (c->cmd_type == CMD_RWREQ) {
3369 complete_command(h, c, 0);
3370 } else if (c->cmd_type == CMD_IOCTL_PEND) {
3371 complete(c->waiting);
3372 }
3373 # ifdef CONFIG_CISS_SCSI_TAPE
3374 else if (c->cmd_type == CMD_SCSI)
3375 complete_scsi_command(c, 0, a1);
3376 # endif
3377 continue;
3378 }
3379 }
3380 }
3381
3382 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
3383 return IRQ_HANDLED;
3384 }
3385
3386 /**
3387 * add_to_scan_list() - add controller to rescan queue
3388 * @h: Pointer to the controller.
3389 *
3390 * Adds the controller to the rescan queue if not already on the queue.
3391 *
3392 * returns 1 if added to the queue, 0 if skipped (could be on the
3393 * queue already, or the controller could be initializing or shutting
3394 * down).
3395 **/
3396 static int add_to_scan_list(struct ctlr_info *h)
3397 {
3398 struct ctlr_info *test_h;
3399 int found = 0;
3400 int ret = 0;
3401
3402 if (h->busy_initializing)
3403 return 0;
3404
3405 if (!mutex_trylock(&h->busy_shutting_down))
3406 return 0;
3407
3408 mutex_lock(&scan_mutex);
3409 list_for_each_entry(test_h, &scan_q, scan_list) {
3410 if (test_h == h) {
3411 found = 1;
3412 break;
3413 }
3414 }
3415 if (!found && !h->busy_scanning) {
3416 INIT_COMPLETION(h->scan_wait);
3417 list_add_tail(&h->scan_list, &scan_q);
3418 ret = 1;
3419 }
3420 mutex_unlock(&scan_mutex);
3421 mutex_unlock(&h->busy_shutting_down);
3422
3423 return ret;
3424 }
3425
3426 /**
3427 * remove_from_scan_list() - remove controller from rescan queue
3428 * @h: Pointer to the controller.
3429 *
3430 * Removes the controller from the rescan queue if present. Blocks if
3431 * the controller is currently conducting a rescan. The controller
3432 * can be in one of three states:
3433 * 1. Doesn't need a scan
3434 * 2. On the scan list, but not scanning yet (we remove it)
3435 * 3. Busy scanning (and not on the list). In this case we want to wait for
3436 * the scan to complete to make sure the scanning thread for this
3437 * controller is completely idle.
3438 **/
3439 static void remove_from_scan_list(struct ctlr_info *h)
3440 {
3441 struct ctlr_info *test_h, *tmp_h;
3442
3443 mutex_lock(&scan_mutex);
3444 list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3445 if (test_h == h) { /* state 2. */
3446 list_del(&h->scan_list);
3447 complete_all(&h->scan_wait);
3448 mutex_unlock(&scan_mutex);
3449 return;
3450 }
3451 }
3452 if (h->busy_scanning) { /* state 3. */
3453 mutex_unlock(&scan_mutex);
3454 wait_for_completion(&h->scan_wait);
3455 } else { /* state 1, nothing to do. */
3456 mutex_unlock(&scan_mutex);
3457 }
3458 }
3459
3460 /**
3461 * scan_thread() - kernel thread used to rescan controllers
3462 * @data: Ignored.
3463 *
3464 * A kernel thread used scan for drive topology changes on
3465 * controllers. The thread processes only one controller at a time
3466 * using a queue. Controllers are added to the queue using
3467 * add_to_scan_list() and removed from the queue either after done
3468 * processing or using remove_from_scan_list().
3469 *
3470 * returns 0.
3471 **/
3472 static int scan_thread(void *data)
3473 {
3474 struct ctlr_info *h;
3475
3476 while (1) {
3477 set_current_state(TASK_INTERRUPTIBLE);
3478 schedule();
3479 if (kthread_should_stop())
3480 break;
3481
3482 while (1) {
3483 mutex_lock(&scan_mutex);
3484 if (list_empty(&scan_q)) {
3485 mutex_unlock(&scan_mutex);
3486 break;
3487 }
3488
3489 h = list_entry(scan_q.next,
3490 struct ctlr_info,
3491 scan_list);
3492 list_del(&h->scan_list);
3493 h->busy_scanning = 1;
3494 mutex_unlock(&scan_mutex);
3495
3496 rebuild_lun_table(h, 0, 0);
3497 complete_all(&h->scan_wait);
3498 mutex_lock(&scan_mutex);
3499 h->busy_scanning = 0;
3500 mutex_unlock(&scan_mutex);
3501 }
3502 }
3503
3504 return 0;
3505 }
3506
3507 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3508 {
3509 if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3510 return 0;
3511
3512 switch (c->err_info->SenseInfo[12]) {
3513 case STATE_CHANGED:
3514 printk(KERN_WARNING "cciss%d: a state change "
3515 "detected, command retried\n", h->ctlr);
3516 return 1;
3517 break;
3518 case LUN_FAILED:
3519 printk(KERN_WARNING "cciss%d: LUN failure "
3520 "detected, action required\n", h->ctlr);
3521 return 1;
3522 break;
3523 case REPORT_LUNS_CHANGED:
3524 printk(KERN_WARNING "cciss%d: report LUN data "
3525 "changed\n", h->ctlr);
3526 /*
3527 * Here, we could call add_to_scan_list and wake up the scan thread,
3528 * except that it's quite likely that we will get more than one
3529 * REPORT_LUNS_CHANGED condition in quick succession, which means
3530 * that those which occur after the first one will likely happen
3531 * *during* the scan_thread's rescan. And the rescan code is not
3532 * robust enough to restart in the middle, undoing what it has already
3533 * done, and it's not clear that it's even possible to do this, since
3534 * part of what it does is notify the block layer, which starts
3535 * doing it's own i/o to read partition tables and so on, and the
3536 * driver doesn't have visibility to know what might need undoing.
3537 * In any event, if possible, it is horribly complicated to get right
3538 * so we just don't do it for now.
3539 *
3540 * Note: this REPORT_LUNS_CHANGED condition only occurs on the MSA2012.
3541 */
3542 return 1;
3543 break;
3544 case POWER_OR_RESET:
3545 printk(KERN_WARNING "cciss%d: a power on "
3546 "or device reset detected\n", h->ctlr);
3547 return 1;
3548 break;
3549 case UNIT_ATTENTION_CLEARED:
3550 printk(KERN_WARNING "cciss%d: unit attention "
3551 "cleared by another initiator\n", h->ctlr);
3552 return 1;
3553 break;
3554 default:
3555 printk(KERN_WARNING "cciss%d: unknown "
3556 "unit attention detected\n", h->ctlr);
3557 return 1;
3558 }
3559 }
3560
3561 /*
3562 * We cannot read the structure directly, for portability we must use
3563 * the io functions.
3564 * This is for debug only.
3565 */
3566 #ifdef CCISS_DEBUG
3567 static void print_cfg_table(CfgTable_struct *tb)
3568 {
3569 int i;
3570 char temp_name[17];
3571
3572 printk("Controller Configuration information\n");
3573 printk("------------------------------------\n");
3574 for (i = 0; i < 4; i++)
3575 temp_name[i] = readb(&(tb->Signature[i]));
3576 temp_name[4] = '\0';
3577 printk(" Signature = %s\n", temp_name);
3578 printk(" Spec Number = %d\n", readl(&(tb->SpecValence)));
3579 printk(" Transport methods supported = 0x%x\n",
3580 readl(&(tb->TransportSupport)));
3581 printk(" Transport methods active = 0x%x\n",
3582 readl(&(tb->TransportActive)));
3583 printk(" Requested transport Method = 0x%x\n",
3584 readl(&(tb->HostWrite.TransportRequest)));
3585 printk(" Coalesce Interrupt Delay = 0x%x\n",
3586 readl(&(tb->HostWrite.CoalIntDelay)));
3587 printk(" Coalesce Interrupt Count = 0x%x\n",
3588 readl(&(tb->HostWrite.CoalIntCount)));
3589 printk(" Max outstanding commands = 0x%d\n",
3590 readl(&(tb->CmdsOutMax)));
3591 printk(" Bus Types = 0x%x\n", readl(&(tb->BusTypes)));
3592 for (i = 0; i < 16; i++)
3593 temp_name[i] = readb(&(tb->ServerName[i]));
3594 temp_name[16] = '\0';
3595 printk(" Server Name = %s\n", temp_name);
3596 printk(" Heartbeat Counter = 0x%x\n\n\n", readl(&(tb->HeartBeat)));
3597 }
3598 #endif /* CCISS_DEBUG */
3599
3600 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3601 {
3602 int i, offset, mem_type, bar_type;
3603 if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3604 return 0;
3605 offset = 0;
3606 for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3607 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3608 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3609 offset += 4;
3610 else {
3611 mem_type = pci_resource_flags(pdev, i) &
3612 PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3613 switch (mem_type) {
3614 case PCI_BASE_ADDRESS_MEM_TYPE_32:
3615 case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3616 offset += 4; /* 32 bit */
3617 break;
3618 case PCI_BASE_ADDRESS_MEM_TYPE_64:
3619 offset += 8;
3620 break;
3621 default: /* reserved in PCI 2.2 */
3622 printk(KERN_WARNING
3623 "Base address is invalid\n");
3624 return -1;
3625 break;
3626 }
3627 }
3628 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3629 return i + 1;
3630 }
3631 return -1;
3632 }
3633
3634 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
3635 * controllers that are capable. If not, we use IO-APIC mode.
3636 */
3637
3638 static void __devinit cciss_interrupt_mode(ctlr_info_t *c,
3639 struct pci_dev *pdev, __u32 board_id)
3640 {
3641 #ifdef CONFIG_PCI_MSI
3642 int err;
3643 struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
3644 {0, 2}, {0, 3}
3645 };
3646
3647 /* Some boards advertise MSI but don't really support it */
3648 if ((board_id == 0x40700E11) ||
3649 (board_id == 0x40800E11) ||
3650 (board_id == 0x40820E11) || (board_id == 0x40830E11))
3651 goto default_int_mode;
3652
3653 if (pci_find_capability(pdev, PCI_CAP_ID_MSIX)) {
3654 err = pci_enable_msix(pdev, cciss_msix_entries, 4);
3655 if (!err) {
3656 c->intr[0] = cciss_msix_entries[0].vector;
3657 c->intr[1] = cciss_msix_entries[1].vector;
3658 c->intr[2] = cciss_msix_entries[2].vector;
3659 c->intr[3] = cciss_msix_entries[3].vector;
3660 c->msix_vector = 1;
3661 return;
3662 }
3663 if (err > 0) {
3664 printk(KERN_WARNING "cciss: only %d MSI-X vectors "
3665 "available\n", err);
3666 goto default_int_mode;
3667 } else {
3668 printk(KERN_WARNING "cciss: MSI-X init failed %d\n",
3669 err);
3670 goto default_int_mode;
3671 }
3672 }
3673 if (pci_find_capability(pdev, PCI_CAP_ID_MSI)) {
3674 if (!pci_enable_msi(pdev)) {
3675 c->msi_vector = 1;
3676 } else {
3677 printk(KERN_WARNING "cciss: MSI init failed\n");
3678 }
3679 }
3680 default_int_mode:
3681 #endif /* CONFIG_PCI_MSI */
3682 /* if we get here we're going to use the default interrupt mode */
3683 c->intr[SIMPLE_MODE_INT] = pdev->irq;
3684 return;
3685 }
3686
3687 static int __devinit cciss_pci_init(ctlr_info_t *c, struct pci_dev *pdev)
3688 {
3689 ushort subsystem_vendor_id, subsystem_device_id, command;
3690 __u32 board_id, scratchpad = 0;
3691 __u64 cfg_offset;
3692 __u32 cfg_base_addr;
3693 __u64 cfg_base_addr_index;
3694 int i, prod_index, err;
3695
3696 subsystem_vendor_id = pdev->subsystem_vendor;
3697 subsystem_device_id = pdev->subsystem_device;
3698 board_id = (((__u32) (subsystem_device_id << 16) & 0xffff0000) |
3699 subsystem_vendor_id);
3700
3701 for (i = 0; i < ARRAY_SIZE(products); i++) {
3702 /* Stand aside for hpsa driver on request */
3703 if (cciss_allow_hpsa && products[i].board_id == HPSA_BOUNDARY)
3704 return -ENODEV;
3705 if (board_id == products[i].board_id)
3706 break;
3707 }
3708 prod_index = i;
3709 if (prod_index == ARRAY_SIZE(products)) {
3710 dev_warn(&pdev->dev,
3711 "unrecognized board ID: 0x%08lx, ignoring.\n",
3712 (unsigned long) board_id);
3713 return -ENODEV;
3714 }
3715
3716 /* check to see if controller has been disabled */
3717 /* BEFORE trying to enable it */
3718 (void)pci_read_config_word(pdev, PCI_COMMAND, &command);
3719 if (!(command & 0x02)) {
3720 printk(KERN_WARNING
3721 "cciss: controller appears to be disabled\n");
3722 return -ENODEV;
3723 }
3724
3725 err = pci_enable_device(pdev);
3726 if (err) {
3727 printk(KERN_ERR "cciss: Unable to Enable PCI device\n");
3728 return err;
3729 }
3730
3731 err = pci_request_regions(pdev, "cciss");
3732 if (err) {
3733 printk(KERN_ERR "cciss: Cannot obtain PCI resources, "
3734 "aborting\n");
3735 return err;
3736 }
3737
3738 #ifdef CCISS_DEBUG
3739 printk("command = %x\n", command);
3740 printk("irq = %x\n", pdev->irq);
3741 printk("board_id = %x\n", board_id);
3742 #endif /* CCISS_DEBUG */
3743
3744 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
3745 * else we use the IO-APIC interrupt assigned to us by system ROM.
3746 */
3747 cciss_interrupt_mode(c, pdev, board_id);
3748
3749 /* find the memory BAR */
3750 for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3751 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM)
3752 break;
3753 }
3754 if (i == DEVICE_COUNT_RESOURCE) {
3755 printk(KERN_WARNING "cciss: No memory BAR found\n");
3756 err = -ENODEV;
3757 goto err_out_free_res;
3758 }
3759
3760 c->paddr = pci_resource_start(pdev, i); /* addressing mode bits
3761 * already removed
3762 */
3763
3764 #ifdef CCISS_DEBUG
3765 printk("address 0 = %lx\n", c->paddr);
3766 #endif /* CCISS_DEBUG */
3767 c->vaddr = remap_pci_mem(c->paddr, 0x250);
3768
3769 /* Wait for the board to become ready. (PCI hotplug needs this.)
3770 * We poll for up to 120 secs, once per 100ms. */
3771 for (i = 0; i < 1200; i++) {
3772 scratchpad = readl(c->vaddr + SA5_SCRATCHPAD_OFFSET);
3773 if (scratchpad == CCISS_FIRMWARE_READY)
3774 break;
3775 set_current_state(TASK_INTERRUPTIBLE);
3776 schedule_timeout(msecs_to_jiffies(100)); /* wait 100ms */
3777 }
3778 if (scratchpad != CCISS_FIRMWARE_READY) {
3779 printk(KERN_WARNING "cciss: Board not ready. Timed out.\n");
3780 err = -ENODEV;
3781 goto err_out_free_res;
3782 }
3783
3784 /* get the address index number */
3785 cfg_base_addr = readl(c->vaddr + SA5_CTCFG_OFFSET);
3786 cfg_base_addr &= (__u32) 0x0000ffff;
3787 #ifdef CCISS_DEBUG
3788 printk("cfg base address = %x\n", cfg_base_addr);
3789 #endif /* CCISS_DEBUG */
3790 cfg_base_addr_index = find_PCI_BAR_index(pdev, cfg_base_addr);
3791 #ifdef CCISS_DEBUG
3792 printk("cfg base address index = %llx\n",
3793 (unsigned long long)cfg_base_addr_index);
3794 #endif /* CCISS_DEBUG */
3795 if (cfg_base_addr_index == -1) {
3796 printk(KERN_WARNING "cciss: Cannot find cfg_base_addr_index\n");
3797 err = -ENODEV;
3798 goto err_out_free_res;
3799 }
3800
3801 cfg_offset = readl(c->vaddr + SA5_CTMEM_OFFSET);
3802 #ifdef CCISS_DEBUG
3803 printk("cfg offset = %llx\n", (unsigned long long)cfg_offset);
3804 #endif /* CCISS_DEBUG */
3805 c->cfgtable = remap_pci_mem(pci_resource_start(pdev,
3806 cfg_base_addr_index) +
3807 cfg_offset, sizeof(CfgTable_struct));
3808 c->board_id = board_id;
3809
3810 #ifdef CCISS_DEBUG
3811 print_cfg_table(c->cfgtable);
3812 #endif /* CCISS_DEBUG */
3813
3814 /* Some controllers support Zero Memory Raid (ZMR).
3815 * When configured in ZMR mode the number of supported
3816 * commands drops to 64. So instead of just setting an
3817 * arbitrary value we make the driver a little smarter.
3818 * We read the config table to tell us how many commands
3819 * are supported on the controller then subtract 4 to
3820 * leave a little room for ioctl calls.
3821 */
3822 c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
3823 c->maxsgentries = readl(&(c->cfgtable->MaxSGElements));
3824
3825 /*
3826 * Limit native command to 32 s/g elements to save dma'able memory.
3827 * Howvever spec says if 0, use 31
3828 */
3829
3830 c->max_cmd_sgentries = 31;
3831 if (c->maxsgentries > 512) {
3832 c->max_cmd_sgentries = 32;
3833 c->chainsize = c->maxsgentries - c->max_cmd_sgentries + 1;
3834 c->maxsgentries -= 1; /* account for chain pointer */
3835 } else {
3836 c->maxsgentries = 31; /* Default to traditional value */
3837 c->chainsize = 0; /* traditional */
3838 }
3839
3840 c->product_name = products[prod_index].product_name;
3841 c->access = *(products[prod_index].access);
3842 c->nr_cmds = c->max_commands - 4;
3843 if ((readb(&c->cfgtable->Signature[0]) != 'C') ||
3844 (readb(&c->cfgtable->Signature[1]) != 'I') ||
3845 (readb(&c->cfgtable->Signature[2]) != 'S') ||
3846 (readb(&c->cfgtable->Signature[3]) != 'S')) {
3847 printk("Does not appear to be a valid CISS config table\n");
3848 err = -ENODEV;
3849 goto err_out_free_res;
3850 }
3851 #ifdef CONFIG_X86
3852 {
3853 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
3854 __u32 prefetch;
3855 prefetch = readl(&(c->cfgtable->SCSI_Prefetch));
3856 prefetch |= 0x100;
3857 writel(prefetch, &(c->cfgtable->SCSI_Prefetch));
3858 }
3859 #endif
3860
3861 /* Disabling DMA prefetch and refetch for the P600.
3862 * An ASIC bug may result in accesses to invalid memory addresses.
3863 * We've disabled prefetch for some time now. Testing with XEN
3864 * kernels revealed a bug in the refetch if dom0 resides on a P600.
3865 */
3866 if(board_id == 0x3225103C) {
3867 __u32 dma_prefetch;
3868 __u32 dma_refetch;
3869 dma_prefetch = readl(c->vaddr + I2O_DMA1_CFG);
3870 dma_prefetch |= 0x8000;
3871 writel(dma_prefetch, c->vaddr + I2O_DMA1_CFG);
3872 pci_read_config_dword(pdev, PCI_COMMAND_PARITY, &dma_refetch);
3873 dma_refetch |= 0x1;
3874 pci_write_config_dword(pdev, PCI_COMMAND_PARITY, dma_refetch);
3875 }
3876
3877 #ifdef CCISS_DEBUG
3878 printk("Trying to put board into Simple mode\n");
3879 #endif /* CCISS_DEBUG */
3880 c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
3881 /* Update the field, and then ring the doorbell */
3882 writel(CFGTBL_Trans_Simple, &(c->cfgtable->HostWrite.TransportRequest));
3883 writel(CFGTBL_ChangeReq, c->vaddr + SA5_DOORBELL);
3884
3885 /* under certain very rare conditions, this can take awhile.
3886 * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3887 * as we enter this code.) */
3888 for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3889 if (!(readl(c->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3890 break;
3891 /* delay and try again */
3892 set_current_state(TASK_INTERRUPTIBLE);
3893 schedule_timeout(msecs_to_jiffies(1));
3894 }
3895
3896 #ifdef CCISS_DEBUG
3897 printk(KERN_DEBUG "I counter got to %d %x\n", i,
3898 readl(c->vaddr + SA5_DOORBELL));
3899 #endif /* CCISS_DEBUG */
3900 #ifdef CCISS_DEBUG
3901 print_cfg_table(c->cfgtable);
3902 #endif /* CCISS_DEBUG */
3903
3904 if (!(readl(&(c->cfgtable->TransportActive)) & CFGTBL_Trans_Simple)) {
3905 printk(KERN_WARNING "cciss: unable to get board into"
3906 " simple mode\n");
3907 err = -ENODEV;
3908 goto err_out_free_res;
3909 }
3910 return 0;
3911
3912 err_out_free_res:
3913 /*
3914 * Deliberately omit pci_disable_device(): it does something nasty to
3915 * Smart Array controllers that pci_enable_device does not undo
3916 */
3917 pci_release_regions(pdev);
3918 return err;
3919 }
3920
3921 /* Function to find the first free pointer into our hba[] array
3922 * Returns -1 if no free entries are left.
3923 */
3924 static int alloc_cciss_hba(void)
3925 {
3926 int i;
3927
3928 for (i = 0; i < MAX_CTLR; i++) {
3929 if (!hba[i]) {
3930 ctlr_info_t *p;
3931
3932 p = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
3933 if (!p)
3934 goto Enomem;
3935 hba[i] = p;
3936 return i;
3937 }
3938 }
3939 printk(KERN_WARNING "cciss: This driver supports a maximum"
3940 " of %d controllers.\n", MAX_CTLR);
3941 return -1;
3942 Enomem:
3943 printk(KERN_ERR "cciss: out of memory.\n");
3944 return -1;
3945 }
3946
3947 static void free_hba(int n)
3948 {
3949 ctlr_info_t *h = hba[n];
3950 int i;
3951
3952 hba[n] = NULL;
3953 for (i = 0; i < h->highest_lun + 1; i++)
3954 if (h->gendisk[i] != NULL)
3955 put_disk(h->gendisk[i]);
3956 kfree(h);
3957 }
3958
3959 /* Send a message CDB to the firmware. */
3960 static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
3961 {
3962 typedef struct {
3963 CommandListHeader_struct CommandHeader;
3964 RequestBlock_struct Request;
3965 ErrDescriptor_struct ErrorDescriptor;
3966 } Command;
3967 static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
3968 Command *cmd;
3969 dma_addr_t paddr64;
3970 uint32_t paddr32, tag;
3971 void __iomem *vaddr;
3972 int i, err;
3973
3974 vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
3975 if (vaddr == NULL)
3976 return -ENOMEM;
3977
3978 /* The Inbound Post Queue only accepts 32-bit physical addresses for the
3979 CCISS commands, so they must be allocated from the lower 4GiB of
3980 memory. */
3981 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
3982 if (err) {
3983 iounmap(vaddr);
3984 return -ENOMEM;
3985 }
3986
3987 cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
3988 if (cmd == NULL) {
3989 iounmap(vaddr);
3990 return -ENOMEM;
3991 }
3992
3993 /* This must fit, because of the 32-bit consistent DMA mask. Also,
3994 although there's no guarantee, we assume that the address is at
3995 least 4-byte aligned (most likely, it's page-aligned). */
3996 paddr32 = paddr64;
3997
3998 cmd->CommandHeader.ReplyQueue = 0;
3999 cmd->CommandHeader.SGList = 0;
4000 cmd->CommandHeader.SGTotal = 0;
4001 cmd->CommandHeader.Tag.lower = paddr32;
4002 cmd->CommandHeader.Tag.upper = 0;
4003 memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
4004
4005 cmd->Request.CDBLen = 16;
4006 cmd->Request.Type.Type = TYPE_MSG;
4007 cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
4008 cmd->Request.Type.Direction = XFER_NONE;
4009 cmd->Request.Timeout = 0; /* Don't time out */
4010 cmd->Request.CDB[0] = opcode;
4011 cmd->Request.CDB[1] = type;
4012 memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
4013
4014 cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
4015 cmd->ErrorDescriptor.Addr.upper = 0;
4016 cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
4017
4018 writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
4019
4020 for (i = 0; i < 10; i++) {
4021 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
4022 if ((tag & ~3) == paddr32)
4023 break;
4024 schedule_timeout_uninterruptible(HZ);
4025 }
4026
4027 iounmap(vaddr);
4028
4029 /* we leak the DMA buffer here ... no choice since the controller could
4030 still complete the command. */
4031 if (i == 10) {
4032 printk(KERN_ERR "cciss: controller message %02x:%02x timed out\n",
4033 opcode, type);
4034 return -ETIMEDOUT;
4035 }
4036
4037 pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
4038
4039 if (tag & 2) {
4040 printk(KERN_ERR "cciss: controller message %02x:%02x failed\n",
4041 opcode, type);
4042 return -EIO;
4043 }
4044
4045 printk(KERN_INFO "cciss: controller message %02x:%02x succeeded\n",
4046 opcode, type);
4047 return 0;
4048 }
4049
4050 #define cciss_soft_reset_controller(p) cciss_message(p, 1, 0)
4051 #define cciss_noop(p) cciss_message(p, 3, 0)
4052
4053 static __devinit int cciss_reset_msi(struct pci_dev *pdev)
4054 {
4055 /* the #defines are stolen from drivers/pci/msi.h. */
4056 #define msi_control_reg(base) (base + PCI_MSI_FLAGS)
4057 #define PCI_MSIX_FLAGS_ENABLE (1 << 15)
4058
4059 int pos;
4060 u16 control = 0;
4061
4062 pos = pci_find_capability(pdev, PCI_CAP_ID_MSI);
4063 if (pos) {
4064 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4065 if (control & PCI_MSI_FLAGS_ENABLE) {
4066 printk(KERN_INFO "cciss: resetting MSI\n");
4067 pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSI_FLAGS_ENABLE);
4068 }
4069 }
4070
4071 pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
4072 if (pos) {
4073 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4074 if (control & PCI_MSIX_FLAGS_ENABLE) {
4075 printk(KERN_INFO "cciss: resetting MSI-X\n");
4076 pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSIX_FLAGS_ENABLE);
4077 }
4078 }
4079
4080 return 0;
4081 }
4082
4083 /* This does a hard reset of the controller using PCI power management
4084 * states. */
4085 static __devinit int cciss_hard_reset_controller(struct pci_dev *pdev)
4086 {
4087 u16 pmcsr, saved_config_space[32];
4088 int i, pos;
4089
4090 printk(KERN_INFO "cciss: using PCI PM to reset controller\n");
4091
4092 /* This is very nearly the same thing as
4093
4094 pci_save_state(pci_dev);
4095 pci_set_power_state(pci_dev, PCI_D3hot);
4096 pci_set_power_state(pci_dev, PCI_D0);
4097 pci_restore_state(pci_dev);
4098
4099 but we can't use these nice canned kernel routines on
4100 kexec, because they also check the MSI/MSI-X state in PCI
4101 configuration space and do the wrong thing when it is
4102 set/cleared. Also, the pci_save/restore_state functions
4103 violate the ordering requirements for restoring the
4104 configuration space from the CCISS document (see the
4105 comment below). So we roll our own .... */
4106
4107 for (i = 0; i < 32; i++)
4108 pci_read_config_word(pdev, 2*i, &saved_config_space[i]);
4109
4110 pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4111 if (pos == 0) {
4112 printk(KERN_ERR "cciss_reset_controller: PCI PM not supported\n");
4113 return -ENODEV;
4114 }
4115
4116 /* Quoting from the Open CISS Specification: "The Power
4117 * Management Control/Status Register (CSR) controls the power
4118 * state of the device. The normal operating state is D0,
4119 * CSR=00h. The software off state is D3, CSR=03h. To reset
4120 * the controller, place the interface device in D3 then to
4121 * D0, this causes a secondary PCI reset which will reset the
4122 * controller." */
4123
4124 /* enter the D3hot power management state */
4125 pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4126 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4127 pmcsr |= PCI_D3hot;
4128 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4129
4130 schedule_timeout_uninterruptible(HZ >> 1);
4131
4132 /* enter the D0 power management state */
4133 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4134 pmcsr |= PCI_D0;
4135 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4136
4137 schedule_timeout_uninterruptible(HZ >> 1);
4138
4139 /* Restore the PCI configuration space. The Open CISS
4140 * Specification says, "Restore the PCI Configuration
4141 * Registers, offsets 00h through 60h. It is important to
4142 * restore the command register, 16-bits at offset 04h,
4143 * last. Do not restore the configuration status register,
4144 * 16-bits at offset 06h." Note that the offset is 2*i. */
4145 for (i = 0; i < 32; i++) {
4146 if (i == 2 || i == 3)
4147 continue;
4148 pci_write_config_word(pdev, 2*i, saved_config_space[i]);
4149 }
4150 wmb();
4151 pci_write_config_word(pdev, 4, saved_config_space[2]);
4152
4153 return 0;
4154 }
4155
4156 /*
4157 * This is it. Find all the controllers and register them. I really hate
4158 * stealing all these major device numbers.
4159 * returns the number of block devices registered.
4160 */
4161 static int __devinit cciss_init_one(struct pci_dev *pdev,
4162 const struct pci_device_id *ent)
4163 {
4164 int i;
4165 int j = 0;
4166 int k = 0;
4167 int rc;
4168 int dac, return_code;
4169 InquiryData_struct *inq_buff;
4170
4171 if (reset_devices) {
4172 /* Reset the controller with a PCI power-cycle */
4173 if (cciss_hard_reset_controller(pdev) || cciss_reset_msi(pdev))
4174 return -ENODEV;
4175
4176 /* Now try to get the controller to respond to a no-op. Some
4177 devices (notably the HP Smart Array 5i Controller) need
4178 up to 30 seconds to respond. */
4179 for (i=0; i<30; i++) {
4180 if (cciss_noop(pdev) == 0)
4181 break;
4182
4183 schedule_timeout_uninterruptible(HZ);
4184 }
4185 if (i == 30) {
4186 printk(KERN_ERR "cciss: controller seems dead\n");
4187 return -EBUSY;
4188 }
4189 }
4190
4191 i = alloc_cciss_hba();
4192 if (i < 0)
4193 return -1;
4194
4195 hba[i]->busy_initializing = 1;
4196 INIT_HLIST_HEAD(&hba[i]->cmpQ);
4197 INIT_HLIST_HEAD(&hba[i]->reqQ);
4198 mutex_init(&hba[i]->busy_shutting_down);
4199
4200 if (cciss_pci_init(hba[i], pdev) != 0)
4201 goto clean_no_release_regions;
4202
4203 sprintf(hba[i]->devname, "cciss%d", i);
4204 hba[i]->ctlr = i;
4205 hba[i]->pdev = pdev;
4206
4207 init_completion(&hba[i]->scan_wait);
4208
4209 if (cciss_create_hba_sysfs_entry(hba[i]))
4210 goto clean0;
4211
4212 /* configure PCI DMA stuff */
4213 if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
4214 dac = 1;
4215 else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
4216 dac = 0;
4217 else {
4218 printk(KERN_ERR "cciss: no suitable DMA available\n");
4219 goto clean1;
4220 }
4221
4222 /*
4223 * register with the major number, or get a dynamic major number
4224 * by passing 0 as argument. This is done for greater than
4225 * 8 controller support.
4226 */
4227 if (i < MAX_CTLR_ORIG)
4228 hba[i]->major = COMPAQ_CISS_MAJOR + i;
4229 rc = register_blkdev(hba[i]->major, hba[i]->devname);
4230 if (rc == -EBUSY || rc == -EINVAL) {
4231 printk(KERN_ERR
4232 "cciss: Unable to get major number %d for %s "
4233 "on hba %d\n", hba[i]->major, hba[i]->devname, i);
4234 goto clean1;
4235 } else {
4236 if (i >= MAX_CTLR_ORIG)
4237 hba[i]->major = rc;
4238 }
4239
4240 /* make sure the board interrupts are off */
4241 hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_OFF);
4242 if (request_irq(hba[i]->intr[SIMPLE_MODE_INT], do_cciss_intr,
4243 IRQF_DISABLED | IRQF_SHARED, hba[i]->devname, hba[i])) {
4244 printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
4245 hba[i]->intr[SIMPLE_MODE_INT], hba[i]->devname);
4246 goto clean2;
4247 }
4248
4249 printk(KERN_INFO "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
4250 hba[i]->devname, pdev->device, pci_name(pdev),
4251 hba[i]->intr[SIMPLE_MODE_INT], dac ? "" : " not");
4252
4253 hba[i]->cmd_pool_bits =
4254 kmalloc(DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4255 * sizeof(unsigned long), GFP_KERNEL);
4256 hba[i]->cmd_pool = (CommandList_struct *)
4257 pci_alloc_consistent(hba[i]->pdev,
4258 hba[i]->nr_cmds * sizeof(CommandList_struct),
4259 &(hba[i]->cmd_pool_dhandle));
4260 hba[i]->errinfo_pool = (ErrorInfo_struct *)
4261 pci_alloc_consistent(hba[i]->pdev,
4262 hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4263 &(hba[i]->errinfo_pool_dhandle));
4264 if ((hba[i]->cmd_pool_bits == NULL)
4265 || (hba[i]->cmd_pool == NULL)
4266 || (hba[i]->errinfo_pool == NULL)) {
4267 printk(KERN_ERR "cciss: out of memory");
4268 goto clean4;
4269 }
4270
4271 /* Need space for temp scatter list */
4272 hba[i]->scatter_list = kmalloc(hba[i]->max_commands *
4273 sizeof(struct scatterlist *),
4274 GFP_KERNEL);
4275 for (k = 0; k < hba[i]->nr_cmds; k++) {
4276 hba[i]->scatter_list[k] = kmalloc(sizeof(struct scatterlist) *
4277 hba[i]->maxsgentries,
4278 GFP_KERNEL);
4279 if (hba[i]->scatter_list[k] == NULL) {
4280 printk(KERN_ERR "cciss%d: could not allocate "
4281 "s/g lists\n", i);
4282 goto clean4;
4283 }
4284 }
4285 hba[i]->cmd_sg_list = cciss_allocate_sg_chain_blocks(hba[i],
4286 hba[i]->chainsize, hba[i]->nr_cmds);
4287 if (!hba[i]->cmd_sg_list && hba[i]->chainsize > 0)
4288 goto clean4;
4289
4290 spin_lock_init(&hba[i]->lock);
4291
4292 /* Initialize the pdev driver private data.
4293 have it point to hba[i]. */
4294 pci_set_drvdata(pdev, hba[i]);
4295 /* command and error info recs zeroed out before
4296 they are used */
4297 memset(hba[i]->cmd_pool_bits, 0,
4298 DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4299 * sizeof(unsigned long));
4300
4301 hba[i]->num_luns = 0;
4302 hba[i]->highest_lun = -1;
4303 for (j = 0; j < CISS_MAX_LUN; j++) {
4304 hba[i]->drv[j] = NULL;
4305 hba[i]->gendisk[j] = NULL;
4306 }
4307
4308 cciss_scsi_setup(i);
4309
4310 /* Turn the interrupts on so we can service requests */
4311 hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_ON);
4312
4313 /* Get the firmware version */
4314 inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
4315 if (inq_buff == NULL) {
4316 printk(KERN_ERR "cciss: out of memory\n");
4317 goto clean4;
4318 }
4319
4320 return_code = sendcmd_withirq(CISS_INQUIRY, i, inq_buff,
4321 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
4322 if (return_code == IO_OK) {
4323 hba[i]->firm_ver[0] = inq_buff->data_byte[32];
4324 hba[i]->firm_ver[1] = inq_buff->data_byte[33];
4325 hba[i]->firm_ver[2] = inq_buff->data_byte[34];
4326 hba[i]->firm_ver[3] = inq_buff->data_byte[35];
4327 } else { /* send command failed */
4328 printk(KERN_WARNING "cciss: unable to determine firmware"
4329 " version of controller\n");
4330 }
4331 kfree(inq_buff);
4332
4333 cciss_procinit(i);
4334
4335 hba[i]->cciss_max_sectors = 8192;
4336
4337 rebuild_lun_table(hba[i], 1, 0);
4338 hba[i]->busy_initializing = 0;
4339 return 1;
4340
4341 clean4:
4342 kfree(hba[i]->cmd_pool_bits);
4343 /* Free up sg elements */
4344 for (k = 0; k < hba[i]->nr_cmds; k++)
4345 kfree(hba[i]->scatter_list[k]);
4346 kfree(hba[i]->scatter_list);
4347 cciss_free_sg_chain_blocks(hba[i]->cmd_sg_list, hba[i]->nr_cmds);
4348 if (hba[i]->cmd_pool)
4349 pci_free_consistent(hba[i]->pdev,
4350 hba[i]->nr_cmds * sizeof(CommandList_struct),
4351 hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4352 if (hba[i]->errinfo_pool)
4353 pci_free_consistent(hba[i]->pdev,
4354 hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4355 hba[i]->errinfo_pool,
4356 hba[i]->errinfo_pool_dhandle);
4357 free_irq(hba[i]->intr[SIMPLE_MODE_INT], hba[i]);
4358 clean2:
4359 unregister_blkdev(hba[i]->major, hba[i]->devname);
4360 clean1:
4361 cciss_destroy_hba_sysfs_entry(hba[i]);
4362 clean0:
4363 pci_release_regions(pdev);
4364 clean_no_release_regions:
4365 hba[i]->busy_initializing = 0;
4366
4367 /*
4368 * Deliberately omit pci_disable_device(): it does something nasty to
4369 * Smart Array controllers that pci_enable_device does not undo
4370 */
4371 pci_set_drvdata(pdev, NULL);
4372 free_hba(i);
4373 return -1;
4374 }
4375
4376 static void cciss_shutdown(struct pci_dev *pdev)
4377 {
4378 ctlr_info_t *h;
4379 char *flush_buf;
4380 int return_code;
4381
4382 h = pci_get_drvdata(pdev);
4383 flush_buf = kzalloc(4, GFP_KERNEL);
4384 if (!flush_buf) {
4385 printk(KERN_WARNING
4386 "cciss:%d cache not flushed, out of memory.\n",
4387 h->ctlr);
4388 return;
4389 }
4390 /* write all data in the battery backed cache to disk */
4391 memset(flush_buf, 0, 4);
4392 return_code = sendcmd_withirq(CCISS_CACHE_FLUSH, h->ctlr, flush_buf,
4393 4, 0, CTLR_LUNID, TYPE_CMD);
4394 kfree(flush_buf);
4395 if (return_code != IO_OK)
4396 printk(KERN_WARNING "cciss%d: Error flushing cache\n",
4397 h->ctlr);
4398 h->access.set_intr_mask(h, CCISS_INTR_OFF);
4399 free_irq(h->intr[2], h);
4400 }
4401
4402 static void __devexit cciss_remove_one(struct pci_dev *pdev)
4403 {
4404 ctlr_info_t *tmp_ptr;
4405 int i, j;
4406
4407 if (pci_get_drvdata(pdev) == NULL) {
4408 printk(KERN_ERR "cciss: Unable to remove device \n");
4409 return;
4410 }
4411
4412 tmp_ptr = pci_get_drvdata(pdev);
4413 i = tmp_ptr->ctlr;
4414 if (hba[i] == NULL) {
4415 printk(KERN_ERR "cciss: device appears to "
4416 "already be removed \n");
4417 return;
4418 }
4419
4420 mutex_lock(&hba[i]->busy_shutting_down);
4421
4422 remove_from_scan_list(hba[i]);
4423 remove_proc_entry(hba[i]->devname, proc_cciss);
4424 unregister_blkdev(hba[i]->major, hba[i]->devname);
4425
4426 /* remove it from the disk list */
4427 for (j = 0; j < CISS_MAX_LUN; j++) {
4428 struct gendisk *disk = hba[i]->gendisk[j];
4429 if (disk) {
4430 struct request_queue *q = disk->queue;
4431
4432 if (disk->flags & GENHD_FL_UP) {
4433 cciss_destroy_ld_sysfs_entry(hba[i], j, 1);
4434 del_gendisk(disk);
4435 }
4436 if (q)
4437 blk_cleanup_queue(q);
4438 }
4439 }
4440
4441 #ifdef CONFIG_CISS_SCSI_TAPE
4442 cciss_unregister_scsi(i); /* unhook from SCSI subsystem */
4443 #endif
4444
4445 cciss_shutdown(pdev);
4446
4447 #ifdef CONFIG_PCI_MSI
4448 if (hba[i]->msix_vector)
4449 pci_disable_msix(hba[i]->pdev);
4450 else if (hba[i]->msi_vector)
4451 pci_disable_msi(hba[i]->pdev);
4452 #endif /* CONFIG_PCI_MSI */
4453
4454 iounmap(hba[i]->vaddr);
4455
4456 pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(CommandList_struct),
4457 hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4458 pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4459 hba[i]->errinfo_pool, hba[i]->errinfo_pool_dhandle);
4460 kfree(hba[i]->cmd_pool_bits);
4461 /* Free up sg elements */
4462 for (j = 0; j < hba[i]->nr_cmds; j++)
4463 kfree(hba[i]->scatter_list[j]);
4464 kfree(hba[i]->scatter_list);
4465 cciss_free_sg_chain_blocks(hba[i]->cmd_sg_list, hba[i]->nr_cmds);
4466 /*
4467 * Deliberately omit pci_disable_device(): it does something nasty to
4468 * Smart Array controllers that pci_enable_device does not undo
4469 */
4470 pci_release_regions(pdev);
4471 pci_set_drvdata(pdev, NULL);
4472 cciss_destroy_hba_sysfs_entry(hba[i]);
4473 mutex_unlock(&hba[i]->busy_shutting_down);
4474 free_hba(i);
4475 }
4476
4477 static struct pci_driver cciss_pci_driver = {
4478 .name = "cciss",
4479 .probe = cciss_init_one,
4480 .remove = __devexit_p(cciss_remove_one),
4481 .id_table = cciss_pci_device_id, /* id_table */
4482 .shutdown = cciss_shutdown,
4483 };
4484
4485 /*
4486 * This is it. Register the PCI driver information for the cards we control
4487 * the OS will call our registered routines when it finds one of our cards.
4488 */
4489 static int __init cciss_init(void)
4490 {
4491 int err;
4492
4493 /*
4494 * The hardware requires that commands are aligned on a 64-bit
4495 * boundary. Given that we use pci_alloc_consistent() to allocate an
4496 * array of them, the size must be a multiple of 8 bytes.
4497 */
4498 BUILD_BUG_ON(sizeof(CommandList_struct) % COMMANDLIST_ALIGNMENT);
4499
4500 printk(KERN_INFO DRIVER_NAME "\n");
4501
4502 err = bus_register(&cciss_bus_type);
4503 if (err)
4504 return err;
4505
4506 /* Start the scan thread */
4507 cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
4508 if (IS_ERR(cciss_scan_thread)) {
4509 err = PTR_ERR(cciss_scan_thread);
4510 goto err_bus_unregister;
4511 }
4512
4513 /* Register for our PCI devices */
4514 err = pci_register_driver(&cciss_pci_driver);
4515 if (err)
4516 goto err_thread_stop;
4517
4518 return err;
4519
4520 err_thread_stop:
4521 kthread_stop(cciss_scan_thread);
4522 err_bus_unregister:
4523 bus_unregister(&cciss_bus_type);
4524
4525 return err;
4526 }
4527
4528 static void __exit cciss_cleanup(void)
4529 {
4530 int i;
4531
4532 pci_unregister_driver(&cciss_pci_driver);
4533 /* double check that all controller entrys have been removed */
4534 for (i = 0; i < MAX_CTLR; i++) {
4535 if (hba[i] != NULL) {
4536 printk(KERN_WARNING "cciss: had to remove"
4537 " controller %d\n", i);
4538 cciss_remove_one(hba[i]->pdev);
4539 }
4540 }
4541 kthread_stop(cciss_scan_thread);
4542 remove_proc_entry("driver/cciss", NULL);
4543 bus_unregister(&cciss_bus_type);
4544 }
4545
4546 static void fail_all_cmds(unsigned long ctlr)
4547 {
4548 /* If we get here, the board is apparently dead. */
4549 ctlr_info_t *h = hba[ctlr];
4550 CommandList_struct *c;
4551 unsigned long flags;
4552
4553 printk(KERN_WARNING "cciss%d: controller not responding.\n", h->ctlr);
4554 h->alive = 0; /* the controller apparently died... */
4555
4556 spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
4557
4558 pci_disable_device(h->pdev); /* Make sure it is really dead. */
4559
4560 /* move everything off the request queue onto the completed queue */
4561 while (!hlist_empty(&h->reqQ)) {
4562 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
4563 removeQ(c);
4564 h->Qdepth--;
4565 addQ(&h->cmpQ, c);
4566 }
4567
4568 /* Now, fail everything on the completed queue with a HW error */
4569 while (!hlist_empty(&h->cmpQ)) {
4570 c = hlist_entry(h->cmpQ.first, CommandList_struct, list);
4571 removeQ(c);
4572 if (c->cmd_type != CMD_MSG_STALE)
4573 c->err_info->CommandStatus = CMD_HARDWARE_ERR;
4574 if (c->cmd_type == CMD_RWREQ) {
4575 complete_command(h, c, 0);
4576 } else if (c->cmd_type == CMD_IOCTL_PEND)
4577 complete(c->waiting);
4578 #ifdef CONFIG_CISS_SCSI_TAPE
4579 else if (c->cmd_type == CMD_SCSI)
4580 complete_scsi_command(c, 0, 0);
4581 #endif
4582 }
4583 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
4584 return;
4585 }
4586
4587 module_init(cciss_init);
4588 module_exit(cciss_cleanup);
CJK support for tty framebuffer vt