cciss: simplify scatter gather code
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / block / cciss.c
blobadc517c1381cd4be98c82b6aac55fb687e371bcb
1 /*
2 * Disk Array driver for HP Smart Array controllers.
3 * (C) Copyright 2000, 2007 Hewlett-Packard Development Company, L.P.
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.
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.
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.
19 * Questions/Comments/Bugfixes to iss_storagedev@hp.com
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>
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>
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)
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");
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");
77 #include "cciss_cmd.h"
78 #include "cciss.h"
79 #include <linux/cciss_ioctl.h>
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,}
113 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
115 /* board_id = Subsystem Device ID & Vendor ID
116 * product = Marketing Name for the board
117 * access = Address of the struct of function pointers
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},
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
155 /*define how many times we will try a command because of bus resets */
156 #define MAX_CMD_RETRIES 3
158 #define MAX_CTLR 32
160 /* Originally cciss driver only supports 8 major numbers */
161 #define MAX_CTLR_ORIG 8
163 static ctlr_info_t *hba[MAX_CTLR];
165 static struct task_struct *cciss_scan_thread;
166 static DEFINE_MUTEX(scan_mutex);
167 static LIST_HEAD(scan_q);
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);
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);
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);
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);
209 #ifdef CONFIG_PROC_FS
210 static void cciss_procinit(int i);
211 #else
212 static void cciss_procinit(int i)
215 #endif /* CONFIG_PROC_FS */
217 #ifdef CONFIG_COMPAT
218 static int cciss_compat_ioctl(struct block_device *, fmode_t,
219 unsigned, unsigned long);
220 #endif
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,
235 * Enqueuing and dequeuing functions for cmdlists.
237 static inline void addQ(struct hlist_head *list, CommandList_struct *c)
239 hlist_add_head(&c->list, list);
242 static inline void removeQ(CommandList_struct *c)
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.
252 if (WARN_ON(hlist_unhashed(&c->list))) {
253 c->cmd_type = CMD_MSG_STALE;
254 return;
257 hlist_del_init(&c->list);
260 static void cciss_free_sg_chain_blocks(SGDescriptor_struct **cmd_sg_list,
261 int nr_cmds)
263 int i;
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;
271 kfree(cmd_sg_list);
274 static SGDescriptor_struct **cciss_allocate_sg_chain_blocks(
275 ctlr_info_t *h, int chainsize, int nr_cmds)
277 int j;
278 SGDescriptor_struct **cmd_sg_list;
280 if (chainsize <= 0)
281 return NULL;
283 cmd_sg_list = kmalloc(sizeof(*cmd_sg_list) * nr_cmds, GFP_KERNEL);
284 if (!cmd_sg_list)
285 return NULL;
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;
298 return cmd_sg_list;
299 clean:
300 cciss_free_sg_chain_blocks(cmd_sg_list, nr_cmds);
301 return NULL;
304 #include "cciss_scsi.c" /* For SCSI tape support */
306 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
307 "UNKNOWN"
309 #define RAID_UNKNOWN (sizeof(raid_label) / sizeof(raid_label[0])-1)
311 #ifdef CONFIG_PROC_FS
314 * Report information about this controller.
316 #define ENG_GIG 1000000000
317 #define ENG_GIG_FACTOR (ENG_GIG/512)
318 #define ENGAGE_SCSI "engage scsi"
320 static struct proc_dir_entry *proc_cciss;
322 static void cciss_seq_show_header(struct seq_file *seq)
324 ctlr_info_t *h = seq->private;
326 seq_printf(seq, "%s: HP %s Controller\n"
327 "Board ID: 0x%08lx\n"
328 "Firmware Version: %c%c%c%c\n"
329 "IRQ: %d\n"
330 "Logical drives: %d\n"
331 "Current Q depth: %d\n"
332 "Current # commands on controller: %d\n"
333 "Max Q depth since init: %d\n"
334 "Max # commands on controller since init: %d\n"
335 "Max SG entries since init: %d\n",
336 h->devname,
337 h->product_name,
338 (unsigned long)h->board_id,
339 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
340 h->firm_ver[3], (unsigned int)h->intr[SIMPLE_MODE_INT],
341 h->num_luns,
342 h->Qdepth, h->commands_outstanding,
343 h->maxQsinceinit, h->max_outstanding, h->maxSG);
345 #ifdef CONFIG_CISS_SCSI_TAPE
346 cciss_seq_tape_report(seq, h->ctlr);
347 #endif /* CONFIG_CISS_SCSI_TAPE */
350 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
352 ctlr_info_t *h = seq->private;
353 unsigned ctlr = h->ctlr;
354 unsigned long flags;
356 /* prevent displaying bogus info during configuration
357 * or deconfiguration of a logical volume
359 spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
360 if (h->busy_configuring) {
361 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
362 return ERR_PTR(-EBUSY);
364 h->busy_configuring = 1;
365 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
367 if (*pos == 0)
368 cciss_seq_show_header(seq);
370 return pos;
373 static int cciss_seq_show(struct seq_file *seq, void *v)
375 sector_t vol_sz, vol_sz_frac;
376 ctlr_info_t *h = seq->private;
377 unsigned ctlr = h->ctlr;
378 loff_t *pos = v;
379 drive_info_struct *drv = h->drv[*pos];
381 if (*pos > h->highest_lun)
382 return 0;
384 if (drv == NULL) /* it's possible for h->drv[] to have holes. */
385 return 0;
387 if (drv->heads == 0)
388 return 0;
390 vol_sz = drv->nr_blocks;
391 vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
392 vol_sz_frac *= 100;
393 sector_div(vol_sz_frac, ENG_GIG_FACTOR);
395 if (drv->raid_level < 0 || drv->raid_level > RAID_UNKNOWN)
396 drv->raid_level = RAID_UNKNOWN;
397 seq_printf(seq, "cciss/c%dd%d:"
398 "\t%4u.%02uGB\tRAID %s\n",
399 ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
400 raid_label[drv->raid_level]);
401 return 0;
404 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
406 ctlr_info_t *h = seq->private;
408 if (*pos > h->highest_lun)
409 return NULL;
410 *pos += 1;
412 return pos;
415 static void cciss_seq_stop(struct seq_file *seq, void *v)
417 ctlr_info_t *h = seq->private;
419 /* Only reset h->busy_configuring if we succeeded in setting
420 * it during cciss_seq_start. */
421 if (v == ERR_PTR(-EBUSY))
422 return;
424 h->busy_configuring = 0;
427 static const struct seq_operations cciss_seq_ops = {
428 .start = cciss_seq_start,
429 .show = cciss_seq_show,
430 .next = cciss_seq_next,
431 .stop = cciss_seq_stop,
434 static int cciss_seq_open(struct inode *inode, struct file *file)
436 int ret = seq_open(file, &cciss_seq_ops);
437 struct seq_file *seq = file->private_data;
439 if (!ret)
440 seq->private = PDE(inode)->data;
442 return ret;
445 static ssize_t
446 cciss_proc_write(struct file *file, const char __user *buf,
447 size_t length, loff_t *ppos)
449 int err;
450 char *buffer;
452 #ifndef CONFIG_CISS_SCSI_TAPE
453 return -EINVAL;
454 #endif
456 if (!buf || length > PAGE_SIZE - 1)
457 return -EINVAL;
459 buffer = (char *)__get_free_page(GFP_KERNEL);
460 if (!buffer)
461 return -ENOMEM;
463 err = -EFAULT;
464 if (copy_from_user(buffer, buf, length))
465 goto out;
466 buffer[length] = '\0';
468 #ifdef CONFIG_CISS_SCSI_TAPE
469 if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
470 struct seq_file *seq = file->private_data;
471 ctlr_info_t *h = seq->private;
473 err = cciss_engage_scsi(h->ctlr);
474 if (err == 0)
475 err = length;
476 } else
477 #endif /* CONFIG_CISS_SCSI_TAPE */
478 err = -EINVAL;
479 /* might be nice to have "disengage" too, but it's not
480 safely possible. (only 1 module use count, lock issues.) */
482 out:
483 free_page((unsigned long)buffer);
484 return err;
487 static const struct file_operations cciss_proc_fops = {
488 .owner = THIS_MODULE,
489 .open = cciss_seq_open,
490 .read = seq_read,
491 .llseek = seq_lseek,
492 .release = seq_release,
493 .write = cciss_proc_write,
496 static void __devinit cciss_procinit(int i)
498 struct proc_dir_entry *pde;
500 if (proc_cciss == NULL)
501 proc_cciss = proc_mkdir("driver/cciss", NULL);
502 if (!proc_cciss)
503 return;
504 pde = proc_create_data(hba[i]->devname, S_IWUSR | S_IRUSR | S_IRGRP |
505 S_IROTH, proc_cciss,
506 &cciss_proc_fops, hba[i]);
508 #endif /* CONFIG_PROC_FS */
510 #define MAX_PRODUCT_NAME_LEN 19
512 #define to_hba(n) container_of(n, struct ctlr_info, dev)
513 #define to_drv(n) container_of(n, drive_info_struct, dev)
515 static ssize_t host_store_rescan(struct device *dev,
516 struct device_attribute *attr,
517 const char *buf, size_t count)
519 struct ctlr_info *h = to_hba(dev);
521 add_to_scan_list(h);
522 wake_up_process(cciss_scan_thread);
523 wait_for_completion_interruptible(&h->scan_wait);
525 return count;
527 static DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
529 static ssize_t dev_show_unique_id(struct device *dev,
530 struct device_attribute *attr,
531 char *buf)
533 drive_info_struct *drv = to_drv(dev);
534 struct ctlr_info *h = to_hba(drv->dev.parent);
535 __u8 sn[16];
536 unsigned long flags;
537 int ret = 0;
539 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
540 if (h->busy_configuring)
541 ret = -EBUSY;
542 else
543 memcpy(sn, drv->serial_no, sizeof(sn));
544 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
546 if (ret)
547 return ret;
548 else
549 return snprintf(buf, 16 * 2 + 2,
550 "%02X%02X%02X%02X%02X%02X%02X%02X"
551 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
552 sn[0], sn[1], sn[2], sn[3],
553 sn[4], sn[5], sn[6], sn[7],
554 sn[8], sn[9], sn[10], sn[11],
555 sn[12], sn[13], sn[14], sn[15]);
557 static DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
559 static ssize_t dev_show_vendor(struct device *dev,
560 struct device_attribute *attr,
561 char *buf)
563 drive_info_struct *drv = to_drv(dev);
564 struct ctlr_info *h = to_hba(drv->dev.parent);
565 char vendor[VENDOR_LEN + 1];
566 unsigned long flags;
567 int ret = 0;
569 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
570 if (h->busy_configuring)
571 ret = -EBUSY;
572 else
573 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
574 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
576 if (ret)
577 return ret;
578 else
579 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
581 static DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
583 static ssize_t dev_show_model(struct device *dev,
584 struct device_attribute *attr,
585 char *buf)
587 drive_info_struct *drv = to_drv(dev);
588 struct ctlr_info *h = to_hba(drv->dev.parent);
589 char model[MODEL_LEN + 1];
590 unsigned long flags;
591 int ret = 0;
593 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
594 if (h->busy_configuring)
595 ret = -EBUSY;
596 else
597 memcpy(model, drv->model, MODEL_LEN + 1);
598 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
600 if (ret)
601 return ret;
602 else
603 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
605 static DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
607 static ssize_t dev_show_rev(struct device *dev,
608 struct device_attribute *attr,
609 char *buf)
611 drive_info_struct *drv = to_drv(dev);
612 struct ctlr_info *h = to_hba(drv->dev.parent);
613 char rev[REV_LEN + 1];
614 unsigned long flags;
615 int ret = 0;
617 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
618 if (h->busy_configuring)
619 ret = -EBUSY;
620 else
621 memcpy(rev, drv->rev, REV_LEN + 1);
622 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
624 if (ret)
625 return ret;
626 else
627 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
629 static DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
631 static ssize_t cciss_show_lunid(struct device *dev,
632 struct device_attribute *attr, char *buf)
634 drive_info_struct *drv = to_drv(dev);
635 struct ctlr_info *h = to_hba(drv->dev.parent);
636 unsigned long flags;
637 unsigned char lunid[8];
639 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
640 if (h->busy_configuring) {
641 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
642 return -EBUSY;
644 if (!drv->heads) {
645 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
646 return -ENOTTY;
648 memcpy(lunid, drv->LunID, sizeof(lunid));
649 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
650 return snprintf(buf, 20, "0x%02x%02x%02x%02x%02x%02x%02x%02x\n",
651 lunid[0], lunid[1], lunid[2], lunid[3],
652 lunid[4], lunid[5], lunid[6], lunid[7]);
654 static DEVICE_ATTR(lunid, S_IRUGO, cciss_show_lunid, NULL);
656 static ssize_t cciss_show_raid_level(struct device *dev,
657 struct device_attribute *attr, char *buf)
659 drive_info_struct *drv = to_drv(dev);
660 struct ctlr_info *h = to_hba(drv->dev.parent);
661 int raid;
662 unsigned long flags;
664 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
665 if (h->busy_configuring) {
666 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
667 return -EBUSY;
669 raid = drv->raid_level;
670 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
671 if (raid < 0 || raid > RAID_UNKNOWN)
672 raid = RAID_UNKNOWN;
674 return snprintf(buf, strlen(raid_label[raid]) + 7, "RAID %s\n",
675 raid_label[raid]);
677 static DEVICE_ATTR(raid_level, S_IRUGO, cciss_show_raid_level, NULL);
679 static ssize_t cciss_show_usage_count(struct device *dev,
680 struct device_attribute *attr, char *buf)
682 drive_info_struct *drv = to_drv(dev);
683 struct ctlr_info *h = to_hba(drv->dev.parent);
684 unsigned long flags;
685 int count;
687 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
688 if (h->busy_configuring) {
689 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
690 return -EBUSY;
692 count = drv->usage_count;
693 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
694 return snprintf(buf, 20, "%d\n", count);
696 static DEVICE_ATTR(usage_count, S_IRUGO, cciss_show_usage_count, NULL);
698 static struct attribute *cciss_host_attrs[] = {
699 &dev_attr_rescan.attr,
700 NULL
703 static struct attribute_group cciss_host_attr_group = {
704 .attrs = cciss_host_attrs,
707 static const struct attribute_group *cciss_host_attr_groups[] = {
708 &cciss_host_attr_group,
709 NULL
712 static struct device_type cciss_host_type = {
713 .name = "cciss_host",
714 .groups = cciss_host_attr_groups,
715 .release = cciss_hba_release,
718 static struct attribute *cciss_dev_attrs[] = {
719 &dev_attr_unique_id.attr,
720 &dev_attr_model.attr,
721 &dev_attr_vendor.attr,
722 &dev_attr_rev.attr,
723 &dev_attr_lunid.attr,
724 &dev_attr_raid_level.attr,
725 &dev_attr_usage_count.attr,
726 NULL
729 static struct attribute_group cciss_dev_attr_group = {
730 .attrs = cciss_dev_attrs,
733 static const struct attribute_group *cciss_dev_attr_groups[] = {
734 &cciss_dev_attr_group,
735 NULL
738 static struct device_type cciss_dev_type = {
739 .name = "cciss_device",
740 .groups = cciss_dev_attr_groups,
741 .release = cciss_device_release,
744 static struct bus_type cciss_bus_type = {
745 .name = "cciss",
749 * cciss_hba_release is called when the reference count
750 * of h->dev goes to zero.
752 static void cciss_hba_release(struct device *dev)
755 * nothing to do, but need this to avoid a warning
756 * about not having a release handler from lib/kref.c.
761 * Initialize sysfs entry for each controller. This sets up and registers
762 * the 'cciss#' directory for each individual controller under
763 * /sys/bus/pci/devices/<dev>/.
765 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
767 device_initialize(&h->dev);
768 h->dev.type = &cciss_host_type;
769 h->dev.bus = &cciss_bus_type;
770 dev_set_name(&h->dev, "%s", h->devname);
771 h->dev.parent = &h->pdev->dev;
773 return device_add(&h->dev);
777 * Remove sysfs entries for an hba.
779 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
781 device_del(&h->dev);
782 put_device(&h->dev); /* final put. */
785 /* cciss_device_release is called when the reference count
786 * of h->drv[x]dev goes to zero.
788 static void cciss_device_release(struct device *dev)
790 drive_info_struct *drv = to_drv(dev);
791 kfree(drv);
795 * Initialize sysfs for each logical drive. This sets up and registers
796 * the 'c#d#' directory for each individual logical drive under
797 * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
798 * /sys/block/cciss!c#d# to this entry.
800 static long cciss_create_ld_sysfs_entry(struct ctlr_info *h,
801 int drv_index)
803 struct device *dev;
805 if (h->drv[drv_index]->device_initialized)
806 return 0;
808 dev = &h->drv[drv_index]->dev;
809 device_initialize(dev);
810 dev->type = &cciss_dev_type;
811 dev->bus = &cciss_bus_type;
812 dev_set_name(dev, "c%dd%d", h->ctlr, drv_index);
813 dev->parent = &h->dev;
814 h->drv[drv_index]->device_initialized = 1;
815 return device_add(dev);
819 * Remove sysfs entries for a logical drive.
821 static void cciss_destroy_ld_sysfs_entry(struct ctlr_info *h, int drv_index,
822 int ctlr_exiting)
824 struct device *dev = &h->drv[drv_index]->dev;
826 /* special case for c*d0, we only destroy it on controller exit */
827 if (drv_index == 0 && !ctlr_exiting)
828 return;
830 device_del(dev);
831 put_device(dev); /* the "final" put. */
832 h->drv[drv_index] = NULL;
836 * For operations that cannot sleep, a command block is allocated at init,
837 * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
838 * which ones are free or in use. For operations that can wait for kmalloc
839 * to possible sleep, this routine can be called with get_from_pool set to 0.
840 * cmd_free() MUST be called with a got_from_pool set to 0 if cmd_alloc was.
842 static CommandList_struct *cmd_alloc(ctlr_info_t *h, int get_from_pool)
844 CommandList_struct *c;
845 int i;
846 u64bit temp64;
847 dma_addr_t cmd_dma_handle, err_dma_handle;
849 if (!get_from_pool) {
850 c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
851 sizeof(CommandList_struct), &cmd_dma_handle);
852 if (c == NULL)
853 return NULL;
854 memset(c, 0, sizeof(CommandList_struct));
856 c->cmdindex = -1;
858 c->err_info = (ErrorInfo_struct *)
859 pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
860 &err_dma_handle);
862 if (c->err_info == NULL) {
863 pci_free_consistent(h->pdev,
864 sizeof(CommandList_struct), c, cmd_dma_handle);
865 return NULL;
867 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
868 } else { /* get it out of the controllers pool */
870 do {
871 i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
872 if (i == h->nr_cmds)
873 return NULL;
874 } while (test_and_set_bit
875 (i & (BITS_PER_LONG - 1),
876 h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
877 #ifdef CCISS_DEBUG
878 printk(KERN_DEBUG "cciss: using command buffer %d\n", i);
879 #endif
880 c = h->cmd_pool + i;
881 memset(c, 0, sizeof(CommandList_struct));
882 cmd_dma_handle = h->cmd_pool_dhandle
883 + i * sizeof(CommandList_struct);
884 c->err_info = h->errinfo_pool + i;
885 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
886 err_dma_handle = h->errinfo_pool_dhandle
887 + i * sizeof(ErrorInfo_struct);
888 h->nr_allocs++;
890 c->cmdindex = i;
893 INIT_HLIST_NODE(&c->list);
894 c->busaddr = (__u32) cmd_dma_handle;
895 temp64.val = (__u64) err_dma_handle;
896 c->ErrDesc.Addr.lower = temp64.val32.lower;
897 c->ErrDesc.Addr.upper = temp64.val32.upper;
898 c->ErrDesc.Len = sizeof(ErrorInfo_struct);
900 c->ctlr = h->ctlr;
901 return c;
905 * Frees a command block that was previously allocated with cmd_alloc().
907 static void cmd_free(ctlr_info_t *h, CommandList_struct *c, int got_from_pool)
909 int i;
910 u64bit temp64;
912 if (!got_from_pool) {
913 temp64.val32.lower = c->ErrDesc.Addr.lower;
914 temp64.val32.upper = c->ErrDesc.Addr.upper;
915 pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
916 c->err_info, (dma_addr_t) temp64.val);
917 pci_free_consistent(h->pdev, sizeof(CommandList_struct),
918 c, (dma_addr_t) c->busaddr);
919 } else {
920 i = c - h->cmd_pool;
921 clear_bit(i & (BITS_PER_LONG - 1),
922 h->cmd_pool_bits + (i / BITS_PER_LONG));
923 h->nr_frees++;
927 static inline ctlr_info_t *get_host(struct gendisk *disk)
929 return disk->queue->queuedata;
932 static inline drive_info_struct *get_drv(struct gendisk *disk)
934 return disk->private_data;
938 * Open. Make sure the device is really there.
940 static int cciss_open(struct block_device *bdev, fmode_t mode)
942 ctlr_info_t *host = get_host(bdev->bd_disk);
943 drive_info_struct *drv = get_drv(bdev->bd_disk);
945 #ifdef CCISS_DEBUG
946 printk(KERN_DEBUG "cciss_open %s\n", bdev->bd_disk->disk_name);
947 #endif /* CCISS_DEBUG */
949 if (drv->busy_configuring)
950 return -EBUSY;
952 * Root is allowed to open raw volume zero even if it's not configured
953 * so array config can still work. Root is also allowed to open any
954 * volume that has a LUN ID, so it can issue IOCTL to reread the
955 * disk information. I don't think I really like this
956 * but I'm already using way to many device nodes to claim another one
957 * for "raw controller".
959 if (drv->heads == 0) {
960 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
961 /* if not node 0 make sure it is a partition = 0 */
962 if (MINOR(bdev->bd_dev) & 0x0f) {
963 return -ENXIO;
964 /* if it is, make sure we have a LUN ID */
965 } else if (memcmp(drv->LunID, CTLR_LUNID,
966 sizeof(drv->LunID))) {
967 return -ENXIO;
970 if (!capable(CAP_SYS_ADMIN))
971 return -EPERM;
973 drv->usage_count++;
974 host->usage_count++;
975 return 0;
979 * Close. Sync first.
981 static int cciss_release(struct gendisk *disk, fmode_t mode)
983 ctlr_info_t *host = get_host(disk);
984 drive_info_struct *drv = get_drv(disk);
986 #ifdef CCISS_DEBUG
987 printk(KERN_DEBUG "cciss_release %s\n", disk->disk_name);
988 #endif /* CCISS_DEBUG */
990 drv->usage_count--;
991 host->usage_count--;
992 return 0;
995 #ifdef CONFIG_COMPAT
997 static int do_ioctl(struct block_device *bdev, fmode_t mode,
998 unsigned cmd, unsigned long arg)
1000 int ret;
1001 lock_kernel();
1002 ret = cciss_ioctl(bdev, mode, cmd, arg);
1003 unlock_kernel();
1004 return ret;
1007 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1008 unsigned cmd, unsigned long arg);
1009 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1010 unsigned cmd, unsigned long arg);
1012 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
1013 unsigned cmd, unsigned long arg)
1015 switch (cmd) {
1016 case CCISS_GETPCIINFO:
1017 case CCISS_GETINTINFO:
1018 case CCISS_SETINTINFO:
1019 case CCISS_GETNODENAME:
1020 case CCISS_SETNODENAME:
1021 case CCISS_GETHEARTBEAT:
1022 case CCISS_GETBUSTYPES:
1023 case CCISS_GETFIRMVER:
1024 case CCISS_GETDRIVVER:
1025 case CCISS_REVALIDVOLS:
1026 case CCISS_DEREGDISK:
1027 case CCISS_REGNEWDISK:
1028 case CCISS_REGNEWD:
1029 case CCISS_RESCANDISK:
1030 case CCISS_GETLUNINFO:
1031 return do_ioctl(bdev, mode, cmd, arg);
1033 case CCISS_PASSTHRU32:
1034 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
1035 case CCISS_BIG_PASSTHRU32:
1036 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
1038 default:
1039 return -ENOIOCTLCMD;
1043 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1044 unsigned cmd, unsigned long arg)
1046 IOCTL32_Command_struct __user *arg32 =
1047 (IOCTL32_Command_struct __user *) arg;
1048 IOCTL_Command_struct arg64;
1049 IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
1050 int err;
1051 u32 cp;
1053 err = 0;
1054 err |=
1055 copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1056 sizeof(arg64.LUN_info));
1057 err |=
1058 copy_from_user(&arg64.Request, &arg32->Request,
1059 sizeof(arg64.Request));
1060 err |=
1061 copy_from_user(&arg64.error_info, &arg32->error_info,
1062 sizeof(arg64.error_info));
1063 err |= get_user(arg64.buf_size, &arg32->buf_size);
1064 err |= get_user(cp, &arg32->buf);
1065 arg64.buf = compat_ptr(cp);
1066 err |= copy_to_user(p, &arg64, sizeof(arg64));
1068 if (err)
1069 return -EFAULT;
1071 err = do_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
1072 if (err)
1073 return err;
1074 err |=
1075 copy_in_user(&arg32->error_info, &p->error_info,
1076 sizeof(arg32->error_info));
1077 if (err)
1078 return -EFAULT;
1079 return err;
1082 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1083 unsigned cmd, unsigned long arg)
1085 BIG_IOCTL32_Command_struct __user *arg32 =
1086 (BIG_IOCTL32_Command_struct __user *) arg;
1087 BIG_IOCTL_Command_struct arg64;
1088 BIG_IOCTL_Command_struct __user *p =
1089 compat_alloc_user_space(sizeof(arg64));
1090 int err;
1091 u32 cp;
1093 err = 0;
1094 err |=
1095 copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1096 sizeof(arg64.LUN_info));
1097 err |=
1098 copy_from_user(&arg64.Request, &arg32->Request,
1099 sizeof(arg64.Request));
1100 err |=
1101 copy_from_user(&arg64.error_info, &arg32->error_info,
1102 sizeof(arg64.error_info));
1103 err |= get_user(arg64.buf_size, &arg32->buf_size);
1104 err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1105 err |= get_user(cp, &arg32->buf);
1106 arg64.buf = compat_ptr(cp);
1107 err |= copy_to_user(p, &arg64, sizeof(arg64));
1109 if (err)
1110 return -EFAULT;
1112 err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1113 if (err)
1114 return err;
1115 err |=
1116 copy_in_user(&arg32->error_info, &p->error_info,
1117 sizeof(arg32->error_info));
1118 if (err)
1119 return -EFAULT;
1120 return err;
1122 #endif
1124 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1126 drive_info_struct *drv = get_drv(bdev->bd_disk);
1128 if (!drv->cylinders)
1129 return -ENXIO;
1131 geo->heads = drv->heads;
1132 geo->sectors = drv->sectors;
1133 geo->cylinders = drv->cylinders;
1134 return 0;
1137 static void check_ioctl_unit_attention(ctlr_info_t *host, CommandList_struct *c)
1139 if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1140 c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1141 (void)check_for_unit_attention(host, c);
1144 * ioctl
1146 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1147 unsigned int cmd, unsigned long arg)
1149 struct gendisk *disk = bdev->bd_disk;
1150 ctlr_info_t *host = get_host(disk);
1151 drive_info_struct *drv = get_drv(disk);
1152 int ctlr = host->ctlr;
1153 void __user *argp = (void __user *)arg;
1155 #ifdef CCISS_DEBUG
1156 printk(KERN_DEBUG "cciss_ioctl: Called with cmd=%x %lx\n", cmd, arg);
1157 #endif /* CCISS_DEBUG */
1159 switch (cmd) {
1160 case CCISS_GETPCIINFO:
1162 cciss_pci_info_struct pciinfo;
1164 if (!arg)
1165 return -EINVAL;
1166 pciinfo.domain = pci_domain_nr(host->pdev->bus);
1167 pciinfo.bus = host->pdev->bus->number;
1168 pciinfo.dev_fn = host->pdev->devfn;
1169 pciinfo.board_id = host->board_id;
1170 if (copy_to_user
1171 (argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1172 return -EFAULT;
1173 return 0;
1175 case CCISS_GETINTINFO:
1177 cciss_coalint_struct intinfo;
1178 if (!arg)
1179 return -EINVAL;
1180 intinfo.delay =
1181 readl(&host->cfgtable->HostWrite.CoalIntDelay);
1182 intinfo.count =
1183 readl(&host->cfgtable->HostWrite.CoalIntCount);
1184 if (copy_to_user
1185 (argp, &intinfo, sizeof(cciss_coalint_struct)))
1186 return -EFAULT;
1187 return 0;
1189 case CCISS_SETINTINFO:
1191 cciss_coalint_struct intinfo;
1192 unsigned long flags;
1193 int i;
1195 if (!arg)
1196 return -EINVAL;
1197 if (!capable(CAP_SYS_ADMIN))
1198 return -EPERM;
1199 if (copy_from_user
1200 (&intinfo, argp, sizeof(cciss_coalint_struct)))
1201 return -EFAULT;
1202 if ((intinfo.delay == 0) && (intinfo.count == 0))
1204 // printk("cciss_ioctl: delay and count cannot be 0\n");
1205 return -EINVAL;
1207 spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1208 /* Update the field, and then ring the doorbell */
1209 writel(intinfo.delay,
1210 &(host->cfgtable->HostWrite.CoalIntDelay));
1211 writel(intinfo.count,
1212 &(host->cfgtable->HostWrite.CoalIntCount));
1213 writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1215 for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1216 if (!(readl(host->vaddr + SA5_DOORBELL)
1217 & CFGTBL_ChangeReq))
1218 break;
1219 /* delay and try again */
1220 udelay(1000);
1222 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1223 if (i >= MAX_IOCTL_CONFIG_WAIT)
1224 return -EAGAIN;
1225 return 0;
1227 case CCISS_GETNODENAME:
1229 NodeName_type NodeName;
1230 int i;
1232 if (!arg)
1233 return -EINVAL;
1234 for (i = 0; i < 16; i++)
1235 NodeName[i] =
1236 readb(&host->cfgtable->ServerName[i]);
1237 if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1238 return -EFAULT;
1239 return 0;
1241 case CCISS_SETNODENAME:
1243 NodeName_type NodeName;
1244 unsigned long flags;
1245 int i;
1247 if (!arg)
1248 return -EINVAL;
1249 if (!capable(CAP_SYS_ADMIN))
1250 return -EPERM;
1252 if (copy_from_user
1253 (NodeName, argp, sizeof(NodeName_type)))
1254 return -EFAULT;
1256 spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1258 /* Update the field, and then ring the doorbell */
1259 for (i = 0; i < 16; i++)
1260 writeb(NodeName[i],
1261 &host->cfgtable->ServerName[i]);
1263 writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1265 for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1266 if (!(readl(host->vaddr + SA5_DOORBELL)
1267 & CFGTBL_ChangeReq))
1268 break;
1269 /* delay and try again */
1270 udelay(1000);
1272 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1273 if (i >= MAX_IOCTL_CONFIG_WAIT)
1274 return -EAGAIN;
1275 return 0;
1278 case CCISS_GETHEARTBEAT:
1280 Heartbeat_type heartbeat;
1282 if (!arg)
1283 return -EINVAL;
1284 heartbeat = readl(&host->cfgtable->HeartBeat);
1285 if (copy_to_user
1286 (argp, &heartbeat, sizeof(Heartbeat_type)))
1287 return -EFAULT;
1288 return 0;
1290 case CCISS_GETBUSTYPES:
1292 BusTypes_type BusTypes;
1294 if (!arg)
1295 return -EINVAL;
1296 BusTypes = readl(&host->cfgtable->BusTypes);
1297 if (copy_to_user
1298 (argp, &BusTypes, sizeof(BusTypes_type)))
1299 return -EFAULT;
1300 return 0;
1302 case CCISS_GETFIRMVER:
1304 FirmwareVer_type firmware;
1306 if (!arg)
1307 return -EINVAL;
1308 memcpy(firmware, host->firm_ver, 4);
1310 if (copy_to_user
1311 (argp, firmware, sizeof(FirmwareVer_type)))
1312 return -EFAULT;
1313 return 0;
1315 case CCISS_GETDRIVVER:
1317 DriverVer_type DriverVer = DRIVER_VERSION;
1319 if (!arg)
1320 return -EINVAL;
1322 if (copy_to_user
1323 (argp, &DriverVer, sizeof(DriverVer_type)))
1324 return -EFAULT;
1325 return 0;
1328 case CCISS_DEREGDISK:
1329 case CCISS_REGNEWD:
1330 case CCISS_REVALIDVOLS:
1331 return rebuild_lun_table(host, 0, 1);
1333 case CCISS_GETLUNINFO:{
1334 LogvolInfo_struct luninfo;
1336 memcpy(&luninfo.LunID, drv->LunID,
1337 sizeof(luninfo.LunID));
1338 luninfo.num_opens = drv->usage_count;
1339 luninfo.num_parts = 0;
1340 if (copy_to_user(argp, &luninfo,
1341 sizeof(LogvolInfo_struct)))
1342 return -EFAULT;
1343 return 0;
1345 case CCISS_PASSTHRU:
1347 IOCTL_Command_struct iocommand;
1348 CommandList_struct *c;
1349 char *buff = NULL;
1350 u64bit temp64;
1351 unsigned long flags;
1352 DECLARE_COMPLETION_ONSTACK(wait);
1354 if (!arg)
1355 return -EINVAL;
1357 if (!capable(CAP_SYS_RAWIO))
1358 return -EPERM;
1360 if (copy_from_user
1361 (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1362 return -EFAULT;
1363 if ((iocommand.buf_size < 1) &&
1364 (iocommand.Request.Type.Direction != XFER_NONE)) {
1365 return -EINVAL;
1367 #if 0 /* 'buf_size' member is 16-bits, and always smaller than kmalloc limit */
1368 /* Check kmalloc limits */
1369 if (iocommand.buf_size > 128000)
1370 return -EINVAL;
1371 #endif
1372 if (iocommand.buf_size > 0) {
1373 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1374 if (buff == NULL)
1375 return -EFAULT;
1377 if (iocommand.Request.Type.Direction == XFER_WRITE) {
1378 /* Copy the data into the buffer we created */
1379 if (copy_from_user
1380 (buff, iocommand.buf, iocommand.buf_size)) {
1381 kfree(buff);
1382 return -EFAULT;
1384 } else {
1385 memset(buff, 0, iocommand.buf_size);
1387 if ((c = cmd_alloc(host, 0)) == NULL) {
1388 kfree(buff);
1389 return -ENOMEM;
1391 /* Fill in the command type */
1392 c->cmd_type = CMD_IOCTL_PEND;
1393 /* Fill in Command Header */
1394 c->Header.ReplyQueue = 0; /* unused in simple mode */
1395 if (iocommand.buf_size > 0) /* buffer to fill */
1397 c->Header.SGList = 1;
1398 c->Header.SGTotal = 1;
1399 } else /* no buffers to fill */
1401 c->Header.SGList = 0;
1402 c->Header.SGTotal = 0;
1404 c->Header.LUN = iocommand.LUN_info;
1405 /* use the kernel address the cmd block for tag */
1406 c->Header.Tag.lower = c->busaddr;
1408 /* Fill in Request block */
1409 c->Request = iocommand.Request;
1411 /* Fill in the scatter gather information */
1412 if (iocommand.buf_size > 0) {
1413 temp64.val = pci_map_single(host->pdev, buff,
1414 iocommand.buf_size,
1415 PCI_DMA_BIDIRECTIONAL);
1416 c->SG[0].Addr.lower = temp64.val32.lower;
1417 c->SG[0].Addr.upper = temp64.val32.upper;
1418 c->SG[0].Len = iocommand.buf_size;
1419 c->SG[0].Ext = 0; /* we are not chaining */
1421 c->waiting = &wait;
1423 /* Put the request on the tail of the request queue */
1424 spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1425 addQ(&host->reqQ, c);
1426 host->Qdepth++;
1427 start_io(host);
1428 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1430 wait_for_completion(&wait);
1432 /* unlock the buffers from DMA */
1433 temp64.val32.lower = c->SG[0].Addr.lower;
1434 temp64.val32.upper = c->SG[0].Addr.upper;
1435 pci_unmap_single(host->pdev, (dma_addr_t) temp64.val,
1436 iocommand.buf_size,
1437 PCI_DMA_BIDIRECTIONAL);
1439 check_ioctl_unit_attention(host, c);
1441 /* Copy the error information out */
1442 iocommand.error_info = *(c->err_info);
1443 if (copy_to_user
1444 (argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1445 kfree(buff);
1446 cmd_free(host, c, 0);
1447 return -EFAULT;
1450 if (iocommand.Request.Type.Direction == XFER_READ) {
1451 /* Copy the data out of the buffer we created */
1452 if (copy_to_user
1453 (iocommand.buf, buff, iocommand.buf_size)) {
1454 kfree(buff);
1455 cmd_free(host, c, 0);
1456 return -EFAULT;
1459 kfree(buff);
1460 cmd_free(host, c, 0);
1461 return 0;
1463 case CCISS_BIG_PASSTHRU:{
1464 BIG_IOCTL_Command_struct *ioc;
1465 CommandList_struct *c;
1466 unsigned char **buff = NULL;
1467 int *buff_size = NULL;
1468 u64bit temp64;
1469 unsigned long flags;
1470 BYTE sg_used = 0;
1471 int status = 0;
1472 int i;
1473 DECLARE_COMPLETION_ONSTACK(wait);
1474 __u32 left;
1475 __u32 sz;
1476 BYTE __user *data_ptr;
1478 if (!arg)
1479 return -EINVAL;
1480 if (!capable(CAP_SYS_RAWIO))
1481 return -EPERM;
1482 ioc = (BIG_IOCTL_Command_struct *)
1483 kmalloc(sizeof(*ioc), GFP_KERNEL);
1484 if (!ioc) {
1485 status = -ENOMEM;
1486 goto cleanup1;
1488 if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1489 status = -EFAULT;
1490 goto cleanup1;
1492 if ((ioc->buf_size < 1) &&
1493 (ioc->Request.Type.Direction != XFER_NONE)) {
1494 status = -EINVAL;
1495 goto cleanup1;
1497 /* Check kmalloc limits using all SGs */
1498 if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1499 status = -EINVAL;
1500 goto cleanup1;
1502 if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1503 status = -EINVAL;
1504 goto cleanup1;
1506 buff =
1507 kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1508 if (!buff) {
1509 status = -ENOMEM;
1510 goto cleanup1;
1512 buff_size = kmalloc(MAXSGENTRIES * sizeof(int),
1513 GFP_KERNEL);
1514 if (!buff_size) {
1515 status = -ENOMEM;
1516 goto cleanup1;
1518 left = ioc->buf_size;
1519 data_ptr = ioc->buf;
1520 while (left) {
1521 sz = (left >
1522 ioc->malloc_size) ? ioc->
1523 malloc_size : left;
1524 buff_size[sg_used] = sz;
1525 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1526 if (buff[sg_used] == NULL) {
1527 status = -ENOMEM;
1528 goto cleanup1;
1530 if (ioc->Request.Type.Direction == XFER_WRITE) {
1531 if (copy_from_user
1532 (buff[sg_used], data_ptr, sz)) {
1533 status = -EFAULT;
1534 goto cleanup1;
1536 } else {
1537 memset(buff[sg_used], 0, sz);
1539 left -= sz;
1540 data_ptr += sz;
1541 sg_used++;
1543 if ((c = cmd_alloc(host, 0)) == NULL) {
1544 status = -ENOMEM;
1545 goto cleanup1;
1547 c->cmd_type = CMD_IOCTL_PEND;
1548 c->Header.ReplyQueue = 0;
1550 if (ioc->buf_size > 0) {
1551 c->Header.SGList = sg_used;
1552 c->Header.SGTotal = sg_used;
1553 } else {
1554 c->Header.SGList = 0;
1555 c->Header.SGTotal = 0;
1557 c->Header.LUN = ioc->LUN_info;
1558 c->Header.Tag.lower = c->busaddr;
1560 c->Request = ioc->Request;
1561 if (ioc->buf_size > 0) {
1562 int i;
1563 for (i = 0; i < sg_used; i++) {
1564 temp64.val =
1565 pci_map_single(host->pdev, buff[i],
1566 buff_size[i],
1567 PCI_DMA_BIDIRECTIONAL);
1568 c->SG[i].Addr.lower =
1569 temp64.val32.lower;
1570 c->SG[i].Addr.upper =
1571 temp64.val32.upper;
1572 c->SG[i].Len = buff_size[i];
1573 c->SG[i].Ext = 0; /* we are not chaining */
1576 c->waiting = &wait;
1577 /* Put the request on the tail of the request queue */
1578 spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1579 addQ(&host->reqQ, c);
1580 host->Qdepth++;
1581 start_io(host);
1582 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1583 wait_for_completion(&wait);
1584 /* unlock the buffers from DMA */
1585 for (i = 0; i < sg_used; i++) {
1586 temp64.val32.lower = c->SG[i].Addr.lower;
1587 temp64.val32.upper = c->SG[i].Addr.upper;
1588 pci_unmap_single(host->pdev,
1589 (dma_addr_t) temp64.val, buff_size[i],
1590 PCI_DMA_BIDIRECTIONAL);
1592 check_ioctl_unit_attention(host, c);
1593 /* Copy the error information out */
1594 ioc->error_info = *(c->err_info);
1595 if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1596 cmd_free(host, c, 0);
1597 status = -EFAULT;
1598 goto cleanup1;
1600 if (ioc->Request.Type.Direction == XFER_READ) {
1601 /* Copy the data out of the buffer we created */
1602 BYTE __user *ptr = ioc->buf;
1603 for (i = 0; i < sg_used; i++) {
1604 if (copy_to_user
1605 (ptr, buff[i], buff_size[i])) {
1606 cmd_free(host, c, 0);
1607 status = -EFAULT;
1608 goto cleanup1;
1610 ptr += buff_size[i];
1613 cmd_free(host, c, 0);
1614 status = 0;
1615 cleanup1:
1616 if (buff) {
1617 for (i = 0; i < sg_used; i++)
1618 kfree(buff[i]);
1619 kfree(buff);
1621 kfree(buff_size);
1622 kfree(ioc);
1623 return status;
1626 /* scsi_cmd_ioctl handles these, below, though some are not */
1627 /* very meaningful for cciss. SG_IO is the main one people want. */
1629 case SG_GET_VERSION_NUM:
1630 case SG_SET_TIMEOUT:
1631 case SG_GET_TIMEOUT:
1632 case SG_GET_RESERVED_SIZE:
1633 case SG_SET_RESERVED_SIZE:
1634 case SG_EMULATED_HOST:
1635 case SG_IO:
1636 case SCSI_IOCTL_SEND_COMMAND:
1637 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
1639 /* scsi_cmd_ioctl would normally handle these, below, but */
1640 /* they aren't a good fit for cciss, as CD-ROMs are */
1641 /* not supported, and we don't have any bus/target/lun */
1642 /* which we present to the kernel. */
1644 case CDROM_SEND_PACKET:
1645 case CDROMCLOSETRAY:
1646 case CDROMEJECT:
1647 case SCSI_IOCTL_GET_IDLUN:
1648 case SCSI_IOCTL_GET_BUS_NUMBER:
1649 default:
1650 return -ENOTTY;
1654 static void cciss_check_queues(ctlr_info_t *h)
1656 int start_queue = h->next_to_run;
1657 int i;
1659 /* check to see if we have maxed out the number of commands that can
1660 * be placed on the queue. If so then exit. We do this check here
1661 * in case the interrupt we serviced was from an ioctl and did not
1662 * free any new commands.
1664 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1665 return;
1667 /* We have room on the queue for more commands. Now we need to queue
1668 * them up. We will also keep track of the next queue to run so
1669 * that every queue gets a chance to be started first.
1671 for (i = 0; i < h->highest_lun + 1; i++) {
1672 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1673 /* make sure the disk has been added and the drive is real
1674 * because this can be called from the middle of init_one.
1676 if (!h->drv[curr_queue])
1677 continue;
1678 if (!(h->drv[curr_queue]->queue) ||
1679 !(h->drv[curr_queue]->heads))
1680 continue;
1681 blk_start_queue(h->gendisk[curr_queue]->queue);
1683 /* check to see if we have maxed out the number of commands
1684 * that can be placed on the queue.
1686 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1687 if (curr_queue == start_queue) {
1688 h->next_to_run =
1689 (start_queue + 1) % (h->highest_lun + 1);
1690 break;
1691 } else {
1692 h->next_to_run = curr_queue;
1693 break;
1699 static void cciss_softirq_done(struct request *rq)
1701 CommandList_struct *cmd = rq->completion_data;
1702 ctlr_info_t *h = hba[cmd->ctlr];
1703 SGDescriptor_struct *curr_sg = cmd->SG;
1704 unsigned long flags;
1705 u64bit temp64;
1706 int i, ddir;
1707 int sg_index = 0;
1709 if (cmd->Request.Type.Direction == XFER_READ)
1710 ddir = PCI_DMA_FROMDEVICE;
1711 else
1712 ddir = PCI_DMA_TODEVICE;
1714 /* command did not need to be retried */
1715 /* unmap the DMA mapping for all the scatter gather elements */
1716 for (i = 0; i < cmd->Header.SGList; i++) {
1717 if (curr_sg[sg_index].Ext == CCISS_SG_CHAIN) {
1718 temp64.val32.lower = cmd->SG[i].Addr.lower;
1719 temp64.val32.upper = cmd->SG[i].Addr.upper;
1720 pci_dma_sync_single_for_cpu(h->pdev, temp64.val,
1721 cmd->SG[i].Len, ddir);
1722 pci_unmap_single(h->pdev, temp64.val,
1723 cmd->SG[i].Len, ddir);
1724 /* Point to the next block */
1725 curr_sg = h->cmd_sg_list[cmd->cmdindex];
1726 sg_index = 0;
1728 temp64.val32.lower = curr_sg[sg_index].Addr.lower;
1729 temp64.val32.upper = curr_sg[sg_index].Addr.upper;
1730 pci_unmap_page(h->pdev, temp64.val, curr_sg[sg_index].Len,
1731 ddir);
1732 ++sg_index;
1735 #ifdef CCISS_DEBUG
1736 printk("Done with %p\n", rq);
1737 #endif /* CCISS_DEBUG */
1739 /* set the residual count for pc requests */
1740 if (blk_pc_request(rq))
1741 rq->resid_len = cmd->err_info->ResidualCnt;
1743 blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1745 spin_lock_irqsave(&h->lock, flags);
1746 cmd_free(h, cmd, 1);
1747 cciss_check_queues(h);
1748 spin_unlock_irqrestore(&h->lock, flags);
1751 static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1752 unsigned char scsi3addr[], uint32_t log_unit)
1754 memcpy(scsi3addr, h->drv[log_unit]->LunID,
1755 sizeof(h->drv[log_unit]->LunID));
1758 /* This function gets the SCSI vendor, model, and revision of a logical drive
1759 * via the inquiry page 0. Model, vendor, and rev are set to empty strings if
1760 * they cannot be read.
1762 static void cciss_get_device_descr(int ctlr, int logvol,
1763 char *vendor, char *model, char *rev)
1765 int rc;
1766 InquiryData_struct *inq_buf;
1767 unsigned char scsi3addr[8];
1769 *vendor = '\0';
1770 *model = '\0';
1771 *rev = '\0';
1773 inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1774 if (!inq_buf)
1775 return;
1777 log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1778 rc = sendcmd_withirq(CISS_INQUIRY, ctlr, inq_buf, sizeof(*inq_buf), 0,
1779 scsi3addr, TYPE_CMD);
1780 if (rc == IO_OK) {
1781 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1782 vendor[VENDOR_LEN] = '\0';
1783 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1784 model[MODEL_LEN] = '\0';
1785 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1786 rev[REV_LEN] = '\0';
1789 kfree(inq_buf);
1790 return;
1793 /* This function gets the serial number of a logical drive via
1794 * inquiry page 0x83. Serial no. is 16 bytes. If the serial
1795 * number cannot be had, for whatever reason, 16 bytes of 0xff
1796 * are returned instead.
1798 static void cciss_get_serial_no(int ctlr, int logvol,
1799 unsigned char *serial_no, int buflen)
1801 #define PAGE_83_INQ_BYTES 64
1802 int rc;
1803 unsigned char *buf;
1804 unsigned char scsi3addr[8];
1806 if (buflen > 16)
1807 buflen = 16;
1808 memset(serial_no, 0xff, buflen);
1809 buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1810 if (!buf)
1811 return;
1812 memset(serial_no, 0, buflen);
1813 log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1814 rc = sendcmd_withirq(CISS_INQUIRY, ctlr, buf,
1815 PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1816 if (rc == IO_OK)
1817 memcpy(serial_no, &buf[8], buflen);
1818 kfree(buf);
1819 return;
1823 * cciss_add_disk sets up the block device queue for a logical drive
1825 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1826 int drv_index)
1828 disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1829 if (!disk->queue)
1830 goto init_queue_failure;
1831 sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1832 disk->major = h->major;
1833 disk->first_minor = drv_index << NWD_SHIFT;
1834 disk->fops = &cciss_fops;
1835 if (cciss_create_ld_sysfs_entry(h, drv_index))
1836 goto cleanup_queue;
1837 disk->private_data = h->drv[drv_index];
1838 disk->driverfs_dev = &h->drv[drv_index]->dev;
1840 /* Set up queue information */
1841 blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1843 /* This is a hardware imposed limit. */
1844 blk_queue_max_segments(disk->queue, h->maxsgentries);
1846 blk_queue_max_hw_sectors(disk->queue, h->cciss_max_sectors);
1848 blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1850 disk->queue->queuedata = h;
1852 blk_queue_logical_block_size(disk->queue,
1853 h->drv[drv_index]->block_size);
1855 /* Make sure all queue data is written out before */
1856 /* setting h->drv[drv_index]->queue, as setting this */
1857 /* allows the interrupt handler to start the queue */
1858 wmb();
1859 h->drv[drv_index]->queue = disk->queue;
1860 add_disk(disk);
1861 return 0;
1863 cleanup_queue:
1864 blk_cleanup_queue(disk->queue);
1865 disk->queue = NULL;
1866 init_queue_failure:
1867 return -1;
1870 /* This function will check the usage_count of the drive to be updated/added.
1871 * If the usage_count is zero and it is a heretofore unknown drive, or,
1872 * the drive's capacity, geometry, or serial number has changed,
1873 * then the drive information will be updated and the disk will be
1874 * re-registered with the kernel. If these conditions don't hold,
1875 * then it will be left alone for the next reboot. The exception to this
1876 * is disk 0 which will always be left registered with the kernel since it
1877 * is also the controller node. Any changes to disk 0 will show up on
1878 * the next reboot.
1880 static void cciss_update_drive_info(int ctlr, int drv_index, int first_time,
1881 int via_ioctl)
1883 ctlr_info_t *h = hba[ctlr];
1884 struct gendisk *disk;
1885 InquiryData_struct *inq_buff = NULL;
1886 unsigned int block_size;
1887 sector_t total_size;
1888 unsigned long flags = 0;
1889 int ret = 0;
1890 drive_info_struct *drvinfo;
1892 /* Get information about the disk and modify the driver structure */
1893 inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1894 drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
1895 if (inq_buff == NULL || drvinfo == NULL)
1896 goto mem_msg;
1898 /* testing to see if 16-byte CDBs are already being used */
1899 if (h->cciss_read == CCISS_READ_16) {
1900 cciss_read_capacity_16(h->ctlr, drv_index,
1901 &total_size, &block_size);
1903 } else {
1904 cciss_read_capacity(ctlr, drv_index, &total_size, &block_size);
1905 /* if read_capacity returns all F's this volume is >2TB */
1906 /* in size so we switch to 16-byte CDB's for all */
1907 /* read/write ops */
1908 if (total_size == 0xFFFFFFFFULL) {
1909 cciss_read_capacity_16(ctlr, drv_index,
1910 &total_size, &block_size);
1911 h->cciss_read = CCISS_READ_16;
1912 h->cciss_write = CCISS_WRITE_16;
1913 } else {
1914 h->cciss_read = CCISS_READ_10;
1915 h->cciss_write = CCISS_WRITE_10;
1919 cciss_geometry_inquiry(ctlr, drv_index, total_size, block_size,
1920 inq_buff, drvinfo);
1921 drvinfo->block_size = block_size;
1922 drvinfo->nr_blocks = total_size + 1;
1924 cciss_get_device_descr(ctlr, drv_index, drvinfo->vendor,
1925 drvinfo->model, drvinfo->rev);
1926 cciss_get_serial_no(ctlr, drv_index, drvinfo->serial_no,
1927 sizeof(drvinfo->serial_no));
1928 /* Save the lunid in case we deregister the disk, below. */
1929 memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
1930 sizeof(drvinfo->LunID));
1932 /* Is it the same disk we already know, and nothing's changed? */
1933 if (h->drv[drv_index]->raid_level != -1 &&
1934 ((memcmp(drvinfo->serial_no,
1935 h->drv[drv_index]->serial_no, 16) == 0) &&
1936 drvinfo->block_size == h->drv[drv_index]->block_size &&
1937 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
1938 drvinfo->heads == h->drv[drv_index]->heads &&
1939 drvinfo->sectors == h->drv[drv_index]->sectors &&
1940 drvinfo->cylinders == h->drv[drv_index]->cylinders))
1941 /* The disk is unchanged, nothing to update */
1942 goto freeret;
1944 /* If we get here it's not the same disk, or something's changed,
1945 * so we need to * deregister it, and re-register it, if it's not
1946 * in use.
1947 * If the disk already exists then deregister it before proceeding
1948 * (unless it's the first disk (for the controller node).
1950 if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
1951 printk(KERN_WARNING "disk %d has changed.\n", drv_index);
1952 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1953 h->drv[drv_index]->busy_configuring = 1;
1954 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1956 /* deregister_disk sets h->drv[drv_index]->queue = NULL
1957 * which keeps the interrupt handler from starting
1958 * the queue.
1960 ret = deregister_disk(h, drv_index, 0, via_ioctl);
1963 /* If the disk is in use return */
1964 if (ret)
1965 goto freeret;
1967 /* Save the new information from cciss_geometry_inquiry
1968 * and serial number inquiry. If the disk was deregistered
1969 * above, then h->drv[drv_index] will be NULL.
1971 if (h->drv[drv_index] == NULL) {
1972 drvinfo->device_initialized = 0;
1973 h->drv[drv_index] = drvinfo;
1974 drvinfo = NULL; /* so it won't be freed below. */
1975 } else {
1976 /* special case for cxd0 */
1977 h->drv[drv_index]->block_size = drvinfo->block_size;
1978 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
1979 h->drv[drv_index]->heads = drvinfo->heads;
1980 h->drv[drv_index]->sectors = drvinfo->sectors;
1981 h->drv[drv_index]->cylinders = drvinfo->cylinders;
1982 h->drv[drv_index]->raid_level = drvinfo->raid_level;
1983 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
1984 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
1985 VENDOR_LEN + 1);
1986 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
1987 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
1990 ++h->num_luns;
1991 disk = h->gendisk[drv_index];
1992 set_capacity(disk, h->drv[drv_index]->nr_blocks);
1994 /* If it's not disk 0 (drv_index != 0)
1995 * or if it was disk 0, but there was previously
1996 * no actual corresponding configured logical drive
1997 * (raid_leve == -1) then we want to update the
1998 * logical drive's information.
2000 if (drv_index || first_time) {
2001 if (cciss_add_disk(h, disk, drv_index) != 0) {
2002 cciss_free_gendisk(h, drv_index);
2003 cciss_free_drive_info(h, drv_index);
2004 printk(KERN_WARNING "cciss:%d could not update "
2005 "disk %d\n", h->ctlr, drv_index);
2006 --h->num_luns;
2010 freeret:
2011 kfree(inq_buff);
2012 kfree(drvinfo);
2013 return;
2014 mem_msg:
2015 printk(KERN_ERR "cciss: out of memory\n");
2016 goto freeret;
2019 /* This function will find the first index of the controllers drive array
2020 * that has a null drv pointer and allocate the drive info struct and
2021 * will return that index This is where new drives will be added.
2022 * If the index to be returned is greater than the highest_lun index for
2023 * the controller then highest_lun is set * to this new index.
2024 * If there are no available indexes or if tha allocation fails, then -1
2025 * is returned. * "controller_node" is used to know if this is a real
2026 * logical drive, or just the controller node, which determines if this
2027 * counts towards highest_lun.
2029 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
2031 int i;
2032 drive_info_struct *drv;
2034 /* Search for an empty slot for our drive info */
2035 for (i = 0; i < CISS_MAX_LUN; i++) {
2037 /* if not cxd0 case, and it's occupied, skip it. */
2038 if (h->drv[i] && i != 0)
2039 continue;
2041 * If it's cxd0 case, and drv is alloc'ed already, and a
2042 * disk is configured there, skip it.
2044 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2045 continue;
2048 * We've found an empty slot. Update highest_lun
2049 * provided this isn't just the fake cxd0 controller node.
2051 if (i > h->highest_lun && !controller_node)
2052 h->highest_lun = i;
2054 /* If adding a real disk at cxd0, and it's already alloc'ed */
2055 if (i == 0 && h->drv[i] != NULL)
2056 return i;
2059 * Found an empty slot, not already alloc'ed. Allocate it.
2060 * Mark it with raid_level == -1, so we know it's new later on.
2062 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2063 if (!drv)
2064 return -1;
2065 drv->raid_level = -1; /* so we know it's new */
2066 h->drv[i] = drv;
2067 return i;
2069 return -1;
2072 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2074 kfree(h->drv[drv_index]);
2075 h->drv[drv_index] = NULL;
2078 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2080 put_disk(h->gendisk[drv_index]);
2081 h->gendisk[drv_index] = NULL;
2084 /* cciss_add_gendisk finds a free hba[]->drv structure
2085 * and allocates a gendisk if needed, and sets the lunid
2086 * in the drvinfo structure. It returns the index into
2087 * the ->drv[] array, or -1 if none are free.
2088 * is_controller_node indicates whether highest_lun should
2089 * count this disk, or if it's only being added to provide
2090 * a means to talk to the controller in case no logical
2091 * drives have yet been configured.
2093 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2094 int controller_node)
2096 int drv_index;
2098 drv_index = cciss_alloc_drive_info(h, controller_node);
2099 if (drv_index == -1)
2100 return -1;
2102 /*Check if the gendisk needs to be allocated */
2103 if (!h->gendisk[drv_index]) {
2104 h->gendisk[drv_index] =
2105 alloc_disk(1 << NWD_SHIFT);
2106 if (!h->gendisk[drv_index]) {
2107 printk(KERN_ERR "cciss%d: could not "
2108 "allocate a new disk %d\n",
2109 h->ctlr, drv_index);
2110 goto err_free_drive_info;
2113 memcpy(h->drv[drv_index]->LunID, lunid,
2114 sizeof(h->drv[drv_index]->LunID));
2115 if (cciss_create_ld_sysfs_entry(h, drv_index))
2116 goto err_free_disk;
2117 /* Don't need to mark this busy because nobody */
2118 /* else knows about this disk yet to contend */
2119 /* for access to it. */
2120 h->drv[drv_index]->busy_configuring = 0;
2121 wmb();
2122 return drv_index;
2124 err_free_disk:
2125 cciss_free_gendisk(h, drv_index);
2126 err_free_drive_info:
2127 cciss_free_drive_info(h, drv_index);
2128 return -1;
2131 /* This is for the special case of a controller which
2132 * has no logical drives. In this case, we still need
2133 * to register a disk so the controller can be accessed
2134 * by the Array Config Utility.
2136 static void cciss_add_controller_node(ctlr_info_t *h)
2138 struct gendisk *disk;
2139 int drv_index;
2141 if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2142 return;
2144 drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2145 if (drv_index == -1)
2146 goto error;
2147 h->drv[drv_index]->block_size = 512;
2148 h->drv[drv_index]->nr_blocks = 0;
2149 h->drv[drv_index]->heads = 0;
2150 h->drv[drv_index]->sectors = 0;
2151 h->drv[drv_index]->cylinders = 0;
2152 h->drv[drv_index]->raid_level = -1;
2153 memset(h->drv[drv_index]->serial_no, 0, 16);
2154 disk = h->gendisk[drv_index];
2155 if (cciss_add_disk(h, disk, drv_index) == 0)
2156 return;
2157 cciss_free_gendisk(h, drv_index);
2158 cciss_free_drive_info(h, drv_index);
2159 error:
2160 printk(KERN_WARNING "cciss%d: could not "
2161 "add disk 0.\n", h->ctlr);
2162 return;
2165 /* This function will add and remove logical drives from the Logical
2166 * drive array of the controller and maintain persistency of ordering
2167 * so that mount points are preserved until the next reboot. This allows
2168 * for the removal of logical drives in the middle of the drive array
2169 * without a re-ordering of those drives.
2170 * INPUT
2171 * h = The controller to perform the operations on
2173 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2174 int via_ioctl)
2176 int ctlr = h->ctlr;
2177 int num_luns;
2178 ReportLunData_struct *ld_buff = NULL;
2179 int return_code;
2180 int listlength = 0;
2181 int i;
2182 int drv_found;
2183 int drv_index = 0;
2184 unsigned char lunid[8] = CTLR_LUNID;
2185 unsigned long flags;
2187 if (!capable(CAP_SYS_RAWIO))
2188 return -EPERM;
2190 /* Set busy_configuring flag for this operation */
2191 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2192 if (h->busy_configuring) {
2193 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2194 return -EBUSY;
2196 h->busy_configuring = 1;
2197 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2199 ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2200 if (ld_buff == NULL)
2201 goto mem_msg;
2203 return_code = sendcmd_withirq(CISS_REPORT_LOG, ctlr, ld_buff,
2204 sizeof(ReportLunData_struct),
2205 0, CTLR_LUNID, TYPE_CMD);
2207 if (return_code == IO_OK)
2208 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2209 else { /* reading number of logical volumes failed */
2210 printk(KERN_WARNING "cciss: report logical volume"
2211 " command failed\n");
2212 listlength = 0;
2213 goto freeret;
2216 num_luns = listlength / 8; /* 8 bytes per entry */
2217 if (num_luns > CISS_MAX_LUN) {
2218 num_luns = CISS_MAX_LUN;
2219 printk(KERN_WARNING "cciss: more luns configured"
2220 " on controller than can be handled by"
2221 " this driver.\n");
2224 if (num_luns == 0)
2225 cciss_add_controller_node(h);
2227 /* Compare controller drive array to driver's drive array
2228 * to see if any drives are missing on the controller due
2229 * to action of Array Config Utility (user deletes drive)
2230 * and deregister logical drives which have disappeared.
2232 for (i = 0; i <= h->highest_lun; i++) {
2233 int j;
2234 drv_found = 0;
2236 /* skip holes in the array from already deleted drives */
2237 if (h->drv[i] == NULL)
2238 continue;
2240 for (j = 0; j < num_luns; j++) {
2241 memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2242 if (memcmp(h->drv[i]->LunID, lunid,
2243 sizeof(lunid)) == 0) {
2244 drv_found = 1;
2245 break;
2248 if (!drv_found) {
2249 /* Deregister it from the OS, it's gone. */
2250 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2251 h->drv[i]->busy_configuring = 1;
2252 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2253 return_code = deregister_disk(h, i, 1, via_ioctl);
2254 if (h->drv[i] != NULL)
2255 h->drv[i]->busy_configuring = 0;
2259 /* Compare controller drive array to driver's drive array.
2260 * Check for updates in the drive information and any new drives
2261 * on the controller due to ACU adding logical drives, or changing
2262 * a logical drive's size, etc. Reregister any new/changed drives
2264 for (i = 0; i < num_luns; i++) {
2265 int j;
2267 drv_found = 0;
2269 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2270 /* Find if the LUN is already in the drive array
2271 * of the driver. If so then update its info
2272 * if not in use. If it does not exist then find
2273 * the first free index and add it.
2275 for (j = 0; j <= h->highest_lun; j++) {
2276 if (h->drv[j] != NULL &&
2277 memcmp(h->drv[j]->LunID, lunid,
2278 sizeof(h->drv[j]->LunID)) == 0) {
2279 drv_index = j;
2280 drv_found = 1;
2281 break;
2285 /* check if the drive was found already in the array */
2286 if (!drv_found) {
2287 drv_index = cciss_add_gendisk(h, lunid, 0);
2288 if (drv_index == -1)
2289 goto freeret;
2291 cciss_update_drive_info(ctlr, drv_index, first_time,
2292 via_ioctl);
2293 } /* end for */
2295 freeret:
2296 kfree(ld_buff);
2297 h->busy_configuring = 0;
2298 /* We return -1 here to tell the ACU that we have registered/updated
2299 * all of the drives that we can and to keep it from calling us
2300 * additional times.
2302 return -1;
2303 mem_msg:
2304 printk(KERN_ERR "cciss: out of memory\n");
2305 h->busy_configuring = 0;
2306 goto freeret;
2309 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2311 /* zero out the disk size info */
2312 drive_info->nr_blocks = 0;
2313 drive_info->block_size = 0;
2314 drive_info->heads = 0;
2315 drive_info->sectors = 0;
2316 drive_info->cylinders = 0;
2317 drive_info->raid_level = -1;
2318 memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2319 memset(drive_info->model, 0, sizeof(drive_info->model));
2320 memset(drive_info->rev, 0, sizeof(drive_info->rev));
2321 memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2323 * don't clear the LUNID though, we need to remember which
2324 * one this one is.
2328 /* This function will deregister the disk and it's queue from the
2329 * kernel. It must be called with the controller lock held and the
2330 * drv structures busy_configuring flag set. It's parameters are:
2332 * disk = This is the disk to be deregistered
2333 * drv = This is the drive_info_struct associated with the disk to be
2334 * deregistered. It contains information about the disk used
2335 * by the driver.
2336 * clear_all = This flag determines whether or not the disk information
2337 * is going to be completely cleared out and the highest_lun
2338 * reset. Sometimes we want to clear out information about
2339 * the disk in preparation for re-adding it. In this case
2340 * the highest_lun should be left unchanged and the LunID
2341 * should not be cleared.
2342 * via_ioctl
2343 * This indicates whether we've reached this path via ioctl.
2344 * This affects the maximum usage count allowed for c0d0 to be messed with.
2345 * If this path is reached via ioctl(), then the max_usage_count will
2346 * be 1, as the process calling ioctl() has got to have the device open.
2347 * If we get here via sysfs, then the max usage count will be zero.
2349 static int deregister_disk(ctlr_info_t *h, int drv_index,
2350 int clear_all, int via_ioctl)
2352 int i;
2353 struct gendisk *disk;
2354 drive_info_struct *drv;
2355 int recalculate_highest_lun;
2357 if (!capable(CAP_SYS_RAWIO))
2358 return -EPERM;
2360 drv = h->drv[drv_index];
2361 disk = h->gendisk[drv_index];
2363 /* make sure logical volume is NOT is use */
2364 if (clear_all || (h->gendisk[0] == disk)) {
2365 if (drv->usage_count > via_ioctl)
2366 return -EBUSY;
2367 } else if (drv->usage_count > 0)
2368 return -EBUSY;
2370 recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2372 /* invalidate the devices and deregister the disk. If it is disk
2373 * zero do not deregister it but just zero out it's values. This
2374 * allows us to delete disk zero but keep the controller registered.
2376 if (h->gendisk[0] != disk) {
2377 struct request_queue *q = disk->queue;
2378 if (disk->flags & GENHD_FL_UP) {
2379 cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2380 del_gendisk(disk);
2382 if (q)
2383 blk_cleanup_queue(q);
2384 /* If clear_all is set then we are deleting the logical
2385 * drive, not just refreshing its info. For drives
2386 * other than disk 0 we will call put_disk. We do not
2387 * do this for disk 0 as we need it to be able to
2388 * configure the controller.
2390 if (clear_all){
2391 /* This isn't pretty, but we need to find the
2392 * disk in our array and NULL our the pointer.
2393 * This is so that we will call alloc_disk if
2394 * this index is used again later.
2396 for (i=0; i < CISS_MAX_LUN; i++){
2397 if (h->gendisk[i] == disk) {
2398 h->gendisk[i] = NULL;
2399 break;
2402 put_disk(disk);
2404 } else {
2405 set_capacity(disk, 0);
2406 cciss_clear_drive_info(drv);
2409 --h->num_luns;
2411 /* if it was the last disk, find the new hightest lun */
2412 if (clear_all && recalculate_highest_lun) {
2413 int i, newhighest = -1;
2414 for (i = 0; i <= h->highest_lun; i++) {
2415 /* if the disk has size > 0, it is available */
2416 if (h->drv[i] && h->drv[i]->heads)
2417 newhighest = i;
2419 h->highest_lun = newhighest;
2421 return 0;
2424 static int fill_cmd(CommandList_struct *c, __u8 cmd, int ctlr, void *buff,
2425 size_t size, __u8 page_code, unsigned char *scsi3addr,
2426 int cmd_type)
2428 ctlr_info_t *h = hba[ctlr];
2429 u64bit buff_dma_handle;
2430 int status = IO_OK;
2432 c->cmd_type = CMD_IOCTL_PEND;
2433 c->Header.ReplyQueue = 0;
2434 if (buff != NULL) {
2435 c->Header.SGList = 1;
2436 c->Header.SGTotal = 1;
2437 } else {
2438 c->Header.SGList = 0;
2439 c->Header.SGTotal = 0;
2441 c->Header.Tag.lower = c->busaddr;
2442 memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2444 c->Request.Type.Type = cmd_type;
2445 if (cmd_type == TYPE_CMD) {
2446 switch (cmd) {
2447 case CISS_INQUIRY:
2448 /* are we trying to read a vital product page */
2449 if (page_code != 0) {
2450 c->Request.CDB[1] = 0x01;
2451 c->Request.CDB[2] = page_code;
2453 c->Request.CDBLen = 6;
2454 c->Request.Type.Attribute = ATTR_SIMPLE;
2455 c->Request.Type.Direction = XFER_READ;
2456 c->Request.Timeout = 0;
2457 c->Request.CDB[0] = CISS_INQUIRY;
2458 c->Request.CDB[4] = size & 0xFF;
2459 break;
2460 case CISS_REPORT_LOG:
2461 case CISS_REPORT_PHYS:
2462 /* Talking to controller so It's a physical command
2463 mode = 00 target = 0. Nothing to write.
2465 c->Request.CDBLen = 12;
2466 c->Request.Type.Attribute = ATTR_SIMPLE;
2467 c->Request.Type.Direction = XFER_READ;
2468 c->Request.Timeout = 0;
2469 c->Request.CDB[0] = cmd;
2470 c->Request.CDB[6] = (size >> 24) & 0xFF; /* MSB */
2471 c->Request.CDB[7] = (size >> 16) & 0xFF;
2472 c->Request.CDB[8] = (size >> 8) & 0xFF;
2473 c->Request.CDB[9] = size & 0xFF;
2474 break;
2476 case CCISS_READ_CAPACITY:
2477 c->Request.CDBLen = 10;
2478 c->Request.Type.Attribute = ATTR_SIMPLE;
2479 c->Request.Type.Direction = XFER_READ;
2480 c->Request.Timeout = 0;
2481 c->Request.CDB[0] = cmd;
2482 break;
2483 case CCISS_READ_CAPACITY_16:
2484 c->Request.CDBLen = 16;
2485 c->Request.Type.Attribute = ATTR_SIMPLE;
2486 c->Request.Type.Direction = XFER_READ;
2487 c->Request.Timeout = 0;
2488 c->Request.CDB[0] = cmd;
2489 c->Request.CDB[1] = 0x10;
2490 c->Request.CDB[10] = (size >> 24) & 0xFF;
2491 c->Request.CDB[11] = (size >> 16) & 0xFF;
2492 c->Request.CDB[12] = (size >> 8) & 0xFF;
2493 c->Request.CDB[13] = size & 0xFF;
2494 c->Request.Timeout = 0;
2495 c->Request.CDB[0] = cmd;
2496 break;
2497 case CCISS_CACHE_FLUSH:
2498 c->Request.CDBLen = 12;
2499 c->Request.Type.Attribute = ATTR_SIMPLE;
2500 c->Request.Type.Direction = XFER_WRITE;
2501 c->Request.Timeout = 0;
2502 c->Request.CDB[0] = BMIC_WRITE;
2503 c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2504 break;
2505 case TEST_UNIT_READY:
2506 c->Request.CDBLen = 6;
2507 c->Request.Type.Attribute = ATTR_SIMPLE;
2508 c->Request.Type.Direction = XFER_NONE;
2509 c->Request.Timeout = 0;
2510 break;
2511 default:
2512 printk(KERN_WARNING
2513 "cciss%d: Unknown Command 0x%c\n", ctlr, cmd);
2514 return IO_ERROR;
2516 } else if (cmd_type == TYPE_MSG) {
2517 switch (cmd) {
2518 case 0: /* ABORT message */
2519 c->Request.CDBLen = 12;
2520 c->Request.Type.Attribute = ATTR_SIMPLE;
2521 c->Request.Type.Direction = XFER_WRITE;
2522 c->Request.Timeout = 0;
2523 c->Request.CDB[0] = cmd; /* abort */
2524 c->Request.CDB[1] = 0; /* abort a command */
2525 /* buff contains the tag of the command to abort */
2526 memcpy(&c->Request.CDB[4], buff, 8);
2527 break;
2528 case 1: /* RESET message */
2529 c->Request.CDBLen = 16;
2530 c->Request.Type.Attribute = ATTR_SIMPLE;
2531 c->Request.Type.Direction = XFER_NONE;
2532 c->Request.Timeout = 0;
2533 memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2534 c->Request.CDB[0] = cmd; /* reset */
2535 c->Request.CDB[1] = 0x03; /* reset a target */
2536 break;
2537 case 3: /* No-Op message */
2538 c->Request.CDBLen = 1;
2539 c->Request.Type.Attribute = ATTR_SIMPLE;
2540 c->Request.Type.Direction = XFER_WRITE;
2541 c->Request.Timeout = 0;
2542 c->Request.CDB[0] = cmd;
2543 break;
2544 default:
2545 printk(KERN_WARNING
2546 "cciss%d: unknown message type %d\n", ctlr, cmd);
2547 return IO_ERROR;
2549 } else {
2550 printk(KERN_WARNING
2551 "cciss%d: unknown command type %d\n", ctlr, cmd_type);
2552 return IO_ERROR;
2554 /* Fill in the scatter gather information */
2555 if (size > 0) {
2556 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2557 buff, size,
2558 PCI_DMA_BIDIRECTIONAL);
2559 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2560 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2561 c->SG[0].Len = size;
2562 c->SG[0].Ext = 0; /* we are not chaining */
2564 return status;
2567 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2569 switch (c->err_info->ScsiStatus) {
2570 case SAM_STAT_GOOD:
2571 return IO_OK;
2572 case SAM_STAT_CHECK_CONDITION:
2573 switch (0xf & c->err_info->SenseInfo[2]) {
2574 case 0: return IO_OK; /* no sense */
2575 case 1: return IO_OK; /* recovered error */
2576 default:
2577 if (check_for_unit_attention(h, c))
2578 return IO_NEEDS_RETRY;
2579 printk(KERN_WARNING "cciss%d: cmd 0x%02x "
2580 "check condition, sense key = 0x%02x\n",
2581 h->ctlr, c->Request.CDB[0],
2582 c->err_info->SenseInfo[2]);
2584 break;
2585 default:
2586 printk(KERN_WARNING "cciss%d: cmd 0x%02x"
2587 "scsi status = 0x%02x\n", h->ctlr,
2588 c->Request.CDB[0], c->err_info->ScsiStatus);
2589 break;
2591 return IO_ERROR;
2594 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2596 int return_status = IO_OK;
2598 if (c->err_info->CommandStatus == CMD_SUCCESS)
2599 return IO_OK;
2601 switch (c->err_info->CommandStatus) {
2602 case CMD_TARGET_STATUS:
2603 return_status = check_target_status(h, c);
2604 break;
2605 case CMD_DATA_UNDERRUN:
2606 case CMD_DATA_OVERRUN:
2607 /* expected for inquiry and report lun commands */
2608 break;
2609 case CMD_INVALID:
2610 printk(KERN_WARNING "cciss: cmd 0x%02x is "
2611 "reported invalid\n", c->Request.CDB[0]);
2612 return_status = IO_ERROR;
2613 break;
2614 case CMD_PROTOCOL_ERR:
2615 printk(KERN_WARNING "cciss: cmd 0x%02x has "
2616 "protocol error \n", c->Request.CDB[0]);
2617 return_status = IO_ERROR;
2618 break;
2619 case CMD_HARDWARE_ERR:
2620 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2621 " hardware error\n", c->Request.CDB[0]);
2622 return_status = IO_ERROR;
2623 break;
2624 case CMD_CONNECTION_LOST:
2625 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2626 "connection lost\n", c->Request.CDB[0]);
2627 return_status = IO_ERROR;
2628 break;
2629 case CMD_ABORTED:
2630 printk(KERN_WARNING "cciss: cmd 0x%02x was "
2631 "aborted\n", c->Request.CDB[0]);
2632 return_status = IO_ERROR;
2633 break;
2634 case CMD_ABORT_FAILED:
2635 printk(KERN_WARNING "cciss: cmd 0x%02x reports "
2636 "abort failed\n", c->Request.CDB[0]);
2637 return_status = IO_ERROR;
2638 break;
2639 case CMD_UNSOLICITED_ABORT:
2640 printk(KERN_WARNING
2641 "cciss%d: unsolicited abort 0x%02x\n", h->ctlr,
2642 c->Request.CDB[0]);
2643 return_status = IO_NEEDS_RETRY;
2644 break;
2645 default:
2646 printk(KERN_WARNING "cciss: cmd 0x%02x returned "
2647 "unknown status %x\n", c->Request.CDB[0],
2648 c->err_info->CommandStatus);
2649 return_status = IO_ERROR;
2651 return return_status;
2654 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2655 int attempt_retry)
2657 DECLARE_COMPLETION_ONSTACK(wait);
2658 u64bit buff_dma_handle;
2659 unsigned long flags;
2660 int return_status = IO_OK;
2662 resend_cmd2:
2663 c->waiting = &wait;
2664 /* Put the request on the tail of the queue and send it */
2665 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2666 addQ(&h->reqQ, c);
2667 h->Qdepth++;
2668 start_io(h);
2669 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2671 wait_for_completion(&wait);
2673 if (c->err_info->CommandStatus == 0 || !attempt_retry)
2674 goto command_done;
2676 return_status = process_sendcmd_error(h, c);
2678 if (return_status == IO_NEEDS_RETRY &&
2679 c->retry_count < MAX_CMD_RETRIES) {
2680 printk(KERN_WARNING "cciss%d: retrying 0x%02x\n", h->ctlr,
2681 c->Request.CDB[0]);
2682 c->retry_count++;
2683 /* erase the old error information */
2684 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2685 return_status = IO_OK;
2686 INIT_COMPLETION(wait);
2687 goto resend_cmd2;
2690 command_done:
2691 /* unlock the buffers from DMA */
2692 buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2693 buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2694 pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2695 c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2696 return return_status;
2699 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
2700 __u8 page_code, unsigned char scsi3addr[],
2701 int cmd_type)
2703 ctlr_info_t *h = hba[ctlr];
2704 CommandList_struct *c;
2705 int return_status;
2707 c = cmd_alloc(h, 0);
2708 if (!c)
2709 return -ENOMEM;
2710 return_status = fill_cmd(c, cmd, ctlr, buff, size, page_code,
2711 scsi3addr, cmd_type);
2712 if (return_status == IO_OK)
2713 return_status = sendcmd_withirq_core(h, c, 1);
2715 cmd_free(h, c, 0);
2716 return return_status;
2719 static void cciss_geometry_inquiry(int ctlr, int logvol,
2720 sector_t total_size,
2721 unsigned int block_size,
2722 InquiryData_struct *inq_buff,
2723 drive_info_struct *drv)
2725 int return_code;
2726 unsigned long t;
2727 unsigned char scsi3addr[8];
2729 memset(inq_buff, 0, sizeof(InquiryData_struct));
2730 log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2731 return_code = sendcmd_withirq(CISS_INQUIRY, ctlr, inq_buff,
2732 sizeof(*inq_buff), 0xC1, scsi3addr, TYPE_CMD);
2733 if (return_code == IO_OK) {
2734 if (inq_buff->data_byte[8] == 0xFF) {
2735 printk(KERN_WARNING
2736 "cciss: reading geometry failed, volume "
2737 "does not support reading geometry\n");
2738 drv->heads = 255;
2739 drv->sectors = 32; /* Sectors per track */
2740 drv->cylinders = total_size + 1;
2741 drv->raid_level = RAID_UNKNOWN;
2742 } else {
2743 drv->heads = inq_buff->data_byte[6];
2744 drv->sectors = inq_buff->data_byte[7];
2745 drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2746 drv->cylinders += inq_buff->data_byte[5];
2747 drv->raid_level = inq_buff->data_byte[8];
2749 drv->block_size = block_size;
2750 drv->nr_blocks = total_size + 1;
2751 t = drv->heads * drv->sectors;
2752 if (t > 1) {
2753 sector_t real_size = total_size + 1;
2754 unsigned long rem = sector_div(real_size, t);
2755 if (rem)
2756 real_size++;
2757 drv->cylinders = real_size;
2759 } else { /* Get geometry failed */
2760 printk(KERN_WARNING "cciss: reading geometry failed\n");
2764 static void
2765 cciss_read_capacity(int ctlr, int logvol, sector_t *total_size,
2766 unsigned int *block_size)
2768 ReadCapdata_struct *buf;
2769 int return_code;
2770 unsigned char scsi3addr[8];
2772 buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2773 if (!buf) {
2774 printk(KERN_WARNING "cciss: out of memory\n");
2775 return;
2778 log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2779 return_code = sendcmd_withirq(CCISS_READ_CAPACITY, ctlr, buf,
2780 sizeof(ReadCapdata_struct), 0, scsi3addr, TYPE_CMD);
2781 if (return_code == IO_OK) {
2782 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2783 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2784 } else { /* read capacity command failed */
2785 printk(KERN_WARNING "cciss: read capacity failed\n");
2786 *total_size = 0;
2787 *block_size = BLOCK_SIZE;
2789 kfree(buf);
2792 static void cciss_read_capacity_16(int ctlr, int logvol,
2793 sector_t *total_size, unsigned int *block_size)
2795 ReadCapdata_struct_16 *buf;
2796 int return_code;
2797 unsigned char scsi3addr[8];
2799 buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2800 if (!buf) {
2801 printk(KERN_WARNING "cciss: out of memory\n");
2802 return;
2805 log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2806 return_code = sendcmd_withirq(CCISS_READ_CAPACITY_16,
2807 ctlr, buf, sizeof(ReadCapdata_struct_16),
2808 0, scsi3addr, TYPE_CMD);
2809 if (return_code == IO_OK) {
2810 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2811 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2812 } else { /* read capacity command failed */
2813 printk(KERN_WARNING "cciss: read capacity failed\n");
2814 *total_size = 0;
2815 *block_size = BLOCK_SIZE;
2817 printk(KERN_INFO " blocks= %llu block_size= %d\n",
2818 (unsigned long long)*total_size+1, *block_size);
2819 kfree(buf);
2822 static int cciss_revalidate(struct gendisk *disk)
2824 ctlr_info_t *h = get_host(disk);
2825 drive_info_struct *drv = get_drv(disk);
2826 int logvol;
2827 int FOUND = 0;
2828 unsigned int block_size;
2829 sector_t total_size;
2830 InquiryData_struct *inq_buff = NULL;
2832 for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
2833 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2834 sizeof(drv->LunID)) == 0) {
2835 FOUND = 1;
2836 break;
2840 if (!FOUND)
2841 return 1;
2843 inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2844 if (inq_buff == NULL) {
2845 printk(KERN_WARNING "cciss: out of memory\n");
2846 return 1;
2848 if (h->cciss_read == CCISS_READ_10) {
2849 cciss_read_capacity(h->ctlr, logvol,
2850 &total_size, &block_size);
2851 } else {
2852 cciss_read_capacity_16(h->ctlr, logvol,
2853 &total_size, &block_size);
2855 cciss_geometry_inquiry(h->ctlr, logvol, total_size, block_size,
2856 inq_buff, drv);
2858 blk_queue_logical_block_size(drv->queue, drv->block_size);
2859 set_capacity(disk, drv->nr_blocks);
2861 kfree(inq_buff);
2862 return 0;
2866 * Map (physical) PCI mem into (virtual) kernel space
2868 static void __iomem *remap_pci_mem(ulong base, ulong size)
2870 ulong page_base = ((ulong) base) & PAGE_MASK;
2871 ulong page_offs = ((ulong) base) - page_base;
2872 void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2874 return page_remapped ? (page_remapped + page_offs) : NULL;
2878 * Takes jobs of the Q and sends them to the hardware, then puts it on
2879 * the Q to wait for completion.
2881 static void start_io(ctlr_info_t *h)
2883 CommandList_struct *c;
2885 while (!hlist_empty(&h->reqQ)) {
2886 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
2887 /* can't do anything if fifo is full */
2888 if ((h->access.fifo_full(h))) {
2889 printk(KERN_WARNING "cciss: fifo full\n");
2890 break;
2893 /* Get the first entry from the Request Q */
2894 removeQ(c);
2895 h->Qdepth--;
2897 /* Tell the controller execute command */
2898 h->access.submit_command(h, c);
2900 /* Put job onto the completed Q */
2901 addQ(&h->cmpQ, c);
2905 /* Assumes that CCISS_LOCK(h->ctlr) is held. */
2906 /* Zeros out the error record and then resends the command back */
2907 /* to the controller */
2908 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2910 /* erase the old error information */
2911 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2913 /* add it to software queue and then send it to the controller */
2914 addQ(&h->reqQ, c);
2915 h->Qdepth++;
2916 if (h->Qdepth > h->maxQsinceinit)
2917 h->maxQsinceinit = h->Qdepth;
2919 start_io(h);
2922 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
2923 unsigned int msg_byte, unsigned int host_byte,
2924 unsigned int driver_byte)
2926 /* inverse of macros in scsi.h */
2927 return (scsi_status_byte & 0xff) |
2928 ((msg_byte & 0xff) << 8) |
2929 ((host_byte & 0xff) << 16) |
2930 ((driver_byte & 0xff) << 24);
2933 static inline int evaluate_target_status(ctlr_info_t *h,
2934 CommandList_struct *cmd, int *retry_cmd)
2936 unsigned char sense_key;
2937 unsigned char status_byte, msg_byte, host_byte, driver_byte;
2938 int error_value;
2940 *retry_cmd = 0;
2941 /* If we get in here, it means we got "target status", that is, scsi status */
2942 status_byte = cmd->err_info->ScsiStatus;
2943 driver_byte = DRIVER_OK;
2944 msg_byte = cmd->err_info->CommandStatus; /* correct? seems too device specific */
2946 if (blk_pc_request(cmd->rq))
2947 host_byte = DID_PASSTHROUGH;
2948 else
2949 host_byte = DID_OK;
2951 error_value = make_status_bytes(status_byte, msg_byte,
2952 host_byte, driver_byte);
2954 if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
2955 if (!blk_pc_request(cmd->rq))
2956 printk(KERN_WARNING "cciss: cmd %p "
2957 "has SCSI Status 0x%x\n",
2958 cmd, cmd->err_info->ScsiStatus);
2959 return error_value;
2962 /* check the sense key */
2963 sense_key = 0xf & cmd->err_info->SenseInfo[2];
2964 /* no status or recovered error */
2965 if (((sense_key == 0x0) || (sense_key == 0x1)) && !blk_pc_request(cmd->rq))
2966 error_value = 0;
2968 if (check_for_unit_attention(h, cmd)) {
2969 *retry_cmd = !blk_pc_request(cmd->rq);
2970 return 0;
2973 if (!blk_pc_request(cmd->rq)) { /* Not SG_IO or similar? */
2974 if (error_value != 0)
2975 printk(KERN_WARNING "cciss: cmd %p has CHECK CONDITION"
2976 " sense key = 0x%x\n", cmd, sense_key);
2977 return error_value;
2980 /* SG_IO or similar, copy sense data back */
2981 if (cmd->rq->sense) {
2982 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
2983 cmd->rq->sense_len = cmd->err_info->SenseLen;
2984 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
2985 cmd->rq->sense_len);
2986 } else
2987 cmd->rq->sense_len = 0;
2989 return error_value;
2992 /* checks the status of the job and calls complete buffers to mark all
2993 * buffers for the completed job. Note that this function does not need
2994 * to hold the hba/queue lock.
2996 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
2997 int timeout)
2999 int retry_cmd = 0;
3000 struct request *rq = cmd->rq;
3002 rq->errors = 0;
3004 if (timeout)
3005 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
3007 if (cmd->err_info->CommandStatus == 0) /* no error has occurred */
3008 goto after_error_processing;
3010 switch (cmd->err_info->CommandStatus) {
3011 case CMD_TARGET_STATUS:
3012 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
3013 break;
3014 case CMD_DATA_UNDERRUN:
3015 if (blk_fs_request(cmd->rq)) {
3016 printk(KERN_WARNING "cciss: cmd %p has"
3017 " completed with data underrun "
3018 "reported\n", cmd);
3019 cmd->rq->resid_len = cmd->err_info->ResidualCnt;
3021 break;
3022 case CMD_DATA_OVERRUN:
3023 if (blk_fs_request(cmd->rq))
3024 printk(KERN_WARNING "cciss: cmd %p has"
3025 " completed with data overrun "
3026 "reported\n", cmd);
3027 break;
3028 case CMD_INVALID:
3029 printk(KERN_WARNING "cciss: cmd %p is "
3030 "reported invalid\n", cmd);
3031 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3032 cmd->err_info->CommandStatus, DRIVER_OK,
3033 blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3034 break;
3035 case CMD_PROTOCOL_ERR:
3036 printk(KERN_WARNING "cciss: cmd %p has "
3037 "protocol error \n", cmd);
3038 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3039 cmd->err_info->CommandStatus, DRIVER_OK,
3040 blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3041 break;
3042 case CMD_HARDWARE_ERR:
3043 printk(KERN_WARNING "cciss: cmd %p had "
3044 " hardware error\n", cmd);
3045 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3046 cmd->err_info->CommandStatus, DRIVER_OK,
3047 blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3048 break;
3049 case CMD_CONNECTION_LOST:
3050 printk(KERN_WARNING "cciss: cmd %p had "
3051 "connection lost\n", cmd);
3052 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3053 cmd->err_info->CommandStatus, DRIVER_OK,
3054 blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3055 break;
3056 case CMD_ABORTED:
3057 printk(KERN_WARNING "cciss: cmd %p was "
3058 "aborted\n", cmd);
3059 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3060 cmd->err_info->CommandStatus, DRIVER_OK,
3061 blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
3062 break;
3063 case CMD_ABORT_FAILED:
3064 printk(KERN_WARNING "cciss: cmd %p reports "
3065 "abort failed\n", cmd);
3066 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3067 cmd->err_info->CommandStatus, DRIVER_OK,
3068 blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3069 break;
3070 case CMD_UNSOLICITED_ABORT:
3071 printk(KERN_WARNING "cciss%d: unsolicited "
3072 "abort %p\n", h->ctlr, cmd);
3073 if (cmd->retry_count < MAX_CMD_RETRIES) {
3074 retry_cmd = 1;
3075 printk(KERN_WARNING
3076 "cciss%d: retrying %p\n", h->ctlr, cmd);
3077 cmd->retry_count++;
3078 } else
3079 printk(KERN_WARNING
3080 "cciss%d: %p retried too "
3081 "many times\n", h->ctlr, cmd);
3082 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3083 cmd->err_info->CommandStatus, DRIVER_OK,
3084 blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
3085 break;
3086 case CMD_TIMEOUT:
3087 printk(KERN_WARNING "cciss: cmd %p timedout\n", cmd);
3088 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3089 cmd->err_info->CommandStatus, DRIVER_OK,
3090 blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3091 break;
3092 default:
3093 printk(KERN_WARNING "cciss: cmd %p returned "
3094 "unknown status %x\n", cmd,
3095 cmd->err_info->CommandStatus);
3096 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3097 cmd->err_info->CommandStatus, DRIVER_OK,
3098 blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3101 after_error_processing:
3103 /* We need to return this command */
3104 if (retry_cmd) {
3105 resend_cciss_cmd(h, cmd);
3106 return;
3108 cmd->rq->completion_data = cmd;
3109 blk_complete_request(cmd->rq);
3113 * Get a request and submit it to the controller.
3115 static void do_cciss_request(struct request_queue *q)
3117 ctlr_info_t *h = q->queuedata;
3118 CommandList_struct *c;
3119 sector_t start_blk;
3120 int seg;
3121 struct request *creq;
3122 u64bit temp64;
3123 struct scatterlist *tmp_sg;
3124 SGDescriptor_struct *curr_sg;
3125 drive_info_struct *drv;
3126 int i, dir;
3127 int nseg = 0;
3128 int sg_index = 0;
3129 int chained = 0;
3131 /* We call start_io here in case there is a command waiting on the
3132 * queue that has not been sent.
3134 if (blk_queue_plugged(q))
3135 goto startio;
3137 queue:
3138 creq = blk_peek_request(q);
3139 if (!creq)
3140 goto startio;
3142 BUG_ON(creq->nr_phys_segments > h->maxsgentries);
3144 if ((c = cmd_alloc(h, 1)) == NULL)
3145 goto full;
3147 blk_start_request(creq);
3149 tmp_sg = h->scatter_list[c->cmdindex];
3150 spin_unlock_irq(q->queue_lock);
3152 c->cmd_type = CMD_RWREQ;
3153 c->rq = creq;
3155 /* fill in the request */
3156 drv = creq->rq_disk->private_data;
3157 c->Header.ReplyQueue = 0; /* unused in simple mode */
3158 /* got command from pool, so use the command block index instead */
3159 /* for direct lookups. */
3160 /* The first 2 bits are reserved for controller error reporting. */
3161 c->Header.Tag.lower = (c->cmdindex << 3);
3162 c->Header.Tag.lower |= 0x04; /* flag for direct lookup. */
3163 memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3164 c->Request.CDBLen = 10; /* 12 byte commands not in FW yet; */
3165 c->Request.Type.Type = TYPE_CMD; /* It is a command. */
3166 c->Request.Type.Attribute = ATTR_SIMPLE;
3167 c->Request.Type.Direction =
3168 (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3169 c->Request.Timeout = 0; /* Don't time out */
3170 c->Request.CDB[0] =
3171 (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3172 start_blk = blk_rq_pos(creq);
3173 #ifdef CCISS_DEBUG
3174 printk(KERN_DEBUG "ciss: sector =%d nr_sectors=%d\n",
3175 (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3176 #endif /* CCISS_DEBUG */
3178 sg_init_table(tmp_sg, h->maxsgentries);
3179 seg = blk_rq_map_sg(q, creq, tmp_sg);
3181 /* get the DMA records for the setup */
3182 if (c->Request.Type.Direction == XFER_READ)
3183 dir = PCI_DMA_FROMDEVICE;
3184 else
3185 dir = PCI_DMA_TODEVICE;
3187 curr_sg = c->SG;
3188 sg_index = 0;
3189 chained = 0;
3191 for (i = 0; i < seg; i++) {
3192 if (((sg_index+1) == (h->max_cmd_sgentries)) &&
3193 !chained && ((seg - i) > 1)) {
3194 nseg = seg - i;
3195 curr_sg[sg_index].Len = (nseg) *
3196 sizeof(SGDescriptor_struct);
3197 curr_sg[sg_index].Ext = CCISS_SG_CHAIN;
3199 /* Point to next chain block. */
3200 curr_sg = h->cmd_sg_list[c->cmdindex];
3201 sg_index = 0;
3202 chained = 1;
3204 curr_sg[sg_index].Len = tmp_sg[i].length;
3205 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3206 tmp_sg[i].offset,
3207 tmp_sg[i].length, dir);
3208 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3209 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3210 curr_sg[sg_index].Ext = 0; /* we are not chaining */
3212 ++sg_index;
3215 if (chained) {
3216 int len;
3217 dma_addr_t dma_addr;
3218 curr_sg = c->SG;
3219 sg_index = h->max_cmd_sgentries - 1;
3220 len = curr_sg[sg_index].Len;
3221 /* Setup pointer to next chain block.
3222 * Fill out last element in current chain
3223 * block with address of next chain block.
3225 temp64.val = pci_map_single(h->pdev,
3226 h->cmd_sg_list[c->cmdindex], len, dir);
3227 dma_addr = temp64.val;
3228 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3229 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3230 pci_dma_sync_single_for_device(h->pdev, dma_addr, len, dir);
3233 /* track how many SG entries we are using */
3234 if (seg > h->maxSG)
3235 h->maxSG = seg;
3237 #ifdef CCISS_DEBUG
3238 printk(KERN_DEBUG "cciss: Submitting %ld sectors in %d segments "
3239 "chained[%d]\n",
3240 blk_rq_sectors(creq), seg, chained);
3241 #endif /* CCISS_DEBUG */
3243 c->Header.SGList = c->Header.SGTotal = seg + chained;
3244 if (seg > h->max_cmd_sgentries)
3245 c->Header.SGList = h->max_cmd_sgentries;
3247 if (likely(blk_fs_request(creq))) {
3248 if(h->cciss_read == CCISS_READ_10) {
3249 c->Request.CDB[1] = 0;
3250 c->Request.CDB[2] = (start_blk >> 24) & 0xff; /* MSB */
3251 c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3252 c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3253 c->Request.CDB[5] = start_blk & 0xff;
3254 c->Request.CDB[6] = 0; /* (sect >> 24) & 0xff; MSB */
3255 c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3256 c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3257 c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3258 } else {
3259 u32 upper32 = upper_32_bits(start_blk);
3261 c->Request.CDBLen = 16;
3262 c->Request.CDB[1]= 0;
3263 c->Request.CDB[2]= (upper32 >> 24) & 0xff; /* MSB */
3264 c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3265 c->Request.CDB[4]= (upper32 >> 8) & 0xff;
3266 c->Request.CDB[5]= upper32 & 0xff;
3267 c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3268 c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3269 c->Request.CDB[8]= (start_blk >> 8) & 0xff;
3270 c->Request.CDB[9]= start_blk & 0xff;
3271 c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3272 c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3273 c->Request.CDB[12]= (blk_rq_sectors(creq) >> 8) & 0xff;
3274 c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3275 c->Request.CDB[14] = c->Request.CDB[15] = 0;
3277 } else if (blk_pc_request(creq)) {
3278 c->Request.CDBLen = creq->cmd_len;
3279 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3280 } else {
3281 printk(KERN_WARNING "cciss%d: bad request type %d\n", h->ctlr, creq->cmd_type);
3282 BUG();
3285 spin_lock_irq(q->queue_lock);
3287 addQ(&h->reqQ, c);
3288 h->Qdepth++;
3289 if (h->Qdepth > h->maxQsinceinit)
3290 h->maxQsinceinit = h->Qdepth;
3292 goto queue;
3293 full:
3294 blk_stop_queue(q);
3295 startio:
3296 /* We will already have the driver lock here so not need
3297 * to lock it.
3299 start_io(h);
3302 static inline unsigned long get_next_completion(ctlr_info_t *h)
3304 return h->access.command_completed(h);
3307 static inline int interrupt_pending(ctlr_info_t *h)
3309 return h->access.intr_pending(h);
3312 static inline long interrupt_not_for_us(ctlr_info_t *h)
3314 return (((h->access.intr_pending(h) == 0) ||
3315 (h->interrupts_enabled == 0)));
3318 static irqreturn_t do_cciss_intr(int irq, void *dev_id)
3320 ctlr_info_t *h = dev_id;
3321 CommandList_struct *c;
3322 unsigned long flags;
3323 __u32 a, a1, a2;
3325 if (interrupt_not_for_us(h))
3326 return IRQ_NONE;
3328 * If there are completed commands in the completion queue,
3329 * we had better do something about it.
3331 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
3332 while (interrupt_pending(h)) {
3333 while ((a = get_next_completion(h)) != FIFO_EMPTY) {
3334 a1 = a;
3335 if ((a & 0x04)) {
3336 a2 = (a >> 3);
3337 if (a2 >= h->nr_cmds) {
3338 printk(KERN_WARNING
3339 "cciss: controller cciss%d failed, stopping.\n",
3340 h->ctlr);
3341 fail_all_cmds(h->ctlr);
3342 return IRQ_HANDLED;
3345 c = h->cmd_pool + a2;
3346 a = c->busaddr;
3348 } else {
3349 struct hlist_node *tmp;
3351 a &= ~3;
3352 c = NULL;
3353 hlist_for_each_entry(c, tmp, &h->cmpQ, list) {
3354 if (c->busaddr == a)
3355 break;
3359 * If we've found the command, take it off the
3360 * completion Q and free it
3362 if (c && c->busaddr == a) {
3363 removeQ(c);
3364 if (c->cmd_type == CMD_RWREQ) {
3365 complete_command(h, c, 0);
3366 } else if (c->cmd_type == CMD_IOCTL_PEND) {
3367 complete(c->waiting);
3369 # ifdef CONFIG_CISS_SCSI_TAPE
3370 else if (c->cmd_type == CMD_SCSI)
3371 complete_scsi_command(c, 0, a1);
3372 # endif
3373 continue;
3378 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
3379 return IRQ_HANDLED;
3383 * add_to_scan_list() - add controller to rescan queue
3384 * @h: Pointer to the controller.
3386 * Adds the controller to the rescan queue if not already on the queue.
3388 * returns 1 if added to the queue, 0 if skipped (could be on the
3389 * queue already, or the controller could be initializing or shutting
3390 * down).
3392 static int add_to_scan_list(struct ctlr_info *h)
3394 struct ctlr_info *test_h;
3395 int found = 0;
3396 int ret = 0;
3398 if (h->busy_initializing)
3399 return 0;
3401 if (!mutex_trylock(&h->busy_shutting_down))
3402 return 0;
3404 mutex_lock(&scan_mutex);
3405 list_for_each_entry(test_h, &scan_q, scan_list) {
3406 if (test_h == h) {
3407 found = 1;
3408 break;
3411 if (!found && !h->busy_scanning) {
3412 INIT_COMPLETION(h->scan_wait);
3413 list_add_tail(&h->scan_list, &scan_q);
3414 ret = 1;
3416 mutex_unlock(&scan_mutex);
3417 mutex_unlock(&h->busy_shutting_down);
3419 return ret;
3423 * remove_from_scan_list() - remove controller from rescan queue
3424 * @h: Pointer to the controller.
3426 * Removes the controller from the rescan queue if present. Blocks if
3427 * the controller is currently conducting a rescan. The controller
3428 * can be in one of three states:
3429 * 1. Doesn't need a scan
3430 * 2. On the scan list, but not scanning yet (we remove it)
3431 * 3. Busy scanning (and not on the list). In this case we want to wait for
3432 * the scan to complete to make sure the scanning thread for this
3433 * controller is completely idle.
3435 static void remove_from_scan_list(struct ctlr_info *h)
3437 struct ctlr_info *test_h, *tmp_h;
3439 mutex_lock(&scan_mutex);
3440 list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3441 if (test_h == h) { /* state 2. */
3442 list_del(&h->scan_list);
3443 complete_all(&h->scan_wait);
3444 mutex_unlock(&scan_mutex);
3445 return;
3448 if (h->busy_scanning) { /* state 3. */
3449 mutex_unlock(&scan_mutex);
3450 wait_for_completion(&h->scan_wait);
3451 } else { /* state 1, nothing to do. */
3452 mutex_unlock(&scan_mutex);
3457 * scan_thread() - kernel thread used to rescan controllers
3458 * @data: Ignored.
3460 * A kernel thread used scan for drive topology changes on
3461 * controllers. The thread processes only one controller at a time
3462 * using a queue. Controllers are added to the queue using
3463 * add_to_scan_list() and removed from the queue either after done
3464 * processing or using remove_from_scan_list().
3466 * returns 0.
3468 static int scan_thread(void *data)
3470 struct ctlr_info *h;
3472 while (1) {
3473 set_current_state(TASK_INTERRUPTIBLE);
3474 schedule();
3475 if (kthread_should_stop())
3476 break;
3478 while (1) {
3479 mutex_lock(&scan_mutex);
3480 if (list_empty(&scan_q)) {
3481 mutex_unlock(&scan_mutex);
3482 break;
3485 h = list_entry(scan_q.next,
3486 struct ctlr_info,
3487 scan_list);
3488 list_del(&h->scan_list);
3489 h->busy_scanning = 1;
3490 mutex_unlock(&scan_mutex);
3492 rebuild_lun_table(h, 0, 0);
3493 complete_all(&h->scan_wait);
3494 mutex_lock(&scan_mutex);
3495 h->busy_scanning = 0;
3496 mutex_unlock(&scan_mutex);
3500 return 0;
3503 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3505 if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3506 return 0;
3508 switch (c->err_info->SenseInfo[12]) {
3509 case STATE_CHANGED:
3510 printk(KERN_WARNING "cciss%d: a state change "
3511 "detected, command retried\n", h->ctlr);
3512 return 1;
3513 break;
3514 case LUN_FAILED:
3515 printk(KERN_WARNING "cciss%d: LUN failure "
3516 "detected, action required\n", h->ctlr);
3517 return 1;
3518 break;
3519 case REPORT_LUNS_CHANGED:
3520 printk(KERN_WARNING "cciss%d: report LUN data "
3521 "changed\n", h->ctlr);
3523 * Here, we could call add_to_scan_list and wake up the scan thread,
3524 * except that it's quite likely that we will get more than one
3525 * REPORT_LUNS_CHANGED condition in quick succession, which means
3526 * that those which occur after the first one will likely happen
3527 * *during* the scan_thread's rescan. And the rescan code is not
3528 * robust enough to restart in the middle, undoing what it has already
3529 * done, and it's not clear that it's even possible to do this, since
3530 * part of what it does is notify the block layer, which starts
3531 * doing it's own i/o to read partition tables and so on, and the
3532 * driver doesn't have visibility to know what might need undoing.
3533 * In any event, if possible, it is horribly complicated to get right
3534 * so we just don't do it for now.
3536 * Note: this REPORT_LUNS_CHANGED condition only occurs on the MSA2012.
3538 return 1;
3539 break;
3540 case POWER_OR_RESET:
3541 printk(KERN_WARNING "cciss%d: a power on "
3542 "or device reset detected\n", h->ctlr);
3543 return 1;
3544 break;
3545 case UNIT_ATTENTION_CLEARED:
3546 printk(KERN_WARNING "cciss%d: unit attention "
3547 "cleared by another initiator\n", h->ctlr);
3548 return 1;
3549 break;
3550 default:
3551 printk(KERN_WARNING "cciss%d: unknown "
3552 "unit attention detected\n", h->ctlr);
3553 return 1;
3558 * We cannot read the structure directly, for portability we must use
3559 * the io functions.
3560 * This is for debug only.
3562 #ifdef CCISS_DEBUG
3563 static void print_cfg_table(CfgTable_struct *tb)
3565 int i;
3566 char temp_name[17];
3568 printk("Controller Configuration information\n");
3569 printk("------------------------------------\n");
3570 for (i = 0; i < 4; i++)
3571 temp_name[i] = readb(&(tb->Signature[i]));
3572 temp_name[4] = '\0';
3573 printk(" Signature = %s\n", temp_name);
3574 printk(" Spec Number = %d\n", readl(&(tb->SpecValence)));
3575 printk(" Transport methods supported = 0x%x\n",
3576 readl(&(tb->TransportSupport)));
3577 printk(" Transport methods active = 0x%x\n",
3578 readl(&(tb->TransportActive)));
3579 printk(" Requested transport Method = 0x%x\n",
3580 readl(&(tb->HostWrite.TransportRequest)));
3581 printk(" Coalesce Interrupt Delay = 0x%x\n",
3582 readl(&(tb->HostWrite.CoalIntDelay)));
3583 printk(" Coalesce Interrupt Count = 0x%x\n",
3584 readl(&(tb->HostWrite.CoalIntCount)));
3585 printk(" Max outstanding commands = 0x%d\n",
3586 readl(&(tb->CmdsOutMax)));
3587 printk(" Bus Types = 0x%x\n", readl(&(tb->BusTypes)));
3588 for (i = 0; i < 16; i++)
3589 temp_name[i] = readb(&(tb->ServerName[i]));
3590 temp_name[16] = '\0';
3591 printk(" Server Name = %s\n", temp_name);
3592 printk(" Heartbeat Counter = 0x%x\n\n\n", readl(&(tb->HeartBeat)));
3594 #endif /* CCISS_DEBUG */
3596 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3598 int i, offset, mem_type, bar_type;
3599 if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3600 return 0;
3601 offset = 0;
3602 for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3603 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3604 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3605 offset += 4;
3606 else {
3607 mem_type = pci_resource_flags(pdev, i) &
3608 PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3609 switch (mem_type) {
3610 case PCI_BASE_ADDRESS_MEM_TYPE_32:
3611 case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3612 offset += 4; /* 32 bit */
3613 break;
3614 case PCI_BASE_ADDRESS_MEM_TYPE_64:
3615 offset += 8;
3616 break;
3617 default: /* reserved in PCI 2.2 */
3618 printk(KERN_WARNING
3619 "Base address is invalid\n");
3620 return -1;
3621 break;
3624 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3625 return i + 1;
3627 return -1;
3630 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
3631 * controllers that are capable. If not, we use IO-APIC mode.
3634 static void __devinit cciss_interrupt_mode(ctlr_info_t *c,
3635 struct pci_dev *pdev, __u32 board_id)
3637 #ifdef CONFIG_PCI_MSI
3638 int err;
3639 struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
3640 {0, 2}, {0, 3}
3643 /* Some boards advertise MSI but don't really support it */
3644 if ((board_id == 0x40700E11) ||
3645 (board_id == 0x40800E11) ||
3646 (board_id == 0x40820E11) || (board_id == 0x40830E11))
3647 goto default_int_mode;
3649 if (pci_find_capability(pdev, PCI_CAP_ID_MSIX)) {
3650 err = pci_enable_msix(pdev, cciss_msix_entries, 4);
3651 if (!err) {
3652 c->intr[0] = cciss_msix_entries[0].vector;
3653 c->intr[1] = cciss_msix_entries[1].vector;
3654 c->intr[2] = cciss_msix_entries[2].vector;
3655 c->intr[3] = cciss_msix_entries[3].vector;
3656 c->msix_vector = 1;
3657 return;
3659 if (err > 0) {
3660 printk(KERN_WARNING "cciss: only %d MSI-X vectors "
3661 "available\n", err);
3662 goto default_int_mode;
3663 } else {
3664 printk(KERN_WARNING "cciss: MSI-X init failed %d\n",
3665 err);
3666 goto default_int_mode;
3669 if (pci_find_capability(pdev, PCI_CAP_ID_MSI)) {
3670 if (!pci_enable_msi(pdev)) {
3671 c->msi_vector = 1;
3672 } else {
3673 printk(KERN_WARNING "cciss: MSI init failed\n");
3676 default_int_mode:
3677 #endif /* CONFIG_PCI_MSI */
3678 /* if we get here we're going to use the default interrupt mode */
3679 c->intr[SIMPLE_MODE_INT] = pdev->irq;
3680 return;
3683 static int __devinit cciss_pci_init(ctlr_info_t *c, struct pci_dev *pdev)
3685 ushort subsystem_vendor_id, subsystem_device_id, command;
3686 __u32 board_id, scratchpad = 0;
3687 __u64 cfg_offset;
3688 __u32 cfg_base_addr;
3689 __u64 cfg_base_addr_index;
3690 int i, prod_index, err;
3692 subsystem_vendor_id = pdev->subsystem_vendor;
3693 subsystem_device_id = pdev->subsystem_device;
3694 board_id = (((__u32) (subsystem_device_id << 16) & 0xffff0000) |
3695 subsystem_vendor_id);
3697 for (i = 0; i < ARRAY_SIZE(products); i++) {
3698 /* Stand aside for hpsa driver on request */
3699 if (cciss_allow_hpsa && products[i].board_id == HPSA_BOUNDARY)
3700 return -ENODEV;
3701 if (board_id == products[i].board_id)
3702 break;
3704 prod_index = i;
3705 if (prod_index == ARRAY_SIZE(products)) {
3706 dev_warn(&pdev->dev,
3707 "unrecognized board ID: 0x%08lx, ignoring.\n",
3708 (unsigned long) board_id);
3709 return -ENODEV;
3712 /* check to see if controller has been disabled */
3713 /* BEFORE trying to enable it */
3714 (void)pci_read_config_word(pdev, PCI_COMMAND, &command);
3715 if (!(command & 0x02)) {
3716 printk(KERN_WARNING
3717 "cciss: controller appears to be disabled\n");
3718 return -ENODEV;
3721 err = pci_enable_device(pdev);
3722 if (err) {
3723 printk(KERN_ERR "cciss: Unable to Enable PCI device\n");
3724 return err;
3727 err = pci_request_regions(pdev, "cciss");
3728 if (err) {
3729 printk(KERN_ERR "cciss: Cannot obtain PCI resources, "
3730 "aborting\n");
3731 return err;
3734 #ifdef CCISS_DEBUG
3735 printk("command = %x\n", command);
3736 printk("irq = %x\n", pdev->irq);
3737 printk("board_id = %x\n", board_id);
3738 #endif /* CCISS_DEBUG */
3740 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
3741 * else we use the IO-APIC interrupt assigned to us by system ROM.
3743 cciss_interrupt_mode(c, pdev, board_id);
3745 /* find the memory BAR */
3746 for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3747 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM)
3748 break;
3750 if (i == DEVICE_COUNT_RESOURCE) {
3751 printk(KERN_WARNING "cciss: No memory BAR found\n");
3752 err = -ENODEV;
3753 goto err_out_free_res;
3756 c->paddr = pci_resource_start(pdev, i); /* addressing mode bits
3757 * already removed
3760 #ifdef CCISS_DEBUG
3761 printk("address 0 = %lx\n", c->paddr);
3762 #endif /* CCISS_DEBUG */
3763 c->vaddr = remap_pci_mem(c->paddr, 0x250);
3765 /* Wait for the board to become ready. (PCI hotplug needs this.)
3766 * We poll for up to 120 secs, once per 100ms. */
3767 for (i = 0; i < 1200; i++) {
3768 scratchpad = readl(c->vaddr + SA5_SCRATCHPAD_OFFSET);
3769 if (scratchpad == CCISS_FIRMWARE_READY)
3770 break;
3771 set_current_state(TASK_INTERRUPTIBLE);
3772 schedule_timeout(msecs_to_jiffies(100)); /* wait 100ms */
3774 if (scratchpad != CCISS_FIRMWARE_READY) {
3775 printk(KERN_WARNING "cciss: Board not ready. Timed out.\n");
3776 err = -ENODEV;
3777 goto err_out_free_res;
3780 /* get the address index number */
3781 cfg_base_addr = readl(c->vaddr + SA5_CTCFG_OFFSET);
3782 cfg_base_addr &= (__u32) 0x0000ffff;
3783 #ifdef CCISS_DEBUG
3784 printk("cfg base address = %x\n", cfg_base_addr);
3785 #endif /* CCISS_DEBUG */
3786 cfg_base_addr_index = find_PCI_BAR_index(pdev, cfg_base_addr);
3787 #ifdef CCISS_DEBUG
3788 printk("cfg base address index = %llx\n",
3789 (unsigned long long)cfg_base_addr_index);
3790 #endif /* CCISS_DEBUG */
3791 if (cfg_base_addr_index == -1) {
3792 printk(KERN_WARNING "cciss: Cannot find cfg_base_addr_index\n");
3793 err = -ENODEV;
3794 goto err_out_free_res;
3797 cfg_offset = readl(c->vaddr + SA5_CTMEM_OFFSET);
3798 #ifdef CCISS_DEBUG
3799 printk("cfg offset = %llx\n", (unsigned long long)cfg_offset);
3800 #endif /* CCISS_DEBUG */
3801 c->cfgtable = remap_pci_mem(pci_resource_start(pdev,
3802 cfg_base_addr_index) +
3803 cfg_offset, sizeof(CfgTable_struct));
3804 c->board_id = board_id;
3806 #ifdef CCISS_DEBUG
3807 print_cfg_table(c->cfgtable);
3808 #endif /* CCISS_DEBUG */
3810 /* Some controllers support Zero Memory Raid (ZMR).
3811 * When configured in ZMR mode the number of supported
3812 * commands drops to 64. So instead of just setting an
3813 * arbitrary value we make the driver a little smarter.
3814 * We read the config table to tell us how many commands
3815 * are supported on the controller then subtract 4 to
3816 * leave a little room for ioctl calls.
3818 c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
3819 c->maxsgentries = readl(&(c->cfgtable->MaxSGElements));
3822 * Limit native command to 32 s/g elements to save dma'able memory.
3823 * Howvever spec says if 0, use 31
3826 c->max_cmd_sgentries = 31;
3827 if (c->maxsgentries > 512) {
3828 c->max_cmd_sgentries = 32;
3829 c->chainsize = c->maxsgentries - c->max_cmd_sgentries + 1;
3830 c->maxsgentries -= 1; /* account for chain pointer */
3831 } else {
3832 c->maxsgentries = 31; /* Default to traditional value */
3833 c->chainsize = 0; /* traditional */
3836 c->product_name = products[prod_index].product_name;
3837 c->access = *(products[prod_index].access);
3838 c->nr_cmds = c->max_commands - 4;
3839 if ((readb(&c->cfgtable->Signature[0]) != 'C') ||
3840 (readb(&c->cfgtable->Signature[1]) != 'I') ||
3841 (readb(&c->cfgtable->Signature[2]) != 'S') ||
3842 (readb(&c->cfgtable->Signature[3]) != 'S')) {
3843 printk("Does not appear to be a valid CISS config table\n");
3844 err = -ENODEV;
3845 goto err_out_free_res;
3847 #ifdef CONFIG_X86
3849 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
3850 __u32 prefetch;
3851 prefetch = readl(&(c->cfgtable->SCSI_Prefetch));
3852 prefetch |= 0x100;
3853 writel(prefetch, &(c->cfgtable->SCSI_Prefetch));
3855 #endif
3857 /* Disabling DMA prefetch and refetch for the P600.
3858 * An ASIC bug may result in accesses to invalid memory addresses.
3859 * We've disabled prefetch for some time now. Testing with XEN
3860 * kernels revealed a bug in the refetch if dom0 resides on a P600.
3862 if(board_id == 0x3225103C) {
3863 __u32 dma_prefetch;
3864 __u32 dma_refetch;
3865 dma_prefetch = readl(c->vaddr + I2O_DMA1_CFG);
3866 dma_prefetch |= 0x8000;
3867 writel(dma_prefetch, c->vaddr + I2O_DMA1_CFG);
3868 pci_read_config_dword(pdev, PCI_COMMAND_PARITY, &dma_refetch);
3869 dma_refetch |= 0x1;
3870 pci_write_config_dword(pdev, PCI_COMMAND_PARITY, dma_refetch);
3873 #ifdef CCISS_DEBUG
3874 printk("Trying to put board into Simple mode\n");
3875 #endif /* CCISS_DEBUG */
3876 c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
3877 /* Update the field, and then ring the doorbell */
3878 writel(CFGTBL_Trans_Simple, &(c->cfgtable->HostWrite.TransportRequest));
3879 writel(CFGTBL_ChangeReq, c->vaddr + SA5_DOORBELL);
3881 /* under certain very rare conditions, this can take awhile.
3882 * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3883 * as we enter this code.) */
3884 for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3885 if (!(readl(c->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3886 break;
3887 /* delay and try again */
3888 set_current_state(TASK_INTERRUPTIBLE);
3889 schedule_timeout(msecs_to_jiffies(1));
3892 #ifdef CCISS_DEBUG
3893 printk(KERN_DEBUG "I counter got to %d %x\n", i,
3894 readl(c->vaddr + SA5_DOORBELL));
3895 #endif /* CCISS_DEBUG */
3896 #ifdef CCISS_DEBUG
3897 print_cfg_table(c->cfgtable);
3898 #endif /* CCISS_DEBUG */
3900 if (!(readl(&(c->cfgtable->TransportActive)) & CFGTBL_Trans_Simple)) {
3901 printk(KERN_WARNING "cciss: unable to get board into"
3902 " simple mode\n");
3903 err = -ENODEV;
3904 goto err_out_free_res;
3906 return 0;
3908 err_out_free_res:
3910 * Deliberately omit pci_disable_device(): it does something nasty to
3911 * Smart Array controllers that pci_enable_device does not undo
3913 pci_release_regions(pdev);
3914 return err;
3917 /* Function to find the first free pointer into our hba[] array
3918 * Returns -1 if no free entries are left.
3920 static int alloc_cciss_hba(void)
3922 int i;
3924 for (i = 0; i < MAX_CTLR; i++) {
3925 if (!hba[i]) {
3926 ctlr_info_t *p;
3928 p = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
3929 if (!p)
3930 goto Enomem;
3931 hba[i] = p;
3932 return i;
3935 printk(KERN_WARNING "cciss: This driver supports a maximum"
3936 " of %d controllers.\n", MAX_CTLR);
3937 return -1;
3938 Enomem:
3939 printk(KERN_ERR "cciss: out of memory.\n");
3940 return -1;
3943 static void free_hba(int n)
3945 ctlr_info_t *h = hba[n];
3946 int i;
3948 hba[n] = NULL;
3949 for (i = 0; i < h->highest_lun + 1; i++)
3950 if (h->gendisk[i] != NULL)
3951 put_disk(h->gendisk[i]);
3952 kfree(h);
3955 /* Send a message CDB to the firmware. */
3956 static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
3958 typedef struct {
3959 CommandListHeader_struct CommandHeader;
3960 RequestBlock_struct Request;
3961 ErrDescriptor_struct ErrorDescriptor;
3962 } Command;
3963 static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
3964 Command *cmd;
3965 dma_addr_t paddr64;
3966 uint32_t paddr32, tag;
3967 void __iomem *vaddr;
3968 int i, err;
3970 vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
3971 if (vaddr == NULL)
3972 return -ENOMEM;
3974 /* The Inbound Post Queue only accepts 32-bit physical addresses for the
3975 CCISS commands, so they must be allocated from the lower 4GiB of
3976 memory. */
3977 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
3978 if (err) {
3979 iounmap(vaddr);
3980 return -ENOMEM;
3983 cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
3984 if (cmd == NULL) {
3985 iounmap(vaddr);
3986 return -ENOMEM;
3989 /* This must fit, because of the 32-bit consistent DMA mask. Also,
3990 although there's no guarantee, we assume that the address is at
3991 least 4-byte aligned (most likely, it's page-aligned). */
3992 paddr32 = paddr64;
3994 cmd->CommandHeader.ReplyQueue = 0;
3995 cmd->CommandHeader.SGList = 0;
3996 cmd->CommandHeader.SGTotal = 0;
3997 cmd->CommandHeader.Tag.lower = paddr32;
3998 cmd->CommandHeader.Tag.upper = 0;
3999 memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
4001 cmd->Request.CDBLen = 16;
4002 cmd->Request.Type.Type = TYPE_MSG;
4003 cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
4004 cmd->Request.Type.Direction = XFER_NONE;
4005 cmd->Request.Timeout = 0; /* Don't time out */
4006 cmd->Request.CDB[0] = opcode;
4007 cmd->Request.CDB[1] = type;
4008 memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
4010 cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
4011 cmd->ErrorDescriptor.Addr.upper = 0;
4012 cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
4014 writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
4016 for (i = 0; i < 10; i++) {
4017 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
4018 if ((tag & ~3) == paddr32)
4019 break;
4020 schedule_timeout_uninterruptible(HZ);
4023 iounmap(vaddr);
4025 /* we leak the DMA buffer here ... no choice since the controller could
4026 still complete the command. */
4027 if (i == 10) {
4028 printk(KERN_ERR "cciss: controller message %02x:%02x timed out\n",
4029 opcode, type);
4030 return -ETIMEDOUT;
4033 pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
4035 if (tag & 2) {
4036 printk(KERN_ERR "cciss: controller message %02x:%02x failed\n",
4037 opcode, type);
4038 return -EIO;
4041 printk(KERN_INFO "cciss: controller message %02x:%02x succeeded\n",
4042 opcode, type);
4043 return 0;
4046 #define cciss_soft_reset_controller(p) cciss_message(p, 1, 0)
4047 #define cciss_noop(p) cciss_message(p, 3, 0)
4049 static __devinit int cciss_reset_msi(struct pci_dev *pdev)
4051 /* the #defines are stolen from drivers/pci/msi.h. */
4052 #define msi_control_reg(base) (base + PCI_MSI_FLAGS)
4053 #define PCI_MSIX_FLAGS_ENABLE (1 << 15)
4055 int pos;
4056 u16 control = 0;
4058 pos = pci_find_capability(pdev, PCI_CAP_ID_MSI);
4059 if (pos) {
4060 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4061 if (control & PCI_MSI_FLAGS_ENABLE) {
4062 printk(KERN_INFO "cciss: resetting MSI\n");
4063 pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSI_FLAGS_ENABLE);
4067 pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
4068 if (pos) {
4069 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4070 if (control & PCI_MSIX_FLAGS_ENABLE) {
4071 printk(KERN_INFO "cciss: resetting MSI-X\n");
4072 pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSIX_FLAGS_ENABLE);
4076 return 0;
4079 /* This does a hard reset of the controller using PCI power management
4080 * states. */
4081 static __devinit int cciss_hard_reset_controller(struct pci_dev *pdev)
4083 u16 pmcsr, saved_config_space[32];
4084 int i, pos;
4086 printk(KERN_INFO "cciss: using PCI PM to reset controller\n");
4088 /* This is very nearly the same thing as
4090 pci_save_state(pci_dev);
4091 pci_set_power_state(pci_dev, PCI_D3hot);
4092 pci_set_power_state(pci_dev, PCI_D0);
4093 pci_restore_state(pci_dev);
4095 but we can't use these nice canned kernel routines on
4096 kexec, because they also check the MSI/MSI-X state in PCI
4097 configuration space and do the wrong thing when it is
4098 set/cleared. Also, the pci_save/restore_state functions
4099 violate the ordering requirements for restoring the
4100 configuration space from the CCISS document (see the
4101 comment below). So we roll our own .... */
4103 for (i = 0; i < 32; i++)
4104 pci_read_config_word(pdev, 2*i, &saved_config_space[i]);
4106 pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4107 if (pos == 0) {
4108 printk(KERN_ERR "cciss_reset_controller: PCI PM not supported\n");
4109 return -ENODEV;
4112 /* Quoting from the Open CISS Specification: "The Power
4113 * Management Control/Status Register (CSR) controls the power
4114 * state of the device. The normal operating state is D0,
4115 * CSR=00h. The software off state is D3, CSR=03h. To reset
4116 * the controller, place the interface device in D3 then to
4117 * D0, this causes a secondary PCI reset which will reset the
4118 * controller." */
4120 /* enter the D3hot power management state */
4121 pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4122 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4123 pmcsr |= PCI_D3hot;
4124 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4126 schedule_timeout_uninterruptible(HZ >> 1);
4128 /* enter the D0 power management state */
4129 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4130 pmcsr |= PCI_D0;
4131 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4133 schedule_timeout_uninterruptible(HZ >> 1);
4135 /* Restore the PCI configuration space. The Open CISS
4136 * Specification says, "Restore the PCI Configuration
4137 * Registers, offsets 00h through 60h. It is important to
4138 * restore the command register, 16-bits at offset 04h,
4139 * last. Do not restore the configuration status register,
4140 * 16-bits at offset 06h." Note that the offset is 2*i. */
4141 for (i = 0; i < 32; i++) {
4142 if (i == 2 || i == 3)
4143 continue;
4144 pci_write_config_word(pdev, 2*i, saved_config_space[i]);
4146 wmb();
4147 pci_write_config_word(pdev, 4, saved_config_space[2]);
4149 return 0;
4153 * This is it. Find all the controllers and register them. I really hate
4154 * stealing all these major device numbers.
4155 * returns the number of block devices registered.
4157 static int __devinit cciss_init_one(struct pci_dev *pdev,
4158 const struct pci_device_id *ent)
4160 int i;
4161 int j = 0;
4162 int k = 0;
4163 int rc;
4164 int dac, return_code;
4165 InquiryData_struct *inq_buff;
4167 if (reset_devices) {
4168 /* Reset the controller with a PCI power-cycle */
4169 if (cciss_hard_reset_controller(pdev) || cciss_reset_msi(pdev))
4170 return -ENODEV;
4172 /* Now try to get the controller to respond to a no-op. Some
4173 devices (notably the HP Smart Array 5i Controller) need
4174 up to 30 seconds to respond. */
4175 for (i=0; i<30; i++) {
4176 if (cciss_noop(pdev) == 0)
4177 break;
4179 schedule_timeout_uninterruptible(HZ);
4181 if (i == 30) {
4182 printk(KERN_ERR "cciss: controller seems dead\n");
4183 return -EBUSY;
4187 i = alloc_cciss_hba();
4188 if (i < 0)
4189 return -1;
4191 hba[i]->busy_initializing = 1;
4192 INIT_HLIST_HEAD(&hba[i]->cmpQ);
4193 INIT_HLIST_HEAD(&hba[i]->reqQ);
4194 mutex_init(&hba[i]->busy_shutting_down);
4196 if (cciss_pci_init(hba[i], pdev) != 0)
4197 goto clean_no_release_regions;
4199 sprintf(hba[i]->devname, "cciss%d", i);
4200 hba[i]->ctlr = i;
4201 hba[i]->pdev = pdev;
4203 init_completion(&hba[i]->scan_wait);
4205 if (cciss_create_hba_sysfs_entry(hba[i]))
4206 goto clean0;
4208 /* configure PCI DMA stuff */
4209 if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
4210 dac = 1;
4211 else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
4212 dac = 0;
4213 else {
4214 printk(KERN_ERR "cciss: no suitable DMA available\n");
4215 goto clean1;
4219 * register with the major number, or get a dynamic major number
4220 * by passing 0 as argument. This is done for greater than
4221 * 8 controller support.
4223 if (i < MAX_CTLR_ORIG)
4224 hba[i]->major = COMPAQ_CISS_MAJOR + i;
4225 rc = register_blkdev(hba[i]->major, hba[i]->devname);
4226 if (rc == -EBUSY || rc == -EINVAL) {
4227 printk(KERN_ERR
4228 "cciss: Unable to get major number %d for %s "
4229 "on hba %d\n", hba[i]->major, hba[i]->devname, i);
4230 goto clean1;
4231 } else {
4232 if (i >= MAX_CTLR_ORIG)
4233 hba[i]->major = rc;
4236 /* make sure the board interrupts are off */
4237 hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_OFF);
4238 if (request_irq(hba[i]->intr[SIMPLE_MODE_INT], do_cciss_intr,
4239 IRQF_DISABLED | IRQF_SHARED, hba[i]->devname, hba[i])) {
4240 printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
4241 hba[i]->intr[SIMPLE_MODE_INT], hba[i]->devname);
4242 goto clean2;
4245 printk(KERN_INFO "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
4246 hba[i]->devname, pdev->device, pci_name(pdev),
4247 hba[i]->intr[SIMPLE_MODE_INT], dac ? "" : " not");
4249 hba[i]->cmd_pool_bits =
4250 kmalloc(DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4251 * sizeof(unsigned long), GFP_KERNEL);
4252 hba[i]->cmd_pool = (CommandList_struct *)
4253 pci_alloc_consistent(hba[i]->pdev,
4254 hba[i]->nr_cmds * sizeof(CommandList_struct),
4255 &(hba[i]->cmd_pool_dhandle));
4256 hba[i]->errinfo_pool = (ErrorInfo_struct *)
4257 pci_alloc_consistent(hba[i]->pdev,
4258 hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4259 &(hba[i]->errinfo_pool_dhandle));
4260 if ((hba[i]->cmd_pool_bits == NULL)
4261 || (hba[i]->cmd_pool == NULL)
4262 || (hba[i]->errinfo_pool == NULL)) {
4263 printk(KERN_ERR "cciss: out of memory");
4264 goto clean4;
4267 /* Need space for temp scatter list */
4268 hba[i]->scatter_list = kmalloc(hba[i]->max_commands *
4269 sizeof(struct scatterlist *),
4270 GFP_KERNEL);
4271 for (k = 0; k < hba[i]->nr_cmds; k++) {
4272 hba[i]->scatter_list[k] = kmalloc(sizeof(struct scatterlist) *
4273 hba[i]->maxsgentries,
4274 GFP_KERNEL);
4275 if (hba[i]->scatter_list[k] == NULL) {
4276 printk(KERN_ERR "cciss%d: could not allocate "
4277 "s/g lists\n", i);
4278 goto clean4;
4281 hba[i]->cmd_sg_list = cciss_allocate_sg_chain_blocks(hba[i],
4282 hba[i]->chainsize, hba[i]->nr_cmds);
4283 if (!hba[i]->cmd_sg_list && hba[i]->chainsize > 0)
4284 goto clean4;
4286 spin_lock_init(&hba[i]->lock);
4288 /* Initialize the pdev driver private data.
4289 have it point to hba[i]. */
4290 pci_set_drvdata(pdev, hba[i]);
4291 /* command and error info recs zeroed out before
4292 they are used */
4293 memset(hba[i]->cmd_pool_bits, 0,
4294 DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4295 * sizeof(unsigned long));
4297 hba[i]->num_luns = 0;
4298 hba[i]->highest_lun = -1;
4299 for (j = 0; j < CISS_MAX_LUN; j++) {
4300 hba[i]->drv[j] = NULL;
4301 hba[i]->gendisk[j] = NULL;
4304 cciss_scsi_setup(i);
4306 /* Turn the interrupts on so we can service requests */
4307 hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_ON);
4309 /* Get the firmware version */
4310 inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
4311 if (inq_buff == NULL) {
4312 printk(KERN_ERR "cciss: out of memory\n");
4313 goto clean4;
4316 return_code = sendcmd_withirq(CISS_INQUIRY, i, inq_buff,
4317 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
4318 if (return_code == IO_OK) {
4319 hba[i]->firm_ver[0] = inq_buff->data_byte[32];
4320 hba[i]->firm_ver[1] = inq_buff->data_byte[33];
4321 hba[i]->firm_ver[2] = inq_buff->data_byte[34];
4322 hba[i]->firm_ver[3] = inq_buff->data_byte[35];
4323 } else { /* send command failed */
4324 printk(KERN_WARNING "cciss: unable to determine firmware"
4325 " version of controller\n");
4327 kfree(inq_buff);
4329 cciss_procinit(i);
4331 hba[i]->cciss_max_sectors = 8192;
4333 rebuild_lun_table(hba[i], 1, 0);
4334 hba[i]->busy_initializing = 0;
4335 return 1;
4337 clean4:
4338 kfree(hba[i]->cmd_pool_bits);
4339 /* Free up sg elements */
4340 for (k = 0; k < hba[i]->nr_cmds; k++)
4341 kfree(hba[i]->scatter_list[k]);
4342 kfree(hba[i]->scatter_list);
4343 cciss_free_sg_chain_blocks(hba[i]->cmd_sg_list, hba[i]->nr_cmds);
4344 if (hba[i]->cmd_pool)
4345 pci_free_consistent(hba[i]->pdev,
4346 hba[i]->nr_cmds * sizeof(CommandList_struct),
4347 hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4348 if (hba[i]->errinfo_pool)
4349 pci_free_consistent(hba[i]->pdev,
4350 hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4351 hba[i]->errinfo_pool,
4352 hba[i]->errinfo_pool_dhandle);
4353 free_irq(hba[i]->intr[SIMPLE_MODE_INT], hba[i]);
4354 clean2:
4355 unregister_blkdev(hba[i]->major, hba[i]->devname);
4356 clean1:
4357 cciss_destroy_hba_sysfs_entry(hba[i]);
4358 clean0:
4359 pci_release_regions(pdev);
4360 clean_no_release_regions:
4361 hba[i]->busy_initializing = 0;
4364 * Deliberately omit pci_disable_device(): it does something nasty to
4365 * Smart Array controllers that pci_enable_device does not undo
4367 pci_set_drvdata(pdev, NULL);
4368 free_hba(i);
4369 return -1;
4372 static void cciss_shutdown(struct pci_dev *pdev)
4374 ctlr_info_t *h;
4375 char *flush_buf;
4376 int return_code;
4378 h = pci_get_drvdata(pdev);
4379 flush_buf = kzalloc(4, GFP_KERNEL);
4380 if (!flush_buf) {
4381 printk(KERN_WARNING
4382 "cciss:%d cache not flushed, out of memory.\n",
4383 h->ctlr);
4384 return;
4386 /* write all data in the battery backed cache to disk */
4387 memset(flush_buf, 0, 4);
4388 return_code = sendcmd_withirq(CCISS_CACHE_FLUSH, h->ctlr, flush_buf,
4389 4, 0, CTLR_LUNID, TYPE_CMD);
4390 kfree(flush_buf);
4391 if (return_code != IO_OK)
4392 printk(KERN_WARNING "cciss%d: Error flushing cache\n",
4393 h->ctlr);
4394 h->access.set_intr_mask(h, CCISS_INTR_OFF);
4395 free_irq(h->intr[2], h);
4398 static void __devexit cciss_remove_one(struct pci_dev *pdev)
4400 ctlr_info_t *tmp_ptr;
4401 int i, j;
4403 if (pci_get_drvdata(pdev) == NULL) {
4404 printk(KERN_ERR "cciss: Unable to remove device \n");
4405 return;
4408 tmp_ptr = pci_get_drvdata(pdev);
4409 i = tmp_ptr->ctlr;
4410 if (hba[i] == NULL) {
4411 printk(KERN_ERR "cciss: device appears to "
4412 "already be removed \n");
4413 return;
4416 mutex_lock(&hba[i]->busy_shutting_down);
4418 remove_from_scan_list(hba[i]);
4419 remove_proc_entry(hba[i]->devname, proc_cciss);
4420 unregister_blkdev(hba[i]->major, hba[i]->devname);
4422 /* remove it from the disk list */
4423 for (j = 0; j < CISS_MAX_LUN; j++) {
4424 struct gendisk *disk = hba[i]->gendisk[j];
4425 if (disk) {
4426 struct request_queue *q = disk->queue;
4428 if (disk->flags & GENHD_FL_UP) {
4429 cciss_destroy_ld_sysfs_entry(hba[i], j, 1);
4430 del_gendisk(disk);
4432 if (q)
4433 blk_cleanup_queue(q);
4437 #ifdef CONFIG_CISS_SCSI_TAPE
4438 cciss_unregister_scsi(i); /* unhook from SCSI subsystem */
4439 #endif
4441 cciss_shutdown(pdev);
4443 #ifdef CONFIG_PCI_MSI
4444 if (hba[i]->msix_vector)
4445 pci_disable_msix(hba[i]->pdev);
4446 else if (hba[i]->msi_vector)
4447 pci_disable_msi(hba[i]->pdev);
4448 #endif /* CONFIG_PCI_MSI */
4450 iounmap(hba[i]->vaddr);
4452 pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(CommandList_struct),
4453 hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4454 pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4455 hba[i]->errinfo_pool, hba[i]->errinfo_pool_dhandle);
4456 kfree(hba[i]->cmd_pool_bits);
4457 /* Free up sg elements */
4458 for (j = 0; j < hba[i]->nr_cmds; j++)
4459 kfree(hba[i]->scatter_list[j]);
4460 kfree(hba[i]->scatter_list);
4461 cciss_free_sg_chain_blocks(hba[i]->cmd_sg_list, hba[i]->nr_cmds);
4463 * Deliberately omit pci_disable_device(): it does something nasty to
4464 * Smart Array controllers that pci_enable_device does not undo
4466 pci_release_regions(pdev);
4467 pci_set_drvdata(pdev, NULL);
4468 cciss_destroy_hba_sysfs_entry(hba[i]);
4469 mutex_unlock(&hba[i]->busy_shutting_down);
4470 free_hba(i);
4473 static struct pci_driver cciss_pci_driver = {
4474 .name = "cciss",
4475 .probe = cciss_init_one,
4476 .remove = __devexit_p(cciss_remove_one),
4477 .id_table = cciss_pci_device_id, /* id_table */
4478 .shutdown = cciss_shutdown,
4482 * This is it. Register the PCI driver information for the cards we control
4483 * the OS will call our registered routines when it finds one of our cards.
4485 static int __init cciss_init(void)
4487 int err;
4490 * The hardware requires that commands are aligned on a 64-bit
4491 * boundary. Given that we use pci_alloc_consistent() to allocate an
4492 * array of them, the size must be a multiple of 8 bytes.
4494 BUILD_BUG_ON(sizeof(CommandList_struct) % COMMANDLIST_ALIGNMENT);
4496 printk(KERN_INFO DRIVER_NAME "\n");
4498 err = bus_register(&cciss_bus_type);
4499 if (err)
4500 return err;
4502 /* Start the scan thread */
4503 cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
4504 if (IS_ERR(cciss_scan_thread)) {
4505 err = PTR_ERR(cciss_scan_thread);
4506 goto err_bus_unregister;
4509 /* Register for our PCI devices */
4510 err = pci_register_driver(&cciss_pci_driver);
4511 if (err)
4512 goto err_thread_stop;
4514 return err;
4516 err_thread_stop:
4517 kthread_stop(cciss_scan_thread);
4518 err_bus_unregister:
4519 bus_unregister(&cciss_bus_type);
4521 return err;
4524 static void __exit cciss_cleanup(void)
4526 int i;
4528 pci_unregister_driver(&cciss_pci_driver);
4529 /* double check that all controller entrys have been removed */
4530 for (i = 0; i < MAX_CTLR; i++) {
4531 if (hba[i] != NULL) {
4532 printk(KERN_WARNING "cciss: had to remove"
4533 " controller %d\n", i);
4534 cciss_remove_one(hba[i]->pdev);
4537 kthread_stop(cciss_scan_thread);
4538 remove_proc_entry("driver/cciss", NULL);
4539 bus_unregister(&cciss_bus_type);
4542 static void fail_all_cmds(unsigned long ctlr)
4544 /* If we get here, the board is apparently dead. */
4545 ctlr_info_t *h = hba[ctlr];
4546 CommandList_struct *c;
4547 unsigned long flags;
4549 printk(KERN_WARNING "cciss%d: controller not responding.\n", h->ctlr);
4550 h->alive = 0; /* the controller apparently died... */
4552 spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
4554 pci_disable_device(h->pdev); /* Make sure it is really dead. */
4556 /* move everything off the request queue onto the completed queue */
4557 while (!hlist_empty(&h->reqQ)) {
4558 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
4559 removeQ(c);
4560 h->Qdepth--;
4561 addQ(&h->cmpQ, c);
4564 /* Now, fail everything on the completed queue with a HW error */
4565 while (!hlist_empty(&h->cmpQ)) {
4566 c = hlist_entry(h->cmpQ.first, CommandList_struct, list);
4567 removeQ(c);
4568 if (c->cmd_type != CMD_MSG_STALE)
4569 c->err_info->CommandStatus = CMD_HARDWARE_ERR;
4570 if (c->cmd_type == CMD_RWREQ) {
4571 complete_command(h, c, 0);
4572 } else if (c->cmd_type == CMD_IOCTL_PEND)
4573 complete(c->waiting);
4574 #ifdef CONFIG_CISS_SCSI_TAPE
4575 else if (c->cmd_type == CMD_SCSI)
4576 complete_scsi_command(c, 0, 0);
4577 #endif
4579 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
4580 return;
4583 module_init(cciss_init);
4584 module_exit(cciss_cleanup);