cciss: Handle special case for sysfs attributes of the first logical drive.
[linux-2.6/btrfs-unstable.git] / drivers / block / cciss.c
blobaa95eeb30206e18b497e87e85e91d04010684ce2
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 #include "cciss_cmd.h"
72 #include "cciss.h"
73 #include <linux/cciss_ioctl.h>
75 /* define the PCI info for the cards we can control */
76 static const struct pci_device_id cciss_pci_device_id[] = {
77 {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISS, 0x0E11, 0x4070},
78 {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4080},
79 {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4082},
80 {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4083},
81 {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x4091},
82 {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409A},
83 {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409B},
84 {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409C},
85 {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409D},
86 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSA, 0x103C, 0x3225},
87 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSC, 0x103C, 0x3223},
88 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSC, 0x103C, 0x3234},
89 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSC, 0x103C, 0x3235},
90 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSD, 0x103C, 0x3211},
91 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSD, 0x103C, 0x3212},
92 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSD, 0x103C, 0x3213},
93 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSD, 0x103C, 0x3214},
94 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSD, 0x103C, 0x3215},
95 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSC, 0x103C, 0x3237},
96 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSC, 0x103C, 0x323D},
97 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSE, 0x103C, 0x3241},
98 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSE, 0x103C, 0x3243},
99 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSE, 0x103C, 0x3245},
100 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSE, 0x103C, 0x3247},
101 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSE, 0x103C, 0x3249},
102 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSE, 0x103C, 0x324A},
103 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSE, 0x103C, 0x324B},
104 {PCI_VENDOR_ID_HP, PCI_ANY_ID, PCI_ANY_ID, PCI_ANY_ID,
105 PCI_CLASS_STORAGE_RAID << 8, 0xffff << 8, 0},
106 {0,}
109 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
111 /* board_id = Subsystem Device ID & Vendor ID
112 * product = Marketing Name for the board
113 * access = Address of the struct of function pointers
115 static struct board_type products[] = {
116 {0x40700E11, "Smart Array 5300", &SA5_access},
117 {0x40800E11, "Smart Array 5i", &SA5B_access},
118 {0x40820E11, "Smart Array 532", &SA5B_access},
119 {0x40830E11, "Smart Array 5312", &SA5B_access},
120 {0x409A0E11, "Smart Array 641", &SA5_access},
121 {0x409B0E11, "Smart Array 642", &SA5_access},
122 {0x409C0E11, "Smart Array 6400", &SA5_access},
123 {0x409D0E11, "Smart Array 6400 EM", &SA5_access},
124 {0x40910E11, "Smart Array 6i", &SA5_access},
125 {0x3225103C, "Smart Array P600", &SA5_access},
126 {0x3223103C, "Smart Array P800", &SA5_access},
127 {0x3234103C, "Smart Array P400", &SA5_access},
128 {0x3235103C, "Smart Array P400i", &SA5_access},
129 {0x3211103C, "Smart Array E200i", &SA5_access},
130 {0x3212103C, "Smart Array E200", &SA5_access},
131 {0x3213103C, "Smart Array E200i", &SA5_access},
132 {0x3214103C, "Smart Array E200i", &SA5_access},
133 {0x3215103C, "Smart Array E200i", &SA5_access},
134 {0x3237103C, "Smart Array E500", &SA5_access},
135 {0x323D103C, "Smart Array P700m", &SA5_access},
136 {0x3241103C, "Smart Array P212", &SA5_access},
137 {0x3243103C, "Smart Array P410", &SA5_access},
138 {0x3245103C, "Smart Array P410i", &SA5_access},
139 {0x3247103C, "Smart Array P411", &SA5_access},
140 {0x3249103C, "Smart Array P812", &SA5_access},
141 {0x324A103C, "Smart Array P712m", &SA5_access},
142 {0x324B103C, "Smart Array P711m", &SA5_access},
143 {0xFFFF103C, "Unknown Smart Array", &SA5_access},
146 /* How long to wait (in milliseconds) for board to go into simple mode */
147 #define MAX_CONFIG_WAIT 30000
148 #define MAX_IOCTL_CONFIG_WAIT 1000
150 /*define how many times we will try a command because of bus resets */
151 #define MAX_CMD_RETRIES 3
153 #define MAX_CTLR 32
155 /* Originally cciss driver only supports 8 major numbers */
156 #define MAX_CTLR_ORIG 8
158 static ctlr_info_t *hba[MAX_CTLR];
160 static struct task_struct *cciss_scan_thread;
161 static DEFINE_MUTEX(scan_mutex);
162 static LIST_HEAD(scan_q);
164 static void do_cciss_request(struct request_queue *q);
165 static irqreturn_t do_cciss_intr(int irq, void *dev_id);
166 static int cciss_open(struct block_device *bdev, fmode_t mode);
167 static int cciss_release(struct gendisk *disk, fmode_t mode);
168 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
169 unsigned int cmd, unsigned long arg);
170 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
172 static int cciss_revalidate(struct gendisk *disk);
173 static int rebuild_lun_table(ctlr_info_t *h, int first_time);
174 static int deregister_disk(ctlr_info_t *h, int drv_index,
175 int clear_all);
177 static void cciss_read_capacity(int ctlr, int logvol, int withirq,
178 sector_t *total_size, unsigned int *block_size);
179 static void cciss_read_capacity_16(int ctlr, int logvol, int withirq,
180 sector_t *total_size, unsigned int *block_size);
181 static void cciss_geometry_inquiry(int ctlr, int logvol,
182 int withirq, sector_t total_size,
183 unsigned int block_size, InquiryData_struct *inq_buff,
184 drive_info_struct *drv);
185 static void __devinit cciss_interrupt_mode(ctlr_info_t *, struct pci_dev *,
186 __u32);
187 static void start_io(ctlr_info_t *h);
188 static int sendcmd(__u8 cmd, int ctlr, void *buff, size_t size,
189 __u8 page_code, unsigned char *scsi3addr, int cmd_type);
190 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
191 __u8 page_code, unsigned char scsi3addr[],
192 int cmd_type);
193 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
194 int attempt_retry);
195 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c);
197 static void fail_all_cmds(unsigned long ctlr);
198 static int add_to_scan_list(struct ctlr_info *h);
199 static int scan_thread(void *data);
200 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c);
201 static void cciss_hba_release(struct device *dev);
202 static void cciss_device_release(struct device *dev);
203 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index);
205 #ifdef CONFIG_PROC_FS
206 static void cciss_procinit(int i);
207 #else
208 static void cciss_procinit(int i)
211 #endif /* CONFIG_PROC_FS */
213 #ifdef CONFIG_COMPAT
214 static int cciss_compat_ioctl(struct block_device *, fmode_t,
215 unsigned, unsigned long);
216 #endif
218 static const struct block_device_operations cciss_fops = {
219 .owner = THIS_MODULE,
220 .open = cciss_open,
221 .release = cciss_release,
222 .locked_ioctl = cciss_ioctl,
223 .getgeo = cciss_getgeo,
224 #ifdef CONFIG_COMPAT
225 .compat_ioctl = cciss_compat_ioctl,
226 #endif
227 .revalidate_disk = cciss_revalidate,
231 * Enqueuing and dequeuing functions for cmdlists.
233 static inline void addQ(struct hlist_head *list, CommandList_struct *c)
235 hlist_add_head(&c->list, list);
238 static inline void removeQ(CommandList_struct *c)
241 * After kexec/dump some commands might still
242 * be in flight, which the firmware will try
243 * to complete. Resetting the firmware doesn't work
244 * with old fw revisions, so we have to mark
245 * them off as 'stale' to prevent the driver from
246 * falling over.
248 if (WARN_ON(hlist_unhashed(&c->list))) {
249 c->cmd_type = CMD_MSG_STALE;
250 return;
253 hlist_del_init(&c->list);
256 #include "cciss_scsi.c" /* For SCSI tape support */
258 #define RAID_UNKNOWN 6
260 #ifdef CONFIG_PROC_FS
263 * Report information about this controller.
265 #define ENG_GIG 1000000000
266 #define ENG_GIG_FACTOR (ENG_GIG/512)
267 #define ENGAGE_SCSI "engage scsi"
268 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
269 "UNKNOWN"
272 static struct proc_dir_entry *proc_cciss;
274 static void cciss_seq_show_header(struct seq_file *seq)
276 ctlr_info_t *h = seq->private;
278 seq_printf(seq, "%s: HP %s Controller\n"
279 "Board ID: 0x%08lx\n"
280 "Firmware Version: %c%c%c%c\n"
281 "IRQ: %d\n"
282 "Logical drives: %d\n"
283 "Current Q depth: %d\n"
284 "Current # commands on controller: %d\n"
285 "Max Q depth since init: %d\n"
286 "Max # commands on controller since init: %d\n"
287 "Max SG entries since init: %d\n",
288 h->devname,
289 h->product_name,
290 (unsigned long)h->board_id,
291 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
292 h->firm_ver[3], (unsigned int)h->intr[SIMPLE_MODE_INT],
293 h->num_luns,
294 h->Qdepth, h->commands_outstanding,
295 h->maxQsinceinit, h->max_outstanding, h->maxSG);
297 #ifdef CONFIG_CISS_SCSI_TAPE
298 cciss_seq_tape_report(seq, h->ctlr);
299 #endif /* CONFIG_CISS_SCSI_TAPE */
302 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
304 ctlr_info_t *h = seq->private;
305 unsigned ctlr = h->ctlr;
306 unsigned long flags;
308 /* prevent displaying bogus info during configuration
309 * or deconfiguration of a logical volume
311 spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
312 if (h->busy_configuring) {
313 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
314 return ERR_PTR(-EBUSY);
316 h->busy_configuring = 1;
317 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
319 if (*pos == 0)
320 cciss_seq_show_header(seq);
322 return pos;
325 static int cciss_seq_show(struct seq_file *seq, void *v)
327 sector_t vol_sz, vol_sz_frac;
328 ctlr_info_t *h = seq->private;
329 unsigned ctlr = h->ctlr;
330 loff_t *pos = v;
331 drive_info_struct *drv = &h->drv[*pos];
333 if (*pos > h->highest_lun)
334 return 0;
336 if (drv->heads == 0)
337 return 0;
339 vol_sz = drv->nr_blocks;
340 vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
341 vol_sz_frac *= 100;
342 sector_div(vol_sz_frac, ENG_GIG_FACTOR);
344 if (drv->raid_level > 5)
345 drv->raid_level = RAID_UNKNOWN;
346 seq_printf(seq, "cciss/c%dd%d:"
347 "\t%4u.%02uGB\tRAID %s\n",
348 ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
349 raid_label[drv->raid_level]);
350 return 0;
353 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
355 ctlr_info_t *h = seq->private;
357 if (*pos > h->highest_lun)
358 return NULL;
359 *pos += 1;
361 return pos;
364 static void cciss_seq_stop(struct seq_file *seq, void *v)
366 ctlr_info_t *h = seq->private;
368 /* Only reset h->busy_configuring if we succeeded in setting
369 * it during cciss_seq_start. */
370 if (v == ERR_PTR(-EBUSY))
371 return;
373 h->busy_configuring = 0;
376 static const struct seq_operations cciss_seq_ops = {
377 .start = cciss_seq_start,
378 .show = cciss_seq_show,
379 .next = cciss_seq_next,
380 .stop = cciss_seq_stop,
383 static int cciss_seq_open(struct inode *inode, struct file *file)
385 int ret = seq_open(file, &cciss_seq_ops);
386 struct seq_file *seq = file->private_data;
388 if (!ret)
389 seq->private = PDE(inode)->data;
391 return ret;
394 static ssize_t
395 cciss_proc_write(struct file *file, const char __user *buf,
396 size_t length, loff_t *ppos)
398 int err;
399 char *buffer;
401 #ifndef CONFIG_CISS_SCSI_TAPE
402 return -EINVAL;
403 #endif
405 if (!buf || length > PAGE_SIZE - 1)
406 return -EINVAL;
408 buffer = (char *)__get_free_page(GFP_KERNEL);
409 if (!buffer)
410 return -ENOMEM;
412 err = -EFAULT;
413 if (copy_from_user(buffer, buf, length))
414 goto out;
415 buffer[length] = '\0';
417 #ifdef CONFIG_CISS_SCSI_TAPE
418 if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
419 struct seq_file *seq = file->private_data;
420 ctlr_info_t *h = seq->private;
421 int rc;
423 rc = cciss_engage_scsi(h->ctlr);
424 if (rc != 0)
425 err = -rc;
426 else
427 err = length;
428 } else
429 #endif /* CONFIG_CISS_SCSI_TAPE */
430 err = -EINVAL;
431 /* might be nice to have "disengage" too, but it's not
432 safely possible. (only 1 module use count, lock issues.) */
434 out:
435 free_page((unsigned long)buffer);
436 return err;
439 static struct file_operations cciss_proc_fops = {
440 .owner = THIS_MODULE,
441 .open = cciss_seq_open,
442 .read = seq_read,
443 .llseek = seq_lseek,
444 .release = seq_release,
445 .write = cciss_proc_write,
448 static void __devinit cciss_procinit(int i)
450 struct proc_dir_entry *pde;
452 if (proc_cciss == NULL)
453 proc_cciss = proc_mkdir("driver/cciss", NULL);
454 if (!proc_cciss)
455 return;
456 pde = proc_create_data(hba[i]->devname, S_IWUSR | S_IRUSR | S_IRGRP |
457 S_IROTH, proc_cciss,
458 &cciss_proc_fops, hba[i]);
460 #endif /* CONFIG_PROC_FS */
462 #define MAX_PRODUCT_NAME_LEN 19
464 #define to_hba(n) container_of(n, struct ctlr_info, dev)
466 static ssize_t host_store_rescan(struct device *dev,
467 struct device_attribute *attr,
468 const char *buf, size_t count)
470 struct ctlr_info *h = to_hba(dev);
472 add_to_scan_list(h);
473 wake_up_process(cciss_scan_thread);
474 wait_for_completion_interruptible(&h->scan_wait);
476 return count;
478 DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
480 static ssize_t dev_show_unique_id(struct device *dev,
481 struct device_attribute *attr,
482 char *buf)
484 drive_info_struct *drv = dev_get_drvdata(dev);
485 struct ctlr_info *h = to_hba(drv->dev->parent);
486 __u8 sn[16];
487 unsigned long flags;
488 int ret = 0;
490 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
491 if (h->busy_configuring)
492 ret = -EBUSY;
493 else
494 memcpy(sn, drv->serial_no, sizeof(sn));
495 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
497 if (ret)
498 return ret;
499 else
500 return snprintf(buf, 16 * 2 + 2,
501 "%02X%02X%02X%02X%02X%02X%02X%02X"
502 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
503 sn[0], sn[1], sn[2], sn[3],
504 sn[4], sn[5], sn[6], sn[7],
505 sn[8], sn[9], sn[10], sn[11],
506 sn[12], sn[13], sn[14], sn[15]);
508 DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
510 static ssize_t dev_show_vendor(struct device *dev,
511 struct device_attribute *attr,
512 char *buf)
514 drive_info_struct *drv = dev_get_drvdata(dev);
515 struct ctlr_info *h = to_hba(drv->dev->parent);
516 char vendor[VENDOR_LEN + 1];
517 unsigned long flags;
518 int ret = 0;
520 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
521 if (h->busy_configuring)
522 ret = -EBUSY;
523 else
524 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
525 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
527 if (ret)
528 return ret;
529 else
530 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
532 DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
534 static ssize_t dev_show_model(struct device *dev,
535 struct device_attribute *attr,
536 char *buf)
538 drive_info_struct *drv = dev_get_drvdata(dev);
539 struct ctlr_info *h = to_hba(drv->dev->parent);
540 char model[MODEL_LEN + 1];
541 unsigned long flags;
542 int ret = 0;
544 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
545 if (h->busy_configuring)
546 ret = -EBUSY;
547 else
548 memcpy(model, drv->model, MODEL_LEN + 1);
549 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
551 if (ret)
552 return ret;
553 else
554 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
556 DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
558 static ssize_t dev_show_rev(struct device *dev,
559 struct device_attribute *attr,
560 char *buf)
562 drive_info_struct *drv = dev_get_drvdata(dev);
563 struct ctlr_info *h = to_hba(drv->dev->parent);
564 char rev[REV_LEN + 1];
565 unsigned long flags;
566 int ret = 0;
568 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
569 if (h->busy_configuring)
570 ret = -EBUSY;
571 else
572 memcpy(rev, drv->rev, REV_LEN + 1);
573 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
575 if (ret)
576 return ret;
577 else
578 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
580 DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
582 static struct attribute *cciss_host_attrs[] = {
583 &dev_attr_rescan.attr,
584 NULL
587 static struct attribute_group cciss_host_attr_group = {
588 .attrs = cciss_host_attrs,
591 static struct attribute_group *cciss_host_attr_groups[] = {
592 &cciss_host_attr_group,
593 NULL
596 static struct device_type cciss_host_type = {
597 .name = "cciss_host",
598 .groups = cciss_host_attr_groups,
599 .release = cciss_hba_release,
602 static struct attribute *cciss_dev_attrs[] = {
603 &dev_attr_unique_id.attr,
604 &dev_attr_model.attr,
605 &dev_attr_vendor.attr,
606 &dev_attr_rev.attr,
607 NULL
610 static struct attribute_group cciss_dev_attr_group = {
611 .attrs = cciss_dev_attrs,
614 static const struct attribute_group *cciss_dev_attr_groups[] = {
615 &cciss_dev_attr_group,
616 NULL
619 static struct device_type cciss_dev_type = {
620 .name = "cciss_device",
621 .groups = cciss_dev_attr_groups,
622 .release = cciss_device_release,
625 static struct bus_type cciss_bus_type = {
626 .name = "cciss",
630 * cciss_hba_release is called when the reference count
631 * of h->dev goes to zero.
633 static void cciss_hba_release(struct device *dev)
636 * nothing to do, but need this to avoid a warning
637 * about not having a release handler from lib/kref.c.
642 * Initialize sysfs entry for each controller. This sets up and registers
643 * the 'cciss#' directory for each individual controller under
644 * /sys/bus/pci/devices/<dev>/.
646 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
648 device_initialize(&h->dev);
649 h->dev.type = &cciss_host_type;
650 h->dev.bus = &cciss_bus_type;
651 dev_set_name(&h->dev, "%s", h->devname);
652 h->dev.parent = &h->pdev->dev;
654 return device_add(&h->dev);
658 * Remove sysfs entries for an hba.
660 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
662 device_del(&h->dev);
663 put_device(&h->dev); /* final put. */
666 /* cciss_device_release is called when the reference count
667 * of h->drv[x].dev goes to zero.
669 static void cciss_device_release(struct device *dev)
671 kfree(dev);
675 * Initialize sysfs for each logical drive. This sets up and registers
676 * the 'c#d#' directory for each individual logical drive under
677 * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
678 * /sys/block/cciss!c#d# to this entry.
680 static long cciss_create_ld_sysfs_entry(struct ctlr_info *h,
681 int drv_index)
683 struct device *dev;
685 /* Special case for c*d0, we only create it once. */
686 if (drv_index == 0 && h->drv[drv_index].dev != NULL)
687 return 0;
689 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
690 if (!dev)
691 return -ENOMEM;
692 device_initialize(dev);
693 dev->type = &cciss_dev_type;
694 dev->bus = &cciss_bus_type;
695 dev_set_name(dev, "c%dd%d", h->ctlr, drv_index);
696 dev->parent = &h->dev;
697 h->drv[drv_index].dev = dev;
698 dev_set_drvdata(dev, &h->drv[drv_index]);
699 return device_add(dev);
703 * Remove sysfs entries for a logical drive.
705 static void cciss_destroy_ld_sysfs_entry(struct ctlr_info *h, int drv_index,
706 int ctlr_exiting)
708 struct device *dev = h->drv[drv_index].dev;
710 /* special case for c*d0, we only destroy it on controller exit */
711 if (drv_index == 0 && !ctlr_exiting)
712 return;
714 device_del(dev);
715 put_device(dev); /* the "final" put. */
716 h->drv[drv_index].dev = NULL;
720 * For operations that cannot sleep, a command block is allocated at init,
721 * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
722 * which ones are free or in use. For operations that can wait for kmalloc
723 * to possible sleep, this routine can be called with get_from_pool set to 0.
724 * cmd_free() MUST be called with a got_from_pool set to 0 if cmd_alloc was.
726 static CommandList_struct *cmd_alloc(ctlr_info_t *h, int get_from_pool)
728 CommandList_struct *c;
729 int i;
730 u64bit temp64;
731 dma_addr_t cmd_dma_handle, err_dma_handle;
733 if (!get_from_pool) {
734 c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
735 sizeof(CommandList_struct), &cmd_dma_handle);
736 if (c == NULL)
737 return NULL;
738 memset(c, 0, sizeof(CommandList_struct));
740 c->cmdindex = -1;
742 c->err_info = (ErrorInfo_struct *)
743 pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
744 &err_dma_handle);
746 if (c->err_info == NULL) {
747 pci_free_consistent(h->pdev,
748 sizeof(CommandList_struct), c, cmd_dma_handle);
749 return NULL;
751 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
752 } else { /* get it out of the controllers pool */
754 do {
755 i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
756 if (i == h->nr_cmds)
757 return NULL;
758 } while (test_and_set_bit
759 (i & (BITS_PER_LONG - 1),
760 h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
761 #ifdef CCISS_DEBUG
762 printk(KERN_DEBUG "cciss: using command buffer %d\n", i);
763 #endif
764 c = h->cmd_pool + i;
765 memset(c, 0, sizeof(CommandList_struct));
766 cmd_dma_handle = h->cmd_pool_dhandle
767 + i * sizeof(CommandList_struct);
768 c->err_info = h->errinfo_pool + i;
769 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
770 err_dma_handle = h->errinfo_pool_dhandle
771 + i * sizeof(ErrorInfo_struct);
772 h->nr_allocs++;
774 c->cmdindex = i;
777 INIT_HLIST_NODE(&c->list);
778 c->busaddr = (__u32) cmd_dma_handle;
779 temp64.val = (__u64) err_dma_handle;
780 c->ErrDesc.Addr.lower = temp64.val32.lower;
781 c->ErrDesc.Addr.upper = temp64.val32.upper;
782 c->ErrDesc.Len = sizeof(ErrorInfo_struct);
784 c->ctlr = h->ctlr;
785 return c;
789 * Frees a command block that was previously allocated with cmd_alloc().
791 static void cmd_free(ctlr_info_t *h, CommandList_struct *c, int got_from_pool)
793 int i;
794 u64bit temp64;
796 if (!got_from_pool) {
797 temp64.val32.lower = c->ErrDesc.Addr.lower;
798 temp64.val32.upper = c->ErrDesc.Addr.upper;
799 pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
800 c->err_info, (dma_addr_t) temp64.val);
801 pci_free_consistent(h->pdev, sizeof(CommandList_struct),
802 c, (dma_addr_t) c->busaddr);
803 } else {
804 i = c - h->cmd_pool;
805 clear_bit(i & (BITS_PER_LONG - 1),
806 h->cmd_pool_bits + (i / BITS_PER_LONG));
807 h->nr_frees++;
811 static inline ctlr_info_t *get_host(struct gendisk *disk)
813 return disk->queue->queuedata;
816 static inline drive_info_struct *get_drv(struct gendisk *disk)
818 return disk->private_data;
822 * Open. Make sure the device is really there.
824 static int cciss_open(struct block_device *bdev, fmode_t mode)
826 ctlr_info_t *host = get_host(bdev->bd_disk);
827 drive_info_struct *drv = get_drv(bdev->bd_disk);
829 #ifdef CCISS_DEBUG
830 printk(KERN_DEBUG "cciss_open %s\n", bdev->bd_disk->disk_name);
831 #endif /* CCISS_DEBUG */
833 if (host->busy_initializing || drv->busy_configuring)
834 return -EBUSY;
836 * Root is allowed to open raw volume zero even if it's not configured
837 * so array config can still work. Root is also allowed to open any
838 * volume that has a LUN ID, so it can issue IOCTL to reread the
839 * disk information. I don't think I really like this
840 * but I'm already using way to many device nodes to claim another one
841 * for "raw controller".
843 if (drv->heads == 0) {
844 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
845 /* if not node 0 make sure it is a partition = 0 */
846 if (MINOR(bdev->bd_dev) & 0x0f) {
847 return -ENXIO;
848 /* if it is, make sure we have a LUN ID */
849 } else if (drv->LunID == 0) {
850 return -ENXIO;
853 if (!capable(CAP_SYS_ADMIN))
854 return -EPERM;
856 drv->usage_count++;
857 host->usage_count++;
858 return 0;
862 * Close. Sync first.
864 static int cciss_release(struct gendisk *disk, fmode_t mode)
866 ctlr_info_t *host = get_host(disk);
867 drive_info_struct *drv = get_drv(disk);
869 #ifdef CCISS_DEBUG
870 printk(KERN_DEBUG "cciss_release %s\n", disk->disk_name);
871 #endif /* CCISS_DEBUG */
873 drv->usage_count--;
874 host->usage_count--;
875 return 0;
878 #ifdef CONFIG_COMPAT
880 static int do_ioctl(struct block_device *bdev, fmode_t mode,
881 unsigned cmd, unsigned long arg)
883 int ret;
884 lock_kernel();
885 ret = cciss_ioctl(bdev, mode, cmd, arg);
886 unlock_kernel();
887 return ret;
890 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
891 unsigned cmd, unsigned long arg);
892 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
893 unsigned cmd, unsigned long arg);
895 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
896 unsigned cmd, unsigned long arg)
898 switch (cmd) {
899 case CCISS_GETPCIINFO:
900 case CCISS_GETINTINFO:
901 case CCISS_SETINTINFO:
902 case CCISS_GETNODENAME:
903 case CCISS_SETNODENAME:
904 case CCISS_GETHEARTBEAT:
905 case CCISS_GETBUSTYPES:
906 case CCISS_GETFIRMVER:
907 case CCISS_GETDRIVVER:
908 case CCISS_REVALIDVOLS:
909 case CCISS_DEREGDISK:
910 case CCISS_REGNEWDISK:
911 case CCISS_REGNEWD:
912 case CCISS_RESCANDISK:
913 case CCISS_GETLUNINFO:
914 return do_ioctl(bdev, mode, cmd, arg);
916 case CCISS_PASSTHRU32:
917 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
918 case CCISS_BIG_PASSTHRU32:
919 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
921 default:
922 return -ENOIOCTLCMD;
926 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
927 unsigned cmd, unsigned long arg)
929 IOCTL32_Command_struct __user *arg32 =
930 (IOCTL32_Command_struct __user *) arg;
931 IOCTL_Command_struct arg64;
932 IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
933 int err;
934 u32 cp;
936 err = 0;
937 err |=
938 copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
939 sizeof(arg64.LUN_info));
940 err |=
941 copy_from_user(&arg64.Request, &arg32->Request,
942 sizeof(arg64.Request));
943 err |=
944 copy_from_user(&arg64.error_info, &arg32->error_info,
945 sizeof(arg64.error_info));
946 err |= get_user(arg64.buf_size, &arg32->buf_size);
947 err |= get_user(cp, &arg32->buf);
948 arg64.buf = compat_ptr(cp);
949 err |= copy_to_user(p, &arg64, sizeof(arg64));
951 if (err)
952 return -EFAULT;
954 err = do_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
955 if (err)
956 return err;
957 err |=
958 copy_in_user(&arg32->error_info, &p->error_info,
959 sizeof(arg32->error_info));
960 if (err)
961 return -EFAULT;
962 return err;
965 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
966 unsigned cmd, unsigned long arg)
968 BIG_IOCTL32_Command_struct __user *arg32 =
969 (BIG_IOCTL32_Command_struct __user *) arg;
970 BIG_IOCTL_Command_struct arg64;
971 BIG_IOCTL_Command_struct __user *p =
972 compat_alloc_user_space(sizeof(arg64));
973 int err;
974 u32 cp;
976 err = 0;
977 err |=
978 copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
979 sizeof(arg64.LUN_info));
980 err |=
981 copy_from_user(&arg64.Request, &arg32->Request,
982 sizeof(arg64.Request));
983 err |=
984 copy_from_user(&arg64.error_info, &arg32->error_info,
985 sizeof(arg64.error_info));
986 err |= get_user(arg64.buf_size, &arg32->buf_size);
987 err |= get_user(arg64.malloc_size, &arg32->malloc_size);
988 err |= get_user(cp, &arg32->buf);
989 arg64.buf = compat_ptr(cp);
990 err |= copy_to_user(p, &arg64, sizeof(arg64));
992 if (err)
993 return -EFAULT;
995 err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
996 if (err)
997 return err;
998 err |=
999 copy_in_user(&arg32->error_info, &p->error_info,
1000 sizeof(arg32->error_info));
1001 if (err)
1002 return -EFAULT;
1003 return err;
1005 #endif
1007 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1009 drive_info_struct *drv = get_drv(bdev->bd_disk);
1011 if (!drv->cylinders)
1012 return -ENXIO;
1014 geo->heads = drv->heads;
1015 geo->sectors = drv->sectors;
1016 geo->cylinders = drv->cylinders;
1017 return 0;
1020 static void check_ioctl_unit_attention(ctlr_info_t *host, CommandList_struct *c)
1022 if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1023 c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1024 (void)check_for_unit_attention(host, c);
1027 * ioctl
1029 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1030 unsigned int cmd, unsigned long arg)
1032 struct gendisk *disk = bdev->bd_disk;
1033 ctlr_info_t *host = get_host(disk);
1034 drive_info_struct *drv = get_drv(disk);
1035 int ctlr = host->ctlr;
1036 void __user *argp = (void __user *)arg;
1038 #ifdef CCISS_DEBUG
1039 printk(KERN_DEBUG "cciss_ioctl: Called with cmd=%x %lx\n", cmd, arg);
1040 #endif /* CCISS_DEBUG */
1042 switch (cmd) {
1043 case CCISS_GETPCIINFO:
1045 cciss_pci_info_struct pciinfo;
1047 if (!arg)
1048 return -EINVAL;
1049 pciinfo.domain = pci_domain_nr(host->pdev->bus);
1050 pciinfo.bus = host->pdev->bus->number;
1051 pciinfo.dev_fn = host->pdev->devfn;
1052 pciinfo.board_id = host->board_id;
1053 if (copy_to_user
1054 (argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1055 return -EFAULT;
1056 return 0;
1058 case CCISS_GETINTINFO:
1060 cciss_coalint_struct intinfo;
1061 if (!arg)
1062 return -EINVAL;
1063 intinfo.delay =
1064 readl(&host->cfgtable->HostWrite.CoalIntDelay);
1065 intinfo.count =
1066 readl(&host->cfgtable->HostWrite.CoalIntCount);
1067 if (copy_to_user
1068 (argp, &intinfo, sizeof(cciss_coalint_struct)))
1069 return -EFAULT;
1070 return 0;
1072 case CCISS_SETINTINFO:
1074 cciss_coalint_struct intinfo;
1075 unsigned long flags;
1076 int i;
1078 if (!arg)
1079 return -EINVAL;
1080 if (!capable(CAP_SYS_ADMIN))
1081 return -EPERM;
1082 if (copy_from_user
1083 (&intinfo, argp, sizeof(cciss_coalint_struct)))
1084 return -EFAULT;
1085 if ((intinfo.delay == 0) && (intinfo.count == 0))
1087 // printk("cciss_ioctl: delay and count cannot be 0\n");
1088 return -EINVAL;
1090 spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1091 /* Update the field, and then ring the doorbell */
1092 writel(intinfo.delay,
1093 &(host->cfgtable->HostWrite.CoalIntDelay));
1094 writel(intinfo.count,
1095 &(host->cfgtable->HostWrite.CoalIntCount));
1096 writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1098 for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1099 if (!(readl(host->vaddr + SA5_DOORBELL)
1100 & CFGTBL_ChangeReq))
1101 break;
1102 /* delay and try again */
1103 udelay(1000);
1105 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1106 if (i >= MAX_IOCTL_CONFIG_WAIT)
1107 return -EAGAIN;
1108 return 0;
1110 case CCISS_GETNODENAME:
1112 NodeName_type NodeName;
1113 int i;
1115 if (!arg)
1116 return -EINVAL;
1117 for (i = 0; i < 16; i++)
1118 NodeName[i] =
1119 readb(&host->cfgtable->ServerName[i]);
1120 if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1121 return -EFAULT;
1122 return 0;
1124 case CCISS_SETNODENAME:
1126 NodeName_type NodeName;
1127 unsigned long flags;
1128 int i;
1130 if (!arg)
1131 return -EINVAL;
1132 if (!capable(CAP_SYS_ADMIN))
1133 return -EPERM;
1135 if (copy_from_user
1136 (NodeName, argp, sizeof(NodeName_type)))
1137 return -EFAULT;
1139 spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1141 /* Update the field, and then ring the doorbell */
1142 for (i = 0; i < 16; i++)
1143 writeb(NodeName[i],
1144 &host->cfgtable->ServerName[i]);
1146 writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1148 for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1149 if (!(readl(host->vaddr + SA5_DOORBELL)
1150 & CFGTBL_ChangeReq))
1151 break;
1152 /* delay and try again */
1153 udelay(1000);
1155 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1156 if (i >= MAX_IOCTL_CONFIG_WAIT)
1157 return -EAGAIN;
1158 return 0;
1161 case CCISS_GETHEARTBEAT:
1163 Heartbeat_type heartbeat;
1165 if (!arg)
1166 return -EINVAL;
1167 heartbeat = readl(&host->cfgtable->HeartBeat);
1168 if (copy_to_user
1169 (argp, &heartbeat, sizeof(Heartbeat_type)))
1170 return -EFAULT;
1171 return 0;
1173 case CCISS_GETBUSTYPES:
1175 BusTypes_type BusTypes;
1177 if (!arg)
1178 return -EINVAL;
1179 BusTypes = readl(&host->cfgtable->BusTypes);
1180 if (copy_to_user
1181 (argp, &BusTypes, sizeof(BusTypes_type)))
1182 return -EFAULT;
1183 return 0;
1185 case CCISS_GETFIRMVER:
1187 FirmwareVer_type firmware;
1189 if (!arg)
1190 return -EINVAL;
1191 memcpy(firmware, host->firm_ver, 4);
1193 if (copy_to_user
1194 (argp, firmware, sizeof(FirmwareVer_type)))
1195 return -EFAULT;
1196 return 0;
1198 case CCISS_GETDRIVVER:
1200 DriverVer_type DriverVer = DRIVER_VERSION;
1202 if (!arg)
1203 return -EINVAL;
1205 if (copy_to_user
1206 (argp, &DriverVer, sizeof(DriverVer_type)))
1207 return -EFAULT;
1208 return 0;
1211 case CCISS_DEREGDISK:
1212 case CCISS_REGNEWD:
1213 case CCISS_REVALIDVOLS:
1214 return rebuild_lun_table(host, 0);
1216 case CCISS_GETLUNINFO:{
1217 LogvolInfo_struct luninfo;
1219 luninfo.LunID = drv->LunID;
1220 luninfo.num_opens = drv->usage_count;
1221 luninfo.num_parts = 0;
1222 if (copy_to_user(argp, &luninfo,
1223 sizeof(LogvolInfo_struct)))
1224 return -EFAULT;
1225 return 0;
1227 case CCISS_PASSTHRU:
1229 IOCTL_Command_struct iocommand;
1230 CommandList_struct *c;
1231 char *buff = NULL;
1232 u64bit temp64;
1233 unsigned long flags;
1234 DECLARE_COMPLETION_ONSTACK(wait);
1236 if (!arg)
1237 return -EINVAL;
1239 if (!capable(CAP_SYS_RAWIO))
1240 return -EPERM;
1242 if (copy_from_user
1243 (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1244 return -EFAULT;
1245 if ((iocommand.buf_size < 1) &&
1246 (iocommand.Request.Type.Direction != XFER_NONE)) {
1247 return -EINVAL;
1249 #if 0 /* 'buf_size' member is 16-bits, and always smaller than kmalloc limit */
1250 /* Check kmalloc limits */
1251 if (iocommand.buf_size > 128000)
1252 return -EINVAL;
1253 #endif
1254 if (iocommand.buf_size > 0) {
1255 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1256 if (buff == NULL)
1257 return -EFAULT;
1259 if (iocommand.Request.Type.Direction == XFER_WRITE) {
1260 /* Copy the data into the buffer we created */
1261 if (copy_from_user
1262 (buff, iocommand.buf, iocommand.buf_size)) {
1263 kfree(buff);
1264 return -EFAULT;
1266 } else {
1267 memset(buff, 0, iocommand.buf_size);
1269 if ((c = cmd_alloc(host, 0)) == NULL) {
1270 kfree(buff);
1271 return -ENOMEM;
1273 // Fill in the command type
1274 c->cmd_type = CMD_IOCTL_PEND;
1275 // Fill in Command Header
1276 c->Header.ReplyQueue = 0; // unused in simple mode
1277 if (iocommand.buf_size > 0) // buffer to fill
1279 c->Header.SGList = 1;
1280 c->Header.SGTotal = 1;
1281 } else // no buffers to fill
1283 c->Header.SGList = 0;
1284 c->Header.SGTotal = 0;
1286 c->Header.LUN = iocommand.LUN_info;
1287 c->Header.Tag.lower = c->busaddr; // use the kernel address the cmd block for tag
1289 // Fill in Request block
1290 c->Request = iocommand.Request;
1292 // Fill in the scatter gather information
1293 if (iocommand.buf_size > 0) {
1294 temp64.val = pci_map_single(host->pdev, buff,
1295 iocommand.buf_size,
1296 PCI_DMA_BIDIRECTIONAL);
1297 c->SG[0].Addr.lower = temp64.val32.lower;
1298 c->SG[0].Addr.upper = temp64.val32.upper;
1299 c->SG[0].Len = iocommand.buf_size;
1300 c->SG[0].Ext = 0; // we are not chaining
1302 c->waiting = &wait;
1304 /* Put the request on the tail of the request queue */
1305 spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1306 addQ(&host->reqQ, c);
1307 host->Qdepth++;
1308 start_io(host);
1309 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1311 wait_for_completion(&wait);
1313 /* unlock the buffers from DMA */
1314 temp64.val32.lower = c->SG[0].Addr.lower;
1315 temp64.val32.upper = c->SG[0].Addr.upper;
1316 pci_unmap_single(host->pdev, (dma_addr_t) temp64.val,
1317 iocommand.buf_size,
1318 PCI_DMA_BIDIRECTIONAL);
1320 check_ioctl_unit_attention(host, c);
1322 /* Copy the error information out */
1323 iocommand.error_info = *(c->err_info);
1324 if (copy_to_user
1325 (argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1326 kfree(buff);
1327 cmd_free(host, c, 0);
1328 return -EFAULT;
1331 if (iocommand.Request.Type.Direction == XFER_READ) {
1332 /* Copy the data out of the buffer we created */
1333 if (copy_to_user
1334 (iocommand.buf, buff, iocommand.buf_size)) {
1335 kfree(buff);
1336 cmd_free(host, c, 0);
1337 return -EFAULT;
1340 kfree(buff);
1341 cmd_free(host, c, 0);
1342 return 0;
1344 case CCISS_BIG_PASSTHRU:{
1345 BIG_IOCTL_Command_struct *ioc;
1346 CommandList_struct *c;
1347 unsigned char **buff = NULL;
1348 int *buff_size = NULL;
1349 u64bit temp64;
1350 unsigned long flags;
1351 BYTE sg_used = 0;
1352 int status = 0;
1353 int i;
1354 DECLARE_COMPLETION_ONSTACK(wait);
1355 __u32 left;
1356 __u32 sz;
1357 BYTE __user *data_ptr;
1359 if (!arg)
1360 return -EINVAL;
1361 if (!capable(CAP_SYS_RAWIO))
1362 return -EPERM;
1363 ioc = (BIG_IOCTL_Command_struct *)
1364 kmalloc(sizeof(*ioc), GFP_KERNEL);
1365 if (!ioc) {
1366 status = -ENOMEM;
1367 goto cleanup1;
1369 if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1370 status = -EFAULT;
1371 goto cleanup1;
1373 if ((ioc->buf_size < 1) &&
1374 (ioc->Request.Type.Direction != XFER_NONE)) {
1375 status = -EINVAL;
1376 goto cleanup1;
1378 /* Check kmalloc limits using all SGs */
1379 if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1380 status = -EINVAL;
1381 goto cleanup1;
1383 if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1384 status = -EINVAL;
1385 goto cleanup1;
1387 buff =
1388 kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1389 if (!buff) {
1390 status = -ENOMEM;
1391 goto cleanup1;
1393 buff_size = kmalloc(MAXSGENTRIES * sizeof(int),
1394 GFP_KERNEL);
1395 if (!buff_size) {
1396 status = -ENOMEM;
1397 goto cleanup1;
1399 left = ioc->buf_size;
1400 data_ptr = ioc->buf;
1401 while (left) {
1402 sz = (left >
1403 ioc->malloc_size) ? ioc->
1404 malloc_size : left;
1405 buff_size[sg_used] = sz;
1406 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1407 if (buff[sg_used] == NULL) {
1408 status = -ENOMEM;
1409 goto cleanup1;
1411 if (ioc->Request.Type.Direction == XFER_WRITE) {
1412 if (copy_from_user
1413 (buff[sg_used], data_ptr, sz)) {
1414 status = -EFAULT;
1415 goto cleanup1;
1417 } else {
1418 memset(buff[sg_used], 0, sz);
1420 left -= sz;
1421 data_ptr += sz;
1422 sg_used++;
1424 if ((c = cmd_alloc(host, 0)) == NULL) {
1425 status = -ENOMEM;
1426 goto cleanup1;
1428 c->cmd_type = CMD_IOCTL_PEND;
1429 c->Header.ReplyQueue = 0;
1431 if (ioc->buf_size > 0) {
1432 c->Header.SGList = sg_used;
1433 c->Header.SGTotal = sg_used;
1434 } else {
1435 c->Header.SGList = 0;
1436 c->Header.SGTotal = 0;
1438 c->Header.LUN = ioc->LUN_info;
1439 c->Header.Tag.lower = c->busaddr;
1441 c->Request = ioc->Request;
1442 if (ioc->buf_size > 0) {
1443 int i;
1444 for (i = 0; i < sg_used; i++) {
1445 temp64.val =
1446 pci_map_single(host->pdev, buff[i],
1447 buff_size[i],
1448 PCI_DMA_BIDIRECTIONAL);
1449 c->SG[i].Addr.lower =
1450 temp64.val32.lower;
1451 c->SG[i].Addr.upper =
1452 temp64.val32.upper;
1453 c->SG[i].Len = buff_size[i];
1454 c->SG[i].Ext = 0; /* we are not chaining */
1457 c->waiting = &wait;
1458 /* Put the request on the tail of the request queue */
1459 spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1460 addQ(&host->reqQ, c);
1461 host->Qdepth++;
1462 start_io(host);
1463 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1464 wait_for_completion(&wait);
1465 /* unlock the buffers from DMA */
1466 for (i = 0; i < sg_used; i++) {
1467 temp64.val32.lower = c->SG[i].Addr.lower;
1468 temp64.val32.upper = c->SG[i].Addr.upper;
1469 pci_unmap_single(host->pdev,
1470 (dma_addr_t) temp64.val, buff_size[i],
1471 PCI_DMA_BIDIRECTIONAL);
1473 check_ioctl_unit_attention(host, c);
1474 /* Copy the error information out */
1475 ioc->error_info = *(c->err_info);
1476 if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1477 cmd_free(host, c, 0);
1478 status = -EFAULT;
1479 goto cleanup1;
1481 if (ioc->Request.Type.Direction == XFER_READ) {
1482 /* Copy the data out of the buffer we created */
1483 BYTE __user *ptr = ioc->buf;
1484 for (i = 0; i < sg_used; i++) {
1485 if (copy_to_user
1486 (ptr, buff[i], buff_size[i])) {
1487 cmd_free(host, c, 0);
1488 status = -EFAULT;
1489 goto cleanup1;
1491 ptr += buff_size[i];
1494 cmd_free(host, c, 0);
1495 status = 0;
1496 cleanup1:
1497 if (buff) {
1498 for (i = 0; i < sg_used; i++)
1499 kfree(buff[i]);
1500 kfree(buff);
1502 kfree(buff_size);
1503 kfree(ioc);
1504 return status;
1507 /* scsi_cmd_ioctl handles these, below, though some are not */
1508 /* very meaningful for cciss. SG_IO is the main one people want. */
1510 case SG_GET_VERSION_NUM:
1511 case SG_SET_TIMEOUT:
1512 case SG_GET_TIMEOUT:
1513 case SG_GET_RESERVED_SIZE:
1514 case SG_SET_RESERVED_SIZE:
1515 case SG_EMULATED_HOST:
1516 case SG_IO:
1517 case SCSI_IOCTL_SEND_COMMAND:
1518 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
1520 /* scsi_cmd_ioctl would normally handle these, below, but */
1521 /* they aren't a good fit for cciss, as CD-ROMs are */
1522 /* not supported, and we don't have any bus/target/lun */
1523 /* which we present to the kernel. */
1525 case CDROM_SEND_PACKET:
1526 case CDROMCLOSETRAY:
1527 case CDROMEJECT:
1528 case SCSI_IOCTL_GET_IDLUN:
1529 case SCSI_IOCTL_GET_BUS_NUMBER:
1530 default:
1531 return -ENOTTY;
1535 static void cciss_check_queues(ctlr_info_t *h)
1537 int start_queue = h->next_to_run;
1538 int i;
1540 /* check to see if we have maxed out the number of commands that can
1541 * be placed on the queue. If so then exit. We do this check here
1542 * in case the interrupt we serviced was from an ioctl and did not
1543 * free any new commands.
1545 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1546 return;
1548 /* We have room on the queue for more commands. Now we need to queue
1549 * them up. We will also keep track of the next queue to run so
1550 * that every queue gets a chance to be started first.
1552 for (i = 0; i < h->highest_lun + 1; i++) {
1553 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1554 /* make sure the disk has been added and the drive is real
1555 * because this can be called from the middle of init_one.
1557 if (!(h->drv[curr_queue].queue) || !(h->drv[curr_queue].heads))
1558 continue;
1559 blk_start_queue(h->gendisk[curr_queue]->queue);
1561 /* check to see if we have maxed out the number of commands
1562 * that can be placed on the queue.
1564 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1565 if (curr_queue == start_queue) {
1566 h->next_to_run =
1567 (start_queue + 1) % (h->highest_lun + 1);
1568 break;
1569 } else {
1570 h->next_to_run = curr_queue;
1571 break;
1577 static void cciss_softirq_done(struct request *rq)
1579 CommandList_struct *cmd = rq->completion_data;
1580 ctlr_info_t *h = hba[cmd->ctlr];
1581 unsigned long flags;
1582 u64bit temp64;
1583 int i, ddir;
1585 if (cmd->Request.Type.Direction == XFER_READ)
1586 ddir = PCI_DMA_FROMDEVICE;
1587 else
1588 ddir = PCI_DMA_TODEVICE;
1590 /* command did not need to be retried */
1591 /* unmap the DMA mapping for all the scatter gather elements */
1592 for (i = 0; i < cmd->Header.SGList; i++) {
1593 temp64.val32.lower = cmd->SG[i].Addr.lower;
1594 temp64.val32.upper = cmd->SG[i].Addr.upper;
1595 pci_unmap_page(h->pdev, temp64.val, cmd->SG[i].Len, ddir);
1598 #ifdef CCISS_DEBUG
1599 printk("Done with %p\n", rq);
1600 #endif /* CCISS_DEBUG */
1602 /* set the residual count for pc requests */
1603 if (blk_pc_request(rq))
1604 rq->resid_len = cmd->err_info->ResidualCnt;
1606 blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1608 spin_lock_irqsave(&h->lock, flags);
1609 cmd_free(h, cmd, 1);
1610 cciss_check_queues(h);
1611 spin_unlock_irqrestore(&h->lock, flags);
1614 static void log_unit_to_scsi3addr(ctlr_info_t *h, unsigned char scsi3addr[],
1615 uint32_t log_unit)
1617 log_unit = h->drv[log_unit].LunID & 0x03fff;
1618 memset(&scsi3addr[4], 0, 4);
1619 memcpy(&scsi3addr[0], &log_unit, 4);
1620 scsi3addr[3] |= 0x40;
1623 /* This function gets the SCSI vendor, model, and revision of a logical drive
1624 * via the inquiry page 0. Model, vendor, and rev are set to empty strings if
1625 * they cannot be read.
1627 static void cciss_get_device_descr(int ctlr, int logvol, int withirq,
1628 char *vendor, char *model, char *rev)
1630 int rc;
1631 InquiryData_struct *inq_buf;
1632 unsigned char scsi3addr[8];
1634 *vendor = '\0';
1635 *model = '\0';
1636 *rev = '\0';
1638 inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1639 if (!inq_buf)
1640 return;
1642 log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1643 if (withirq)
1644 rc = sendcmd_withirq(CISS_INQUIRY, ctlr, inq_buf,
1645 sizeof(InquiryData_struct), 0,
1646 scsi3addr, TYPE_CMD);
1647 else
1648 rc = sendcmd(CISS_INQUIRY, ctlr, inq_buf,
1649 sizeof(InquiryData_struct), 0,
1650 scsi3addr, TYPE_CMD);
1651 if (rc == IO_OK) {
1652 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1653 vendor[VENDOR_LEN] = '\0';
1654 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1655 model[MODEL_LEN] = '\0';
1656 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1657 rev[REV_LEN] = '\0';
1660 kfree(inq_buf);
1661 return;
1664 /* This function gets the serial number of a logical drive via
1665 * inquiry page 0x83. Serial no. is 16 bytes. If the serial
1666 * number cannot be had, for whatever reason, 16 bytes of 0xff
1667 * are returned instead.
1669 static void cciss_get_serial_no(int ctlr, int logvol, int withirq,
1670 unsigned char *serial_no, int buflen)
1672 #define PAGE_83_INQ_BYTES 64
1673 int rc;
1674 unsigned char *buf;
1675 unsigned char scsi3addr[8];
1677 if (buflen > 16)
1678 buflen = 16;
1679 memset(serial_no, 0xff, buflen);
1680 buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1681 if (!buf)
1682 return;
1683 memset(serial_no, 0, buflen);
1684 log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1685 if (withirq)
1686 rc = sendcmd_withirq(CISS_INQUIRY, ctlr, buf,
1687 PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1688 else
1689 rc = sendcmd(CISS_INQUIRY, ctlr, buf,
1690 PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1691 if (rc == IO_OK)
1692 memcpy(serial_no, &buf[8], buflen);
1693 kfree(buf);
1694 return;
1698 * cciss_add_disk sets up the block device queue for a logical drive
1700 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1701 int drv_index)
1703 disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1704 if (!disk->queue)
1705 goto init_queue_failure;
1706 sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1707 disk->major = h->major;
1708 disk->first_minor = drv_index << NWD_SHIFT;
1709 disk->fops = &cciss_fops;
1710 if (h->drv[drv_index].dev == NULL) {
1711 if (cciss_create_ld_sysfs_entry(h, drv_index))
1712 goto cleanup_queue;
1714 disk->private_data = &h->drv[drv_index];
1715 disk->driverfs_dev = h->drv[drv_index].dev;
1717 /* Set up queue information */
1718 blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1720 /* This is a hardware imposed limit. */
1721 blk_queue_max_hw_segments(disk->queue, MAXSGENTRIES);
1723 /* This is a limit in the driver and could be eliminated. */
1724 blk_queue_max_phys_segments(disk->queue, MAXSGENTRIES);
1726 blk_queue_max_sectors(disk->queue, h->cciss_max_sectors);
1728 blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1730 disk->queue->queuedata = h;
1732 blk_queue_logical_block_size(disk->queue,
1733 h->drv[drv_index].block_size);
1735 /* Make sure all queue data is written out before */
1736 /* setting h->drv[drv_index].queue, as setting this */
1737 /* allows the interrupt handler to start the queue */
1738 wmb();
1739 h->drv[drv_index].queue = disk->queue;
1740 add_disk(disk);
1741 return 0;
1743 cleanup_queue:
1744 blk_cleanup_queue(disk->queue);
1745 disk->queue = NULL;
1746 init_queue_failure:
1747 return -1;
1750 /* This function will check the usage_count of the drive to be updated/added.
1751 * If the usage_count is zero and it is a heretofore unknown drive, or,
1752 * the drive's capacity, geometry, or serial number has changed,
1753 * then the drive information will be updated and the disk will be
1754 * re-registered with the kernel. If these conditions don't hold,
1755 * then it will be left alone for the next reboot. The exception to this
1756 * is disk 0 which will always be left registered with the kernel since it
1757 * is also the controller node. Any changes to disk 0 will show up on
1758 * the next reboot.
1760 static void cciss_update_drive_info(int ctlr, int drv_index, int first_time)
1762 ctlr_info_t *h = hba[ctlr];
1763 struct gendisk *disk;
1764 InquiryData_struct *inq_buff = NULL;
1765 unsigned int block_size;
1766 sector_t total_size;
1767 unsigned long flags = 0;
1768 int ret = 0;
1769 drive_info_struct *drvinfo;
1771 /* Get information about the disk and modify the driver structure */
1772 inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1773 drvinfo = kmalloc(sizeof(*drvinfo), GFP_KERNEL);
1774 if (inq_buff == NULL || drvinfo == NULL)
1775 goto mem_msg;
1777 /* testing to see if 16-byte CDBs are already being used */
1778 if (h->cciss_read == CCISS_READ_16) {
1779 cciss_read_capacity_16(h->ctlr, drv_index, 1,
1780 &total_size, &block_size);
1782 } else {
1783 cciss_read_capacity(ctlr, drv_index, 1,
1784 &total_size, &block_size);
1786 /* if read_capacity returns all F's this volume is >2TB */
1787 /* in size so we switch to 16-byte CDB's for all */
1788 /* read/write ops */
1789 if (total_size == 0xFFFFFFFFULL) {
1790 cciss_read_capacity_16(ctlr, drv_index, 1,
1791 &total_size, &block_size);
1792 h->cciss_read = CCISS_READ_16;
1793 h->cciss_write = CCISS_WRITE_16;
1794 } else {
1795 h->cciss_read = CCISS_READ_10;
1796 h->cciss_write = CCISS_WRITE_10;
1800 cciss_geometry_inquiry(ctlr, drv_index, 1, total_size, block_size,
1801 inq_buff, drvinfo);
1802 drvinfo->block_size = block_size;
1803 drvinfo->nr_blocks = total_size + 1;
1805 cciss_get_device_descr(ctlr, drv_index, 1, drvinfo->vendor,
1806 drvinfo->model, drvinfo->rev);
1807 cciss_get_serial_no(ctlr, drv_index, 1, drvinfo->serial_no,
1808 sizeof(drvinfo->serial_no));
1810 /* Is it the same disk we already know, and nothing's changed? */
1811 if (h->drv[drv_index].raid_level != -1 &&
1812 ((memcmp(drvinfo->serial_no,
1813 h->drv[drv_index].serial_no, 16) == 0) &&
1814 drvinfo->block_size == h->drv[drv_index].block_size &&
1815 drvinfo->nr_blocks == h->drv[drv_index].nr_blocks &&
1816 drvinfo->heads == h->drv[drv_index].heads &&
1817 drvinfo->sectors == h->drv[drv_index].sectors &&
1818 drvinfo->cylinders == h->drv[drv_index].cylinders))
1819 /* The disk is unchanged, nothing to update */
1820 goto freeret;
1822 /* If we get here it's not the same disk, or something's changed,
1823 * so we need to * deregister it, and re-register it, if it's not
1824 * in use.
1825 * If the disk already exists then deregister it before proceeding
1826 * (unless it's the first disk (for the controller node).
1828 if (h->drv[drv_index].raid_level != -1 && drv_index != 0) {
1829 printk(KERN_WARNING "disk %d has changed.\n", drv_index);
1830 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1831 h->drv[drv_index].busy_configuring = 1;
1832 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1834 /* deregister_disk sets h->drv[drv_index].queue = NULL
1835 * which keeps the interrupt handler from starting
1836 * the queue.
1838 ret = deregister_disk(h, drv_index, 0);
1839 h->drv[drv_index].busy_configuring = 0;
1842 /* If the disk is in use return */
1843 if (ret)
1844 goto freeret;
1846 /* Save the new information from cciss_geometry_inquiry
1847 * and serial number inquiry.
1849 h->drv[drv_index].block_size = drvinfo->block_size;
1850 h->drv[drv_index].nr_blocks = drvinfo->nr_blocks;
1851 h->drv[drv_index].heads = drvinfo->heads;
1852 h->drv[drv_index].sectors = drvinfo->sectors;
1853 h->drv[drv_index].cylinders = drvinfo->cylinders;
1854 h->drv[drv_index].raid_level = drvinfo->raid_level;
1855 memcpy(h->drv[drv_index].serial_no, drvinfo->serial_no, 16);
1856 memcpy(h->drv[drv_index].vendor, drvinfo->vendor, VENDOR_LEN + 1);
1857 memcpy(h->drv[drv_index].model, drvinfo->model, MODEL_LEN + 1);
1858 memcpy(h->drv[drv_index].rev, drvinfo->rev, REV_LEN + 1);
1860 ++h->num_luns;
1861 disk = h->gendisk[drv_index];
1862 set_capacity(disk, h->drv[drv_index].nr_blocks);
1864 /* If it's not disk 0 (drv_index != 0)
1865 * or if it was disk 0, but there was previously
1866 * no actual corresponding configured logical drive
1867 * (raid_leve == -1) then we want to update the
1868 * logical drive's information.
1870 if (drv_index || first_time) {
1871 if (cciss_add_disk(h, disk, drv_index) != 0) {
1872 cciss_free_gendisk(h, drv_index);
1873 printk(KERN_WARNING "cciss:%d could not update "
1874 "disk %d\n", h->ctlr, drv_index);
1875 --h->num_luns;
1879 freeret:
1880 kfree(inq_buff);
1881 kfree(drvinfo);
1882 return;
1883 mem_msg:
1884 printk(KERN_ERR "cciss: out of memory\n");
1885 goto freeret;
1888 /* This function will find the first index of the controllers drive array
1889 * that has a -1 for the raid_level and will return that index. This is
1890 * where new drives will be added. If the index to be returned is greater
1891 * than the highest_lun index for the controller then highest_lun is set
1892 * to this new index. If there are no available indexes then -1 is returned.
1893 * "controller_node" is used to know if this is a real logical drive, or just
1894 * the controller node, which determines if this counts towards highest_lun.
1896 static int cciss_find_free_drive_index(int ctlr, int controller_node)
1898 int i;
1900 for (i = 0; i < CISS_MAX_LUN; i++) {
1901 if (hba[ctlr]->drv[i].raid_level == -1) {
1902 if (i > hba[ctlr]->highest_lun)
1903 if (!controller_node)
1904 hba[ctlr]->highest_lun = i;
1905 return i;
1908 return -1;
1911 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
1913 put_disk(h->gendisk[drv_index]);
1914 h->gendisk[drv_index] = NULL;
1917 /* cciss_add_gendisk finds a free hba[]->drv structure
1918 * and allocates a gendisk if needed, and sets the lunid
1919 * in the drvinfo structure. It returns the index into
1920 * the ->drv[] array, or -1 if none are free.
1921 * is_controller_node indicates whether highest_lun should
1922 * count this disk, or if it's only being added to provide
1923 * a means to talk to the controller in case no logical
1924 * drives have yet been configured.
1926 static int cciss_add_gendisk(ctlr_info_t *h, __u32 lunid, int controller_node)
1928 int drv_index;
1930 drv_index = cciss_find_free_drive_index(h->ctlr, controller_node);
1931 if (drv_index == -1)
1932 return -1;
1934 /*Check if the gendisk needs to be allocated */
1935 if (!h->gendisk[drv_index]) {
1936 h->gendisk[drv_index] =
1937 alloc_disk(1 << NWD_SHIFT);
1938 if (!h->gendisk[drv_index]) {
1939 printk(KERN_ERR "cciss%d: could not "
1940 "allocate a new disk %d\n",
1941 h->ctlr, drv_index);
1942 return -1;
1945 h->drv[drv_index].LunID = lunid;
1946 if (h->drv[drv_index].dev == NULL) {
1947 if (cciss_create_ld_sysfs_entry(h, drv_index))
1948 goto err_free_disk;
1950 /* Don't need to mark this busy because nobody */
1951 /* else knows about this disk yet to contend */
1952 /* for access to it. */
1953 h->drv[drv_index].busy_configuring = 0;
1954 wmb();
1955 return drv_index;
1957 err_free_disk:
1958 cciss_free_gendisk(h, drv_index);
1959 return -1;
1962 /* This is for the special case of a controller which
1963 * has no logical drives. In this case, we still need
1964 * to register a disk so the controller can be accessed
1965 * by the Array Config Utility.
1967 static void cciss_add_controller_node(ctlr_info_t *h)
1969 struct gendisk *disk;
1970 int drv_index;
1972 if (h->gendisk[0] != NULL) /* already did this? Then bail. */
1973 return;
1975 drv_index = cciss_add_gendisk(h, 0, 1);
1976 if (drv_index == -1)
1977 goto error;
1978 h->drv[drv_index].block_size = 512;
1979 h->drv[drv_index].nr_blocks = 0;
1980 h->drv[drv_index].heads = 0;
1981 h->drv[drv_index].sectors = 0;
1982 h->drv[drv_index].cylinders = 0;
1983 h->drv[drv_index].raid_level = -1;
1984 memset(h->drv[drv_index].serial_no, 0, 16);
1985 disk = h->gendisk[drv_index];
1986 if (cciss_add_disk(h, disk, drv_index) == 0)
1987 return;
1988 cciss_free_gendisk(h, drv_index);
1989 error:
1990 printk(KERN_WARNING "cciss%d: could not "
1991 "add disk 0.\n", h->ctlr);
1992 return;
1995 /* This function will add and remove logical drives from the Logical
1996 * drive array of the controller and maintain persistency of ordering
1997 * so that mount points are preserved until the next reboot. This allows
1998 * for the removal of logical drives in the middle of the drive array
1999 * without a re-ordering of those drives.
2000 * INPUT
2001 * h = The controller to perform the operations on
2003 static int rebuild_lun_table(ctlr_info_t *h, int first_time)
2005 int ctlr = h->ctlr;
2006 int num_luns;
2007 ReportLunData_struct *ld_buff = NULL;
2008 int return_code;
2009 int listlength = 0;
2010 int i;
2011 int drv_found;
2012 int drv_index = 0;
2013 __u32 lunid = 0;
2014 unsigned long flags;
2016 if (!capable(CAP_SYS_RAWIO))
2017 return -EPERM;
2019 /* Set busy_configuring flag for this operation */
2020 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2021 if (h->busy_configuring) {
2022 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2023 return -EBUSY;
2025 h->busy_configuring = 1;
2026 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2028 ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2029 if (ld_buff == NULL)
2030 goto mem_msg;
2032 return_code = sendcmd_withirq(CISS_REPORT_LOG, ctlr, ld_buff,
2033 sizeof(ReportLunData_struct),
2034 0, CTLR_LUNID, TYPE_CMD);
2036 if (return_code == IO_OK)
2037 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2038 else { /* reading number of logical volumes failed */
2039 printk(KERN_WARNING "cciss: report logical volume"
2040 " command failed\n");
2041 listlength = 0;
2042 goto freeret;
2045 num_luns = listlength / 8; /* 8 bytes per entry */
2046 if (num_luns > CISS_MAX_LUN) {
2047 num_luns = CISS_MAX_LUN;
2048 printk(KERN_WARNING "cciss: more luns configured"
2049 " on controller than can be handled by"
2050 " this driver.\n");
2053 if (num_luns == 0)
2054 cciss_add_controller_node(h);
2056 /* Compare controller drive array to driver's drive array
2057 * to see if any drives are missing on the controller due
2058 * to action of Array Config Utility (user deletes drive)
2059 * and deregister logical drives which have disappeared.
2061 for (i = 0; i <= h->highest_lun; i++) {
2062 int j;
2063 drv_found = 0;
2065 /* skip holes in the array from already deleted drives */
2066 if (h->drv[i].raid_level == -1)
2067 continue;
2069 for (j = 0; j < num_luns; j++) {
2070 memcpy(&lunid, &ld_buff->LUN[j][0], 4);
2071 lunid = le32_to_cpu(lunid);
2072 if (h->drv[i].LunID == lunid) {
2073 drv_found = 1;
2074 break;
2077 if (!drv_found) {
2078 /* Deregister it from the OS, it's gone. */
2079 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2080 h->drv[i].busy_configuring = 1;
2081 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2082 return_code = deregister_disk(h, i, 1);
2083 h->drv[i].busy_configuring = 0;
2087 /* Compare controller drive array to driver's drive array.
2088 * Check for updates in the drive information and any new drives
2089 * on the controller due to ACU adding logical drives, or changing
2090 * a logical drive's size, etc. Reregister any new/changed drives
2092 for (i = 0; i < num_luns; i++) {
2093 int j;
2095 drv_found = 0;
2097 memcpy(&lunid, &ld_buff->LUN[i][0], 4);
2098 lunid = le32_to_cpu(lunid);
2100 /* Find if the LUN is already in the drive array
2101 * of the driver. If so then update its info
2102 * if not in use. If it does not exist then find
2103 * the first free index and add it.
2105 for (j = 0; j <= h->highest_lun; j++) {
2106 if (h->drv[j].raid_level != -1 &&
2107 h->drv[j].LunID == lunid) {
2108 drv_index = j;
2109 drv_found = 1;
2110 break;
2114 /* check if the drive was found already in the array */
2115 if (!drv_found) {
2116 drv_index = cciss_add_gendisk(h, lunid, 0);
2117 if (drv_index == -1)
2118 goto freeret;
2120 cciss_update_drive_info(ctlr, drv_index, first_time);
2121 } /* end for */
2123 freeret:
2124 kfree(ld_buff);
2125 h->busy_configuring = 0;
2126 /* We return -1 here to tell the ACU that we have registered/updated
2127 * all of the drives that we can and to keep it from calling us
2128 * additional times.
2130 return -1;
2131 mem_msg:
2132 printk(KERN_ERR "cciss: out of memory\n");
2133 h->busy_configuring = 0;
2134 goto freeret;
2137 /* This function will deregister the disk and it's queue from the
2138 * kernel. It must be called with the controller lock held and the
2139 * drv structures busy_configuring flag set. It's parameters are:
2141 * disk = This is the disk to be deregistered
2142 * drv = This is the drive_info_struct associated with the disk to be
2143 * deregistered. It contains information about the disk used
2144 * by the driver.
2145 * clear_all = This flag determines whether or not the disk information
2146 * is going to be completely cleared out and the highest_lun
2147 * reset. Sometimes we want to clear out information about
2148 * the disk in preparation for re-adding it. In this case
2149 * the highest_lun should be left unchanged and the LunID
2150 * should not be cleared.
2152 static int deregister_disk(ctlr_info_t *h, int drv_index,
2153 int clear_all)
2155 int i;
2156 struct gendisk *disk;
2157 drive_info_struct *drv;
2159 if (!capable(CAP_SYS_RAWIO))
2160 return -EPERM;
2162 drv = &h->drv[drv_index];
2163 disk = h->gendisk[drv_index];
2165 /* make sure logical volume is NOT is use */
2166 if (clear_all || (h->gendisk[0] == disk)) {
2167 if (drv->usage_count > 1)
2168 return -EBUSY;
2169 } else if (drv->usage_count > 0)
2170 return -EBUSY;
2172 /* invalidate the devices and deregister the disk. If it is disk
2173 * zero do not deregister it but just zero out it's values. This
2174 * allows us to delete disk zero but keep the controller registered.
2176 if (h->gendisk[0] != disk) {
2177 struct request_queue *q = disk->queue;
2178 if (disk->flags & GENHD_FL_UP) {
2179 cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2180 del_gendisk(disk);
2182 if (q) {
2183 blk_cleanup_queue(q);
2184 /* Set drv->queue to NULL so that we do not try
2185 * to call blk_start_queue on this queue in the
2186 * interrupt handler
2188 drv->queue = NULL;
2190 /* If clear_all is set then we are deleting the logical
2191 * drive, not just refreshing its info. For drives
2192 * other than disk 0 we will call put_disk. We do not
2193 * do this for disk 0 as we need it to be able to
2194 * configure the controller.
2196 if (clear_all){
2197 /* This isn't pretty, but we need to find the
2198 * disk in our array and NULL our the pointer.
2199 * This is so that we will call alloc_disk if
2200 * this index is used again later.
2202 for (i=0; i < CISS_MAX_LUN; i++){
2203 if (h->gendisk[i] == disk) {
2204 h->gendisk[i] = NULL;
2205 break;
2208 put_disk(disk);
2210 } else {
2211 set_capacity(disk, 0);
2214 --h->num_luns;
2215 /* zero out the disk size info */
2216 drv->nr_blocks = 0;
2217 drv->block_size = 0;
2218 drv->heads = 0;
2219 drv->sectors = 0;
2220 drv->cylinders = 0;
2221 drv->raid_level = -1; /* This can be used as a flag variable to
2222 * indicate that this element of the drive
2223 * array is free.
2225 if (clear_all) {
2226 /* check to see if it was the last disk */
2227 if (drv == h->drv + h->highest_lun) {
2228 /* if so, find the new hightest lun */
2229 int i, newhighest = -1;
2230 for (i = 0; i <= h->highest_lun; i++) {
2231 /* if the disk has size > 0, it is available */
2232 if (h->drv[i].heads)
2233 newhighest = i;
2235 h->highest_lun = newhighest;
2238 drv->LunID = 0;
2240 return 0;
2243 static int fill_cmd(CommandList_struct *c, __u8 cmd, int ctlr, void *buff,
2244 size_t size, __u8 page_code, unsigned char *scsi3addr,
2245 int cmd_type)
2247 ctlr_info_t *h = hba[ctlr];
2248 u64bit buff_dma_handle;
2249 int status = IO_OK;
2251 c->cmd_type = CMD_IOCTL_PEND;
2252 c->Header.ReplyQueue = 0;
2253 if (buff != NULL) {
2254 c->Header.SGList = 1;
2255 c->Header.SGTotal = 1;
2256 } else {
2257 c->Header.SGList = 0;
2258 c->Header.SGTotal = 0;
2260 c->Header.Tag.lower = c->busaddr;
2261 memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2263 c->Request.Type.Type = cmd_type;
2264 if (cmd_type == TYPE_CMD) {
2265 switch (cmd) {
2266 case CISS_INQUIRY:
2267 /* are we trying to read a vital product page */
2268 if (page_code != 0) {
2269 c->Request.CDB[1] = 0x01;
2270 c->Request.CDB[2] = page_code;
2272 c->Request.CDBLen = 6;
2273 c->Request.Type.Attribute = ATTR_SIMPLE;
2274 c->Request.Type.Direction = XFER_READ;
2275 c->Request.Timeout = 0;
2276 c->Request.CDB[0] = CISS_INQUIRY;
2277 c->Request.CDB[4] = size & 0xFF;
2278 break;
2279 case CISS_REPORT_LOG:
2280 case CISS_REPORT_PHYS:
2281 /* Talking to controller so It's a physical command
2282 mode = 00 target = 0. Nothing to write.
2284 c->Request.CDBLen = 12;
2285 c->Request.Type.Attribute = ATTR_SIMPLE;
2286 c->Request.Type.Direction = XFER_READ;
2287 c->Request.Timeout = 0;
2288 c->Request.CDB[0] = cmd;
2289 c->Request.CDB[6] = (size >> 24) & 0xFF; //MSB
2290 c->Request.CDB[7] = (size >> 16) & 0xFF;
2291 c->Request.CDB[8] = (size >> 8) & 0xFF;
2292 c->Request.CDB[9] = size & 0xFF;
2293 break;
2295 case CCISS_READ_CAPACITY:
2296 c->Request.CDBLen = 10;
2297 c->Request.Type.Attribute = ATTR_SIMPLE;
2298 c->Request.Type.Direction = XFER_READ;
2299 c->Request.Timeout = 0;
2300 c->Request.CDB[0] = cmd;
2301 break;
2302 case CCISS_READ_CAPACITY_16:
2303 c->Request.CDBLen = 16;
2304 c->Request.Type.Attribute = ATTR_SIMPLE;
2305 c->Request.Type.Direction = XFER_READ;
2306 c->Request.Timeout = 0;
2307 c->Request.CDB[0] = cmd;
2308 c->Request.CDB[1] = 0x10;
2309 c->Request.CDB[10] = (size >> 24) & 0xFF;
2310 c->Request.CDB[11] = (size >> 16) & 0xFF;
2311 c->Request.CDB[12] = (size >> 8) & 0xFF;
2312 c->Request.CDB[13] = size & 0xFF;
2313 c->Request.Timeout = 0;
2314 c->Request.CDB[0] = cmd;
2315 break;
2316 case CCISS_CACHE_FLUSH:
2317 c->Request.CDBLen = 12;
2318 c->Request.Type.Attribute = ATTR_SIMPLE;
2319 c->Request.Type.Direction = XFER_WRITE;
2320 c->Request.Timeout = 0;
2321 c->Request.CDB[0] = BMIC_WRITE;
2322 c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2323 break;
2324 case TEST_UNIT_READY:
2325 c->Request.CDBLen = 6;
2326 c->Request.Type.Attribute = ATTR_SIMPLE;
2327 c->Request.Type.Direction = XFER_NONE;
2328 c->Request.Timeout = 0;
2329 break;
2330 default:
2331 printk(KERN_WARNING
2332 "cciss%d: Unknown Command 0x%c\n", ctlr, cmd);
2333 return IO_ERROR;
2335 } else if (cmd_type == TYPE_MSG) {
2336 switch (cmd) {
2337 case 0: /* ABORT message */
2338 c->Request.CDBLen = 12;
2339 c->Request.Type.Attribute = ATTR_SIMPLE;
2340 c->Request.Type.Direction = XFER_WRITE;
2341 c->Request.Timeout = 0;
2342 c->Request.CDB[0] = cmd; /* abort */
2343 c->Request.CDB[1] = 0; /* abort a command */
2344 /* buff contains the tag of the command to abort */
2345 memcpy(&c->Request.CDB[4], buff, 8);
2346 break;
2347 case 1: /* RESET message */
2348 c->Request.CDBLen = 16;
2349 c->Request.Type.Attribute = ATTR_SIMPLE;
2350 c->Request.Type.Direction = XFER_NONE;
2351 c->Request.Timeout = 0;
2352 memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2353 c->Request.CDB[0] = cmd; /* reset */
2354 c->Request.CDB[1] = 0x03; /* reset a target */
2355 break;
2356 case 3: /* No-Op message */
2357 c->Request.CDBLen = 1;
2358 c->Request.Type.Attribute = ATTR_SIMPLE;
2359 c->Request.Type.Direction = XFER_WRITE;
2360 c->Request.Timeout = 0;
2361 c->Request.CDB[0] = cmd;
2362 break;
2363 default:
2364 printk(KERN_WARNING
2365 "cciss%d: unknown message type %d\n", ctlr, cmd);
2366 return IO_ERROR;
2368 } else {
2369 printk(KERN_WARNING
2370 "cciss%d: unknown command type %d\n", ctlr, cmd_type);
2371 return IO_ERROR;
2373 /* Fill in the scatter gather information */
2374 if (size > 0) {
2375 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2376 buff, size,
2377 PCI_DMA_BIDIRECTIONAL);
2378 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2379 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2380 c->SG[0].Len = size;
2381 c->SG[0].Ext = 0; /* we are not chaining */
2383 return status;
2386 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2388 switch (c->err_info->ScsiStatus) {
2389 case SAM_STAT_GOOD:
2390 return IO_OK;
2391 case SAM_STAT_CHECK_CONDITION:
2392 switch (0xf & c->err_info->SenseInfo[2]) {
2393 case 0: return IO_OK; /* no sense */
2394 case 1: return IO_OK; /* recovered error */
2395 default:
2396 printk(KERN_WARNING "cciss%d: cmd 0x%02x "
2397 "check condition, sense key = 0x%02x\n",
2398 h->ctlr, c->Request.CDB[0],
2399 c->err_info->SenseInfo[2]);
2401 break;
2402 default:
2403 printk(KERN_WARNING "cciss%d: cmd 0x%02x"
2404 "scsi status = 0x%02x\n", h->ctlr,
2405 c->Request.CDB[0], c->err_info->ScsiStatus);
2406 break;
2408 return IO_ERROR;
2411 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2413 int return_status = IO_OK;
2415 if (c->err_info->CommandStatus == CMD_SUCCESS)
2416 return IO_OK;
2418 switch (c->err_info->CommandStatus) {
2419 case CMD_TARGET_STATUS:
2420 return_status = check_target_status(h, c);
2421 break;
2422 case CMD_DATA_UNDERRUN:
2423 case CMD_DATA_OVERRUN:
2424 /* expected for inquiry and report lun commands */
2425 break;
2426 case CMD_INVALID:
2427 printk(KERN_WARNING "cciss: cmd 0x%02x is "
2428 "reported invalid\n", c->Request.CDB[0]);
2429 return_status = IO_ERROR;
2430 break;
2431 case CMD_PROTOCOL_ERR:
2432 printk(KERN_WARNING "cciss: cmd 0x%02x has "
2433 "protocol error \n", c->Request.CDB[0]);
2434 return_status = IO_ERROR;
2435 break;
2436 case CMD_HARDWARE_ERR:
2437 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2438 " hardware error\n", c->Request.CDB[0]);
2439 return_status = IO_ERROR;
2440 break;
2441 case CMD_CONNECTION_LOST:
2442 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2443 "connection lost\n", c->Request.CDB[0]);
2444 return_status = IO_ERROR;
2445 break;
2446 case CMD_ABORTED:
2447 printk(KERN_WARNING "cciss: cmd 0x%02x was "
2448 "aborted\n", c->Request.CDB[0]);
2449 return_status = IO_ERROR;
2450 break;
2451 case CMD_ABORT_FAILED:
2452 printk(KERN_WARNING "cciss: cmd 0x%02x reports "
2453 "abort failed\n", c->Request.CDB[0]);
2454 return_status = IO_ERROR;
2455 break;
2456 case CMD_UNSOLICITED_ABORT:
2457 printk(KERN_WARNING
2458 "cciss%d: unsolicited abort 0x%02x\n", h->ctlr,
2459 c->Request.CDB[0]);
2460 return_status = IO_NEEDS_RETRY;
2461 break;
2462 default:
2463 printk(KERN_WARNING "cciss: cmd 0x%02x returned "
2464 "unknown status %x\n", c->Request.CDB[0],
2465 c->err_info->CommandStatus);
2466 return_status = IO_ERROR;
2468 return return_status;
2471 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2472 int attempt_retry)
2474 DECLARE_COMPLETION_ONSTACK(wait);
2475 u64bit buff_dma_handle;
2476 unsigned long flags;
2477 int return_status = IO_OK;
2479 resend_cmd2:
2480 c->waiting = &wait;
2481 /* Put the request on the tail of the queue and send it */
2482 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2483 addQ(&h->reqQ, c);
2484 h->Qdepth++;
2485 start_io(h);
2486 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2488 wait_for_completion(&wait);
2490 if (c->err_info->CommandStatus == 0 || !attempt_retry)
2491 goto command_done;
2493 return_status = process_sendcmd_error(h, c);
2495 if (return_status == IO_NEEDS_RETRY &&
2496 c->retry_count < MAX_CMD_RETRIES) {
2497 printk(KERN_WARNING "cciss%d: retrying 0x%02x\n", h->ctlr,
2498 c->Request.CDB[0]);
2499 c->retry_count++;
2500 /* erase the old error information */
2501 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2502 return_status = IO_OK;
2503 INIT_COMPLETION(wait);
2504 goto resend_cmd2;
2507 command_done:
2508 /* unlock the buffers from DMA */
2509 buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2510 buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2511 pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2512 c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2513 return return_status;
2516 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
2517 __u8 page_code, unsigned char scsi3addr[],
2518 int cmd_type)
2520 ctlr_info_t *h = hba[ctlr];
2521 CommandList_struct *c;
2522 int return_status;
2524 c = cmd_alloc(h, 0);
2525 if (!c)
2526 return -ENOMEM;
2527 return_status = fill_cmd(c, cmd, ctlr, buff, size, page_code,
2528 scsi3addr, cmd_type);
2529 if (return_status == IO_OK)
2530 return_status = sendcmd_withirq_core(h, c, 1);
2532 cmd_free(h, c, 0);
2533 return return_status;
2536 static void cciss_geometry_inquiry(int ctlr, int logvol,
2537 int withirq, sector_t total_size,
2538 unsigned int block_size,
2539 InquiryData_struct *inq_buff,
2540 drive_info_struct *drv)
2542 int return_code;
2543 unsigned long t;
2544 unsigned char scsi3addr[8];
2546 memset(inq_buff, 0, sizeof(InquiryData_struct));
2547 log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2548 if (withirq)
2549 return_code = sendcmd_withirq(CISS_INQUIRY, ctlr,
2550 inq_buff, sizeof(*inq_buff),
2551 0xC1, scsi3addr, TYPE_CMD);
2552 else
2553 return_code = sendcmd(CISS_INQUIRY, ctlr, inq_buff,
2554 sizeof(*inq_buff), 0xC1, scsi3addr,
2555 TYPE_CMD);
2556 if (return_code == IO_OK) {
2557 if (inq_buff->data_byte[8] == 0xFF) {
2558 printk(KERN_WARNING
2559 "cciss: reading geometry failed, volume "
2560 "does not support reading geometry\n");
2561 drv->heads = 255;
2562 drv->sectors = 32; // Sectors per track
2563 drv->cylinders = total_size + 1;
2564 drv->raid_level = RAID_UNKNOWN;
2565 } else {
2566 drv->heads = inq_buff->data_byte[6];
2567 drv->sectors = inq_buff->data_byte[7];
2568 drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2569 drv->cylinders += inq_buff->data_byte[5];
2570 drv->raid_level = inq_buff->data_byte[8];
2572 drv->block_size = block_size;
2573 drv->nr_blocks = total_size + 1;
2574 t = drv->heads * drv->sectors;
2575 if (t > 1) {
2576 sector_t real_size = total_size + 1;
2577 unsigned long rem = sector_div(real_size, t);
2578 if (rem)
2579 real_size++;
2580 drv->cylinders = real_size;
2582 } else { /* Get geometry failed */
2583 printk(KERN_WARNING "cciss: reading geometry failed\n");
2585 printk(KERN_INFO " heads=%d, sectors=%d, cylinders=%d\n\n",
2586 drv->heads, drv->sectors, drv->cylinders);
2589 static void
2590 cciss_read_capacity(int ctlr, int logvol, int withirq, sector_t *total_size,
2591 unsigned int *block_size)
2593 ReadCapdata_struct *buf;
2594 int return_code;
2595 unsigned char scsi3addr[8];
2597 buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2598 if (!buf) {
2599 printk(KERN_WARNING "cciss: out of memory\n");
2600 return;
2603 log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2604 if (withirq)
2605 return_code = sendcmd_withirq(CCISS_READ_CAPACITY,
2606 ctlr, buf, sizeof(ReadCapdata_struct),
2607 0, scsi3addr, TYPE_CMD);
2608 else
2609 return_code = sendcmd(CCISS_READ_CAPACITY,
2610 ctlr, buf, sizeof(ReadCapdata_struct),
2611 0, scsi3addr, TYPE_CMD);
2612 if (return_code == IO_OK) {
2613 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2614 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2615 } else { /* read capacity command failed */
2616 printk(KERN_WARNING "cciss: read capacity failed\n");
2617 *total_size = 0;
2618 *block_size = BLOCK_SIZE;
2620 if (*total_size != 0)
2621 printk(KERN_INFO " blocks= %llu block_size= %d\n",
2622 (unsigned long long)*total_size+1, *block_size);
2623 kfree(buf);
2626 static void
2627 cciss_read_capacity_16(int ctlr, int logvol, int withirq, sector_t *total_size, unsigned int *block_size)
2629 ReadCapdata_struct_16 *buf;
2630 int return_code;
2631 unsigned char scsi3addr[8];
2633 buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2634 if (!buf) {
2635 printk(KERN_WARNING "cciss: out of memory\n");
2636 return;
2639 log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2640 if (withirq) {
2641 return_code = sendcmd_withirq(CCISS_READ_CAPACITY_16,
2642 ctlr, buf, sizeof(ReadCapdata_struct_16),
2643 0, scsi3addr, TYPE_CMD);
2645 else {
2646 return_code = sendcmd(CCISS_READ_CAPACITY_16,
2647 ctlr, buf, sizeof(ReadCapdata_struct_16),
2648 0, scsi3addr, TYPE_CMD);
2650 if (return_code == IO_OK) {
2651 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2652 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2653 } else { /* read capacity command failed */
2654 printk(KERN_WARNING "cciss: read capacity failed\n");
2655 *total_size = 0;
2656 *block_size = BLOCK_SIZE;
2658 printk(KERN_INFO " blocks= %llu block_size= %d\n",
2659 (unsigned long long)*total_size+1, *block_size);
2660 kfree(buf);
2663 static int cciss_revalidate(struct gendisk *disk)
2665 ctlr_info_t *h = get_host(disk);
2666 drive_info_struct *drv = get_drv(disk);
2667 int logvol;
2668 int FOUND = 0;
2669 unsigned int block_size;
2670 sector_t total_size;
2671 InquiryData_struct *inq_buff = NULL;
2673 for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
2674 if (h->drv[logvol].LunID == drv->LunID) {
2675 FOUND = 1;
2676 break;
2680 if (!FOUND)
2681 return 1;
2683 inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2684 if (inq_buff == NULL) {
2685 printk(KERN_WARNING "cciss: out of memory\n");
2686 return 1;
2688 if (h->cciss_read == CCISS_READ_10) {
2689 cciss_read_capacity(h->ctlr, logvol, 1,
2690 &total_size, &block_size);
2691 } else {
2692 cciss_read_capacity_16(h->ctlr, logvol, 1,
2693 &total_size, &block_size);
2695 cciss_geometry_inquiry(h->ctlr, logvol, 1, total_size, block_size,
2696 inq_buff, drv);
2698 blk_queue_logical_block_size(drv->queue, drv->block_size);
2699 set_capacity(disk, drv->nr_blocks);
2701 kfree(inq_buff);
2702 return 0;
2706 * Wait polling for a command to complete.
2707 * The memory mapped FIFO is polled for the completion.
2708 * Used only at init time, interrupts from the HBA are disabled.
2710 static unsigned long pollcomplete(int ctlr)
2712 unsigned long done;
2713 int i;
2715 /* Wait (up to 20 seconds) for a command to complete */
2717 for (i = 20 * HZ; i > 0; i--) {
2718 done = hba[ctlr]->access.command_completed(hba[ctlr]);
2719 if (done == FIFO_EMPTY)
2720 schedule_timeout_uninterruptible(1);
2721 else
2722 return done;
2724 /* Invalid address to tell caller we ran out of time */
2725 return 1;
2728 /* Send command c to controller h and poll for it to complete.
2729 * Turns interrupts off on the board. Used at driver init time
2730 * and during SCSI error recovery.
2732 static int sendcmd_core(ctlr_info_t *h, CommandList_struct *c)
2734 int i;
2735 unsigned long complete;
2736 int status = IO_ERROR;
2737 u64bit buff_dma_handle;
2739 resend_cmd1:
2741 /* Disable interrupt on the board. */
2742 h->access.set_intr_mask(h, CCISS_INTR_OFF);
2744 /* Make sure there is room in the command FIFO */
2745 /* Actually it should be completely empty at this time */
2746 /* unless we are in here doing error handling for the scsi */
2747 /* tape side of the driver. */
2748 for (i = 200000; i > 0; i--) {
2749 /* if fifo isn't full go */
2750 if (!(h->access.fifo_full(h)))
2751 break;
2752 udelay(10);
2753 printk(KERN_WARNING "cciss cciss%d: SendCmd FIFO full,"
2754 " waiting!\n", h->ctlr);
2756 h->access.submit_command(h, c); /* Send the cmd */
2757 do {
2758 complete = pollcomplete(h->ctlr);
2760 #ifdef CCISS_DEBUG
2761 printk(KERN_DEBUG "cciss: command completed\n");
2762 #endif /* CCISS_DEBUG */
2764 if (complete == 1) {
2765 printk(KERN_WARNING
2766 "cciss cciss%d: SendCmd Timeout out, "
2767 "No command list address returned!\n", h->ctlr);
2768 status = IO_ERROR;
2769 break;
2772 /* Make sure it's the command we're expecting. */
2773 if ((complete & ~CISS_ERROR_BIT) != c->busaddr) {
2774 printk(KERN_WARNING "cciss%d: Unexpected command "
2775 "completion.\n", h->ctlr);
2776 continue;
2779 /* It is our command. If no error, we're done. */
2780 if (!(complete & CISS_ERROR_BIT)) {
2781 status = IO_OK;
2782 break;
2785 /* There is an error... */
2787 /* if data overrun or underun on Report command ignore it */
2788 if (((c->Request.CDB[0] == CISS_REPORT_LOG) ||
2789 (c->Request.CDB[0] == CISS_REPORT_PHYS) ||
2790 (c->Request.CDB[0] == CISS_INQUIRY)) &&
2791 ((c->err_info->CommandStatus == CMD_DATA_OVERRUN) ||
2792 (c->err_info->CommandStatus == CMD_DATA_UNDERRUN))) {
2793 complete = c->busaddr;
2794 status = IO_OK;
2795 break;
2798 if (c->err_info->CommandStatus == CMD_UNSOLICITED_ABORT) {
2799 printk(KERN_WARNING "cciss%d: unsolicited abort %p\n",
2800 h->ctlr, c);
2801 if (c->retry_count < MAX_CMD_RETRIES) {
2802 printk(KERN_WARNING "cciss%d: retrying %p\n",
2803 h->ctlr, c);
2804 c->retry_count++;
2805 /* erase the old error information */
2806 memset(c->err_info, 0, sizeof(c->err_info));
2807 goto resend_cmd1;
2809 printk(KERN_WARNING "cciss%d: retried %p too many "
2810 "times\n", h->ctlr, c);
2811 status = IO_ERROR;
2812 break;
2815 if (c->err_info->CommandStatus == CMD_UNABORTABLE) {
2816 printk(KERN_WARNING "cciss%d: command could not be "
2817 "aborted.\n", h->ctlr);
2818 status = IO_ERROR;
2819 break;
2822 if (c->err_info->CommandStatus == CMD_TARGET_STATUS) {
2823 status = check_target_status(h, c);
2824 break;
2827 printk(KERN_WARNING "cciss%d: sendcmd error\n", h->ctlr);
2828 printk(KERN_WARNING "cmd = 0x%02x, CommandStatus = 0x%02x\n",
2829 c->Request.CDB[0], c->err_info->CommandStatus);
2830 status = IO_ERROR;
2831 break;
2833 } while (1);
2835 /* unlock the data buffer from DMA */
2836 buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2837 buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2838 pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2839 c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2840 return status;
2844 * Send a command to the controller, and wait for it to complete.
2845 * Used at init time, and during SCSI error recovery.
2847 static int sendcmd(__u8 cmd, int ctlr, void *buff, size_t size,
2848 __u8 page_code, unsigned char *scsi3addr, int cmd_type)
2850 CommandList_struct *c;
2851 int status;
2853 c = cmd_alloc(hba[ctlr], 1);
2854 if (!c) {
2855 printk(KERN_WARNING "cciss: unable to get memory");
2856 return IO_ERROR;
2858 status = fill_cmd(c, cmd, ctlr, buff, size, page_code,
2859 scsi3addr, cmd_type);
2860 if (status == IO_OK)
2861 status = sendcmd_core(hba[ctlr], c);
2862 cmd_free(hba[ctlr], c, 1);
2863 return status;
2867 * Map (physical) PCI mem into (virtual) kernel space
2869 static void __iomem *remap_pci_mem(ulong base, ulong size)
2871 ulong page_base = ((ulong) base) & PAGE_MASK;
2872 ulong page_offs = ((ulong) base) - page_base;
2873 void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2875 return page_remapped ? (page_remapped + page_offs) : NULL;
2879 * Takes jobs of the Q and sends them to the hardware, then puts it on
2880 * the Q to wait for completion.
2882 static void start_io(ctlr_info_t *h)
2884 CommandList_struct *c;
2886 while (!hlist_empty(&h->reqQ)) {
2887 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
2888 /* can't do anything if fifo is full */
2889 if ((h->access.fifo_full(h))) {
2890 printk(KERN_WARNING "cciss: fifo full\n");
2891 break;
2894 /* Get the first entry from the Request Q */
2895 removeQ(c);
2896 h->Qdepth--;
2898 /* Tell the controller execute command */
2899 h->access.submit_command(h, c);
2901 /* Put job onto the completed Q */
2902 addQ(&h->cmpQ, c);
2906 /* Assumes that CCISS_LOCK(h->ctlr) is held. */
2907 /* Zeros out the error record and then resends the command back */
2908 /* to the controller */
2909 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2911 /* erase the old error information */
2912 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2914 /* add it to software queue and then send it to the controller */
2915 addQ(&h->reqQ, c);
2916 h->Qdepth++;
2917 if (h->Qdepth > h->maxQsinceinit)
2918 h->maxQsinceinit = h->Qdepth;
2920 start_io(h);
2923 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
2924 unsigned int msg_byte, unsigned int host_byte,
2925 unsigned int driver_byte)
2927 /* inverse of macros in scsi.h */
2928 return (scsi_status_byte & 0xff) |
2929 ((msg_byte & 0xff) << 8) |
2930 ((host_byte & 0xff) << 16) |
2931 ((driver_byte & 0xff) << 24);
2934 static inline int evaluate_target_status(ctlr_info_t *h,
2935 CommandList_struct *cmd, int *retry_cmd)
2937 unsigned char sense_key;
2938 unsigned char status_byte, msg_byte, host_byte, driver_byte;
2939 int error_value;
2941 *retry_cmd = 0;
2942 /* If we get in here, it means we got "target status", that is, scsi status */
2943 status_byte = cmd->err_info->ScsiStatus;
2944 driver_byte = DRIVER_OK;
2945 msg_byte = cmd->err_info->CommandStatus; /* correct? seems too device specific */
2947 if (blk_pc_request(cmd->rq))
2948 host_byte = DID_PASSTHROUGH;
2949 else
2950 host_byte = DID_OK;
2952 error_value = make_status_bytes(status_byte, msg_byte,
2953 host_byte, driver_byte);
2955 if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
2956 if (!blk_pc_request(cmd->rq))
2957 printk(KERN_WARNING "cciss: cmd %p "
2958 "has SCSI Status 0x%x\n",
2959 cmd, cmd->err_info->ScsiStatus);
2960 return error_value;
2963 /* check the sense key */
2964 sense_key = 0xf & cmd->err_info->SenseInfo[2];
2965 /* no status or recovered error */
2966 if (((sense_key == 0x0) || (sense_key == 0x1)) && !blk_pc_request(cmd->rq))
2967 error_value = 0;
2969 if (check_for_unit_attention(h, cmd)) {
2970 *retry_cmd = !blk_pc_request(cmd->rq);
2971 return 0;
2974 if (!blk_pc_request(cmd->rq)) { /* Not SG_IO or similar? */
2975 if (error_value != 0)
2976 printk(KERN_WARNING "cciss: cmd %p has CHECK CONDITION"
2977 " sense key = 0x%x\n", cmd, sense_key);
2978 return error_value;
2981 /* SG_IO or similar, copy sense data back */
2982 if (cmd->rq->sense) {
2983 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
2984 cmd->rq->sense_len = cmd->err_info->SenseLen;
2985 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
2986 cmd->rq->sense_len);
2987 } else
2988 cmd->rq->sense_len = 0;
2990 return error_value;
2993 /* checks the status of the job and calls complete buffers to mark all
2994 * buffers for the completed job. Note that this function does not need
2995 * to hold the hba/queue lock.
2997 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
2998 int timeout)
3000 int retry_cmd = 0;
3001 struct request *rq = cmd->rq;
3003 rq->errors = 0;
3005 if (timeout)
3006 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
3008 if (cmd->err_info->CommandStatus == 0) /* no error has occurred */
3009 goto after_error_processing;
3011 switch (cmd->err_info->CommandStatus) {
3012 case CMD_TARGET_STATUS:
3013 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
3014 break;
3015 case CMD_DATA_UNDERRUN:
3016 if (blk_fs_request(cmd->rq)) {
3017 printk(KERN_WARNING "cciss: cmd %p has"
3018 " completed with data underrun "
3019 "reported\n", cmd);
3020 cmd->rq->resid_len = cmd->err_info->ResidualCnt;
3022 break;
3023 case CMD_DATA_OVERRUN:
3024 if (blk_fs_request(cmd->rq))
3025 printk(KERN_WARNING "cciss: cmd %p has"
3026 " completed with data overrun "
3027 "reported\n", cmd);
3028 break;
3029 case CMD_INVALID:
3030 printk(KERN_WARNING "cciss: cmd %p is "
3031 "reported invalid\n", cmd);
3032 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3033 cmd->err_info->CommandStatus, DRIVER_OK,
3034 blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3035 break;
3036 case CMD_PROTOCOL_ERR:
3037 printk(KERN_WARNING "cciss: cmd %p has "
3038 "protocol error \n", cmd);
3039 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3040 cmd->err_info->CommandStatus, DRIVER_OK,
3041 blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3042 break;
3043 case CMD_HARDWARE_ERR:
3044 printk(KERN_WARNING "cciss: cmd %p had "
3045 " hardware error\n", cmd);
3046 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3047 cmd->err_info->CommandStatus, DRIVER_OK,
3048 blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3049 break;
3050 case CMD_CONNECTION_LOST:
3051 printk(KERN_WARNING "cciss: cmd %p had "
3052 "connection lost\n", cmd);
3053 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3054 cmd->err_info->CommandStatus, DRIVER_OK,
3055 blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3056 break;
3057 case CMD_ABORTED:
3058 printk(KERN_WARNING "cciss: cmd %p was "
3059 "aborted\n", cmd);
3060 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3061 cmd->err_info->CommandStatus, DRIVER_OK,
3062 blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
3063 break;
3064 case CMD_ABORT_FAILED:
3065 printk(KERN_WARNING "cciss: cmd %p reports "
3066 "abort failed\n", cmd);
3067 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3068 cmd->err_info->CommandStatus, DRIVER_OK,
3069 blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3070 break;
3071 case CMD_UNSOLICITED_ABORT:
3072 printk(KERN_WARNING "cciss%d: unsolicited "
3073 "abort %p\n", h->ctlr, cmd);
3074 if (cmd->retry_count < MAX_CMD_RETRIES) {
3075 retry_cmd = 1;
3076 printk(KERN_WARNING
3077 "cciss%d: retrying %p\n", h->ctlr, cmd);
3078 cmd->retry_count++;
3079 } else
3080 printk(KERN_WARNING
3081 "cciss%d: %p retried too "
3082 "many times\n", h->ctlr, cmd);
3083 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3084 cmd->err_info->CommandStatus, DRIVER_OK,
3085 blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
3086 break;
3087 case CMD_TIMEOUT:
3088 printk(KERN_WARNING "cciss: cmd %p timedout\n", cmd);
3089 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3090 cmd->err_info->CommandStatus, DRIVER_OK,
3091 blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3092 break;
3093 default:
3094 printk(KERN_WARNING "cciss: cmd %p returned "
3095 "unknown status %x\n", cmd,
3096 cmd->err_info->CommandStatus);
3097 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3098 cmd->err_info->CommandStatus, DRIVER_OK,
3099 blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3102 after_error_processing:
3104 /* We need to return this command */
3105 if (retry_cmd) {
3106 resend_cciss_cmd(h, cmd);
3107 return;
3109 cmd->rq->completion_data = cmd;
3110 blk_complete_request(cmd->rq);
3114 * Get a request and submit it to the controller.
3116 static void do_cciss_request(struct request_queue *q)
3118 ctlr_info_t *h = q->queuedata;
3119 CommandList_struct *c;
3120 sector_t start_blk;
3121 int seg;
3122 struct request *creq;
3123 u64bit temp64;
3124 struct scatterlist tmp_sg[MAXSGENTRIES];
3125 drive_info_struct *drv;
3126 int i, dir;
3128 /* We call start_io here in case there is a command waiting on the
3129 * queue that has not been sent.
3131 if (blk_queue_plugged(q))
3132 goto startio;
3134 queue:
3135 creq = blk_peek_request(q);
3136 if (!creq)
3137 goto startio;
3139 BUG_ON(creq->nr_phys_segments > MAXSGENTRIES);
3141 if ((c = cmd_alloc(h, 1)) == NULL)
3142 goto full;
3144 blk_start_request(creq);
3146 spin_unlock_irq(q->queue_lock);
3148 c->cmd_type = CMD_RWREQ;
3149 c->rq = creq;
3151 /* fill in the request */
3152 drv = creq->rq_disk->private_data;
3153 c->Header.ReplyQueue = 0; // unused in simple mode
3154 /* got command from pool, so use the command block index instead */
3155 /* for direct lookups. */
3156 /* The first 2 bits are reserved for controller error reporting. */
3157 c->Header.Tag.lower = (c->cmdindex << 3);
3158 c->Header.Tag.lower |= 0x04; /* flag for direct lookup. */
3159 c->Header.LUN.LogDev.VolId = drv->LunID;
3160 c->Header.LUN.LogDev.Mode = 1;
3161 c->Request.CDBLen = 10; // 12 byte commands not in FW yet;
3162 c->Request.Type.Type = TYPE_CMD; // It is a command.
3163 c->Request.Type.Attribute = ATTR_SIMPLE;
3164 c->Request.Type.Direction =
3165 (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3166 c->Request.Timeout = 0; // Don't time out
3167 c->Request.CDB[0] =
3168 (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3169 start_blk = blk_rq_pos(creq);
3170 #ifdef CCISS_DEBUG
3171 printk(KERN_DEBUG "ciss: sector =%d nr_sectors=%d\n",
3172 (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3173 #endif /* CCISS_DEBUG */
3175 sg_init_table(tmp_sg, MAXSGENTRIES);
3176 seg = blk_rq_map_sg(q, creq, tmp_sg);
3178 /* get the DMA records for the setup */
3179 if (c->Request.Type.Direction == XFER_READ)
3180 dir = PCI_DMA_FROMDEVICE;
3181 else
3182 dir = PCI_DMA_TODEVICE;
3184 for (i = 0; i < seg; i++) {
3185 c->SG[i].Len = tmp_sg[i].length;
3186 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3187 tmp_sg[i].offset,
3188 tmp_sg[i].length, dir);
3189 c->SG[i].Addr.lower = temp64.val32.lower;
3190 c->SG[i].Addr.upper = temp64.val32.upper;
3191 c->SG[i].Ext = 0; // we are not chaining
3193 /* track how many SG entries we are using */
3194 if (seg > h->maxSG)
3195 h->maxSG = seg;
3197 #ifdef CCISS_DEBUG
3198 printk(KERN_DEBUG "cciss: Submitting %u sectors in %d segments\n",
3199 blk_rq_sectors(creq), seg);
3200 #endif /* CCISS_DEBUG */
3202 c->Header.SGList = c->Header.SGTotal = seg;
3203 if (likely(blk_fs_request(creq))) {
3204 if(h->cciss_read == CCISS_READ_10) {
3205 c->Request.CDB[1] = 0;
3206 c->Request.CDB[2] = (start_blk >> 24) & 0xff; //MSB
3207 c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3208 c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3209 c->Request.CDB[5] = start_blk & 0xff;
3210 c->Request.CDB[6] = 0; // (sect >> 24) & 0xff; MSB
3211 c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3212 c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3213 c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3214 } else {
3215 u32 upper32 = upper_32_bits(start_blk);
3217 c->Request.CDBLen = 16;
3218 c->Request.CDB[1]= 0;
3219 c->Request.CDB[2]= (upper32 >> 24) & 0xff; //MSB
3220 c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3221 c->Request.CDB[4]= (upper32 >> 8) & 0xff;
3222 c->Request.CDB[5]= upper32 & 0xff;
3223 c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3224 c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3225 c->Request.CDB[8]= (start_blk >> 8) & 0xff;
3226 c->Request.CDB[9]= start_blk & 0xff;
3227 c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3228 c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3229 c->Request.CDB[12]= (blk_rq_sectors(creq) >> 8) & 0xff;
3230 c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3231 c->Request.CDB[14] = c->Request.CDB[15] = 0;
3233 } else if (blk_pc_request(creq)) {
3234 c->Request.CDBLen = creq->cmd_len;
3235 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3236 } else {
3237 printk(KERN_WARNING "cciss%d: bad request type %d\n", h->ctlr, creq->cmd_type);
3238 BUG();
3241 spin_lock_irq(q->queue_lock);
3243 addQ(&h->reqQ, c);
3244 h->Qdepth++;
3245 if (h->Qdepth > h->maxQsinceinit)
3246 h->maxQsinceinit = h->Qdepth;
3248 goto queue;
3249 full:
3250 blk_stop_queue(q);
3251 startio:
3252 /* We will already have the driver lock here so not need
3253 * to lock it.
3255 start_io(h);
3258 static inline unsigned long get_next_completion(ctlr_info_t *h)
3260 return h->access.command_completed(h);
3263 static inline int interrupt_pending(ctlr_info_t *h)
3265 return h->access.intr_pending(h);
3268 static inline long interrupt_not_for_us(ctlr_info_t *h)
3270 return (((h->access.intr_pending(h) == 0) ||
3271 (h->interrupts_enabled == 0)));
3274 static irqreturn_t do_cciss_intr(int irq, void *dev_id)
3276 ctlr_info_t *h = dev_id;
3277 CommandList_struct *c;
3278 unsigned long flags;
3279 __u32 a, a1, a2;
3281 if (interrupt_not_for_us(h))
3282 return IRQ_NONE;
3284 * If there are completed commands in the completion queue,
3285 * we had better do something about it.
3287 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
3288 while (interrupt_pending(h)) {
3289 while ((a = get_next_completion(h)) != FIFO_EMPTY) {
3290 a1 = a;
3291 if ((a & 0x04)) {
3292 a2 = (a >> 3);
3293 if (a2 >= h->nr_cmds) {
3294 printk(KERN_WARNING
3295 "cciss: controller cciss%d failed, stopping.\n",
3296 h->ctlr);
3297 fail_all_cmds(h->ctlr);
3298 return IRQ_HANDLED;
3301 c = h->cmd_pool + a2;
3302 a = c->busaddr;
3304 } else {
3305 struct hlist_node *tmp;
3307 a &= ~3;
3308 c = NULL;
3309 hlist_for_each_entry(c, tmp, &h->cmpQ, list) {
3310 if (c->busaddr == a)
3311 break;
3315 * If we've found the command, take it off the
3316 * completion Q and free it
3318 if (c && c->busaddr == a) {
3319 removeQ(c);
3320 if (c->cmd_type == CMD_RWREQ) {
3321 complete_command(h, c, 0);
3322 } else if (c->cmd_type == CMD_IOCTL_PEND) {
3323 complete(c->waiting);
3325 # ifdef CONFIG_CISS_SCSI_TAPE
3326 else if (c->cmd_type == CMD_SCSI)
3327 complete_scsi_command(c, 0, a1);
3328 # endif
3329 continue;
3334 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
3335 return IRQ_HANDLED;
3339 * add_to_scan_list() - add controller to rescan queue
3340 * @h: Pointer to the controller.
3342 * Adds the controller to the rescan queue if not already on the queue.
3344 * returns 1 if added to the queue, 0 if skipped (could be on the
3345 * queue already, or the controller could be initializing or shutting
3346 * down).
3348 static int add_to_scan_list(struct ctlr_info *h)
3350 struct ctlr_info *test_h;
3351 int found = 0;
3352 int ret = 0;
3354 if (h->busy_initializing)
3355 return 0;
3357 if (!mutex_trylock(&h->busy_shutting_down))
3358 return 0;
3360 mutex_lock(&scan_mutex);
3361 list_for_each_entry(test_h, &scan_q, scan_list) {
3362 if (test_h == h) {
3363 found = 1;
3364 break;
3367 if (!found && !h->busy_scanning) {
3368 INIT_COMPLETION(h->scan_wait);
3369 list_add_tail(&h->scan_list, &scan_q);
3370 ret = 1;
3372 mutex_unlock(&scan_mutex);
3373 mutex_unlock(&h->busy_shutting_down);
3375 return ret;
3379 * remove_from_scan_list() - remove controller from rescan queue
3380 * @h: Pointer to the controller.
3382 * Removes the controller from the rescan queue if present. Blocks if
3383 * the controller is currently conducting a rescan.
3385 static void remove_from_scan_list(struct ctlr_info *h)
3387 struct ctlr_info *test_h, *tmp_h;
3388 int scanning = 0;
3390 mutex_lock(&scan_mutex);
3391 list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3392 if (test_h == h) {
3393 list_del(&h->scan_list);
3394 complete_all(&h->scan_wait);
3395 mutex_unlock(&scan_mutex);
3396 return;
3399 if (&h->busy_scanning)
3400 scanning = 0;
3401 mutex_unlock(&scan_mutex);
3403 if (scanning)
3404 wait_for_completion(&h->scan_wait);
3408 * scan_thread() - kernel thread used to rescan controllers
3409 * @data: Ignored.
3411 * A kernel thread used scan for drive topology changes on
3412 * controllers. The thread processes only one controller at a time
3413 * using a queue. Controllers are added to the queue using
3414 * add_to_scan_list() and removed from the queue either after done
3415 * processing or using remove_from_scan_list().
3417 * returns 0.
3419 static int scan_thread(void *data)
3421 struct ctlr_info *h;
3423 while (1) {
3424 set_current_state(TASK_INTERRUPTIBLE);
3425 schedule();
3426 if (kthread_should_stop())
3427 break;
3429 while (1) {
3430 mutex_lock(&scan_mutex);
3431 if (list_empty(&scan_q)) {
3432 mutex_unlock(&scan_mutex);
3433 break;
3436 h = list_entry(scan_q.next,
3437 struct ctlr_info,
3438 scan_list);
3439 list_del(&h->scan_list);
3440 h->busy_scanning = 1;
3441 mutex_unlock(&scan_mutex);
3443 if (h) {
3444 rebuild_lun_table(h, 0);
3445 complete_all(&h->scan_wait);
3446 mutex_lock(&scan_mutex);
3447 h->busy_scanning = 0;
3448 mutex_unlock(&scan_mutex);
3453 return 0;
3456 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3458 if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3459 return 0;
3461 switch (c->err_info->SenseInfo[12]) {
3462 case STATE_CHANGED:
3463 printk(KERN_WARNING "cciss%d: a state change "
3464 "detected, command retried\n", h->ctlr);
3465 return 1;
3466 break;
3467 case LUN_FAILED:
3468 printk(KERN_WARNING "cciss%d: LUN failure "
3469 "detected, action required\n", h->ctlr);
3470 return 1;
3471 break;
3472 case REPORT_LUNS_CHANGED:
3473 printk(KERN_WARNING "cciss%d: report LUN data "
3474 "changed\n", h->ctlr);
3475 add_to_scan_list(h);
3476 wake_up_process(cciss_scan_thread);
3477 return 1;
3478 break;
3479 case POWER_OR_RESET:
3480 printk(KERN_WARNING "cciss%d: a power on "
3481 "or device reset detected\n", h->ctlr);
3482 return 1;
3483 break;
3484 case UNIT_ATTENTION_CLEARED:
3485 printk(KERN_WARNING "cciss%d: unit attention "
3486 "cleared by another initiator\n", h->ctlr);
3487 return 1;
3488 break;
3489 default:
3490 printk(KERN_WARNING "cciss%d: unknown "
3491 "unit attention detected\n", h->ctlr);
3492 return 1;
3497 * We cannot read the structure directly, for portability we must use
3498 * the io functions.
3499 * This is for debug only.
3501 #ifdef CCISS_DEBUG
3502 static void print_cfg_table(CfgTable_struct *tb)
3504 int i;
3505 char temp_name[17];
3507 printk("Controller Configuration information\n");
3508 printk("------------------------------------\n");
3509 for (i = 0; i < 4; i++)
3510 temp_name[i] = readb(&(tb->Signature[i]));
3511 temp_name[4] = '\0';
3512 printk(" Signature = %s\n", temp_name);
3513 printk(" Spec Number = %d\n", readl(&(tb->SpecValence)));
3514 printk(" Transport methods supported = 0x%x\n",
3515 readl(&(tb->TransportSupport)));
3516 printk(" Transport methods active = 0x%x\n",
3517 readl(&(tb->TransportActive)));
3518 printk(" Requested transport Method = 0x%x\n",
3519 readl(&(tb->HostWrite.TransportRequest)));
3520 printk(" Coalesce Interrupt Delay = 0x%x\n",
3521 readl(&(tb->HostWrite.CoalIntDelay)));
3522 printk(" Coalesce Interrupt Count = 0x%x\n",
3523 readl(&(tb->HostWrite.CoalIntCount)));
3524 printk(" Max outstanding commands = 0x%d\n",
3525 readl(&(tb->CmdsOutMax)));
3526 printk(" Bus Types = 0x%x\n", readl(&(tb->BusTypes)));
3527 for (i = 0; i < 16; i++)
3528 temp_name[i] = readb(&(tb->ServerName[i]));
3529 temp_name[16] = '\0';
3530 printk(" Server Name = %s\n", temp_name);
3531 printk(" Heartbeat Counter = 0x%x\n\n\n", readl(&(tb->HeartBeat)));
3533 #endif /* CCISS_DEBUG */
3535 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3537 int i, offset, mem_type, bar_type;
3538 if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3539 return 0;
3540 offset = 0;
3541 for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3542 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3543 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3544 offset += 4;
3545 else {
3546 mem_type = pci_resource_flags(pdev, i) &
3547 PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3548 switch (mem_type) {
3549 case PCI_BASE_ADDRESS_MEM_TYPE_32:
3550 case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3551 offset += 4; /* 32 bit */
3552 break;
3553 case PCI_BASE_ADDRESS_MEM_TYPE_64:
3554 offset += 8;
3555 break;
3556 default: /* reserved in PCI 2.2 */
3557 printk(KERN_WARNING
3558 "Base address is invalid\n");
3559 return -1;
3560 break;
3563 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3564 return i + 1;
3566 return -1;
3569 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
3570 * controllers that are capable. If not, we use IO-APIC mode.
3573 static void __devinit cciss_interrupt_mode(ctlr_info_t *c,
3574 struct pci_dev *pdev, __u32 board_id)
3576 #ifdef CONFIG_PCI_MSI
3577 int err;
3578 struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
3579 {0, 2}, {0, 3}
3582 /* Some boards advertise MSI but don't really support it */
3583 if ((board_id == 0x40700E11) ||
3584 (board_id == 0x40800E11) ||
3585 (board_id == 0x40820E11) || (board_id == 0x40830E11))
3586 goto default_int_mode;
3588 if (pci_find_capability(pdev, PCI_CAP_ID_MSIX)) {
3589 err = pci_enable_msix(pdev, cciss_msix_entries, 4);
3590 if (!err) {
3591 c->intr[0] = cciss_msix_entries[0].vector;
3592 c->intr[1] = cciss_msix_entries[1].vector;
3593 c->intr[2] = cciss_msix_entries[2].vector;
3594 c->intr[3] = cciss_msix_entries[3].vector;
3595 c->msix_vector = 1;
3596 return;
3598 if (err > 0) {
3599 printk(KERN_WARNING "cciss: only %d MSI-X vectors "
3600 "available\n", err);
3601 goto default_int_mode;
3602 } else {
3603 printk(KERN_WARNING "cciss: MSI-X init failed %d\n",
3604 err);
3605 goto default_int_mode;
3608 if (pci_find_capability(pdev, PCI_CAP_ID_MSI)) {
3609 if (!pci_enable_msi(pdev)) {
3610 c->msi_vector = 1;
3611 } else {
3612 printk(KERN_WARNING "cciss: MSI init failed\n");
3615 default_int_mode:
3616 #endif /* CONFIG_PCI_MSI */
3617 /* if we get here we're going to use the default interrupt mode */
3618 c->intr[SIMPLE_MODE_INT] = pdev->irq;
3619 return;
3622 static int __devinit cciss_pci_init(ctlr_info_t *c, struct pci_dev *pdev)
3624 ushort subsystem_vendor_id, subsystem_device_id, command;
3625 __u32 board_id, scratchpad = 0;
3626 __u64 cfg_offset;
3627 __u32 cfg_base_addr;
3628 __u64 cfg_base_addr_index;
3629 int i, err;
3631 /* check to see if controller has been disabled */
3632 /* BEFORE trying to enable it */
3633 (void)pci_read_config_word(pdev, PCI_COMMAND, &command);
3634 if (!(command & 0x02)) {
3635 printk(KERN_WARNING
3636 "cciss: controller appears to be disabled\n");
3637 return -ENODEV;
3640 err = pci_enable_device(pdev);
3641 if (err) {
3642 printk(KERN_ERR "cciss: Unable to Enable PCI device\n");
3643 return err;
3646 err = pci_request_regions(pdev, "cciss");
3647 if (err) {
3648 printk(KERN_ERR "cciss: Cannot obtain PCI resources, "
3649 "aborting\n");
3650 return err;
3653 subsystem_vendor_id = pdev->subsystem_vendor;
3654 subsystem_device_id = pdev->subsystem_device;
3655 board_id = (((__u32) (subsystem_device_id << 16) & 0xffff0000) |
3656 subsystem_vendor_id);
3658 #ifdef CCISS_DEBUG
3659 printk("command = %x\n", command);
3660 printk("irq = %x\n", pdev->irq);
3661 printk("board_id = %x\n", board_id);
3662 #endif /* CCISS_DEBUG */
3664 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
3665 * else we use the IO-APIC interrupt assigned to us by system ROM.
3667 cciss_interrupt_mode(c, pdev, board_id);
3669 /* find the memory BAR */
3670 for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3671 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM)
3672 break;
3674 if (i == DEVICE_COUNT_RESOURCE) {
3675 printk(KERN_WARNING "cciss: No memory BAR found\n");
3676 err = -ENODEV;
3677 goto err_out_free_res;
3680 c->paddr = pci_resource_start(pdev, i); /* addressing mode bits
3681 * already removed
3684 #ifdef CCISS_DEBUG
3685 printk("address 0 = %lx\n", c->paddr);
3686 #endif /* CCISS_DEBUG */
3687 c->vaddr = remap_pci_mem(c->paddr, 0x250);
3689 /* Wait for the board to become ready. (PCI hotplug needs this.)
3690 * We poll for up to 120 secs, once per 100ms. */
3691 for (i = 0; i < 1200; i++) {
3692 scratchpad = readl(c->vaddr + SA5_SCRATCHPAD_OFFSET);
3693 if (scratchpad == CCISS_FIRMWARE_READY)
3694 break;
3695 set_current_state(TASK_INTERRUPTIBLE);
3696 schedule_timeout(msecs_to_jiffies(100)); /* wait 100ms */
3698 if (scratchpad != CCISS_FIRMWARE_READY) {
3699 printk(KERN_WARNING "cciss: Board not ready. Timed out.\n");
3700 err = -ENODEV;
3701 goto err_out_free_res;
3704 /* get the address index number */
3705 cfg_base_addr = readl(c->vaddr + SA5_CTCFG_OFFSET);
3706 cfg_base_addr &= (__u32) 0x0000ffff;
3707 #ifdef CCISS_DEBUG
3708 printk("cfg base address = %x\n", cfg_base_addr);
3709 #endif /* CCISS_DEBUG */
3710 cfg_base_addr_index = find_PCI_BAR_index(pdev, cfg_base_addr);
3711 #ifdef CCISS_DEBUG
3712 printk("cfg base address index = %llx\n",
3713 (unsigned long long)cfg_base_addr_index);
3714 #endif /* CCISS_DEBUG */
3715 if (cfg_base_addr_index == -1) {
3716 printk(KERN_WARNING "cciss: Cannot find cfg_base_addr_index\n");
3717 err = -ENODEV;
3718 goto err_out_free_res;
3721 cfg_offset = readl(c->vaddr + SA5_CTMEM_OFFSET);
3722 #ifdef CCISS_DEBUG
3723 printk("cfg offset = %llx\n", (unsigned long long)cfg_offset);
3724 #endif /* CCISS_DEBUG */
3725 c->cfgtable = remap_pci_mem(pci_resource_start(pdev,
3726 cfg_base_addr_index) +
3727 cfg_offset, sizeof(CfgTable_struct));
3728 c->board_id = board_id;
3730 #ifdef CCISS_DEBUG
3731 print_cfg_table(c->cfgtable);
3732 #endif /* CCISS_DEBUG */
3734 /* Some controllers support Zero Memory Raid (ZMR).
3735 * When configured in ZMR mode the number of supported
3736 * commands drops to 64. So instead of just setting an
3737 * arbitrary value we make the driver a little smarter.
3738 * We read the config table to tell us how many commands
3739 * are supported on the controller then subtract 4 to
3740 * leave a little room for ioctl calls.
3742 c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
3743 for (i = 0; i < ARRAY_SIZE(products); i++) {
3744 if (board_id == products[i].board_id) {
3745 c->product_name = products[i].product_name;
3746 c->access = *(products[i].access);
3747 c->nr_cmds = c->max_commands - 4;
3748 break;
3751 if ((readb(&c->cfgtable->Signature[0]) != 'C') ||
3752 (readb(&c->cfgtable->Signature[1]) != 'I') ||
3753 (readb(&c->cfgtable->Signature[2]) != 'S') ||
3754 (readb(&c->cfgtable->Signature[3]) != 'S')) {
3755 printk("Does not appear to be a valid CISS config table\n");
3756 err = -ENODEV;
3757 goto err_out_free_res;
3759 /* We didn't find the controller in our list. We know the
3760 * signature is valid. If it's an HP device let's try to
3761 * bind to the device and fire it up. Otherwise we bail.
3763 if (i == ARRAY_SIZE(products)) {
3764 if (subsystem_vendor_id == PCI_VENDOR_ID_HP) {
3765 c->product_name = products[i-1].product_name;
3766 c->access = *(products[i-1].access);
3767 c->nr_cmds = c->max_commands - 4;
3768 printk(KERN_WARNING "cciss: This is an unknown "
3769 "Smart Array controller.\n"
3770 "cciss: Please update to the latest driver "
3771 "available from www.hp.com.\n");
3772 } else {
3773 printk(KERN_WARNING "cciss: Sorry, I don't know how"
3774 " to access the Smart Array controller %08lx\n"
3775 , (unsigned long)board_id);
3776 err = -ENODEV;
3777 goto err_out_free_res;
3780 #ifdef CONFIG_X86
3782 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
3783 __u32 prefetch;
3784 prefetch = readl(&(c->cfgtable->SCSI_Prefetch));
3785 prefetch |= 0x100;
3786 writel(prefetch, &(c->cfgtable->SCSI_Prefetch));
3788 #endif
3790 /* Disabling DMA prefetch and refetch for the P600.
3791 * An ASIC bug may result in accesses to invalid memory addresses.
3792 * We've disabled prefetch for some time now. Testing with XEN
3793 * kernels revealed a bug in the refetch if dom0 resides on a P600.
3795 if(board_id == 0x3225103C) {
3796 __u32 dma_prefetch;
3797 __u32 dma_refetch;
3798 dma_prefetch = readl(c->vaddr + I2O_DMA1_CFG);
3799 dma_prefetch |= 0x8000;
3800 writel(dma_prefetch, c->vaddr + I2O_DMA1_CFG);
3801 pci_read_config_dword(pdev, PCI_COMMAND_PARITY, &dma_refetch);
3802 dma_refetch |= 0x1;
3803 pci_write_config_dword(pdev, PCI_COMMAND_PARITY, dma_refetch);
3806 #ifdef CCISS_DEBUG
3807 printk("Trying to put board into Simple mode\n");
3808 #endif /* CCISS_DEBUG */
3809 c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
3810 /* Update the field, and then ring the doorbell */
3811 writel(CFGTBL_Trans_Simple, &(c->cfgtable->HostWrite.TransportRequest));
3812 writel(CFGTBL_ChangeReq, c->vaddr + SA5_DOORBELL);
3814 /* under certain very rare conditions, this can take awhile.
3815 * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3816 * as we enter this code.) */
3817 for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3818 if (!(readl(c->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3819 break;
3820 /* delay and try again */
3821 set_current_state(TASK_INTERRUPTIBLE);
3822 schedule_timeout(msecs_to_jiffies(1));
3825 #ifdef CCISS_DEBUG
3826 printk(KERN_DEBUG "I counter got to %d %x\n", i,
3827 readl(c->vaddr + SA5_DOORBELL));
3828 #endif /* CCISS_DEBUG */
3829 #ifdef CCISS_DEBUG
3830 print_cfg_table(c->cfgtable);
3831 #endif /* CCISS_DEBUG */
3833 if (!(readl(&(c->cfgtable->TransportActive)) & CFGTBL_Trans_Simple)) {
3834 printk(KERN_WARNING "cciss: unable to get board into"
3835 " simple mode\n");
3836 err = -ENODEV;
3837 goto err_out_free_res;
3839 return 0;
3841 err_out_free_res:
3843 * Deliberately omit pci_disable_device(): it does something nasty to
3844 * Smart Array controllers that pci_enable_device does not undo
3846 pci_release_regions(pdev);
3847 return err;
3850 /* Function to find the first free pointer into our hba[] array
3851 * Returns -1 if no free entries are left.
3853 static int alloc_cciss_hba(void)
3855 int i;
3857 for (i = 0; i < MAX_CTLR; i++) {
3858 if (!hba[i]) {
3859 ctlr_info_t *p;
3861 p = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
3862 if (!p)
3863 goto Enomem;
3864 hba[i] = p;
3865 return i;
3868 printk(KERN_WARNING "cciss: This driver supports a maximum"
3869 " of %d controllers.\n", MAX_CTLR);
3870 return -1;
3871 Enomem:
3872 printk(KERN_ERR "cciss: out of memory.\n");
3873 return -1;
3876 static void free_hba(int i)
3878 ctlr_info_t *p = hba[i];
3879 int n;
3881 hba[i] = NULL;
3882 for (n = 0; n < CISS_MAX_LUN; n++)
3883 put_disk(p->gendisk[n]);
3884 kfree(p);
3887 /* Send a message CDB to the firmware. */
3888 static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
3890 typedef struct {
3891 CommandListHeader_struct CommandHeader;
3892 RequestBlock_struct Request;
3893 ErrDescriptor_struct ErrorDescriptor;
3894 } Command;
3895 static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
3896 Command *cmd;
3897 dma_addr_t paddr64;
3898 uint32_t paddr32, tag;
3899 void __iomem *vaddr;
3900 int i, err;
3902 vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
3903 if (vaddr == NULL)
3904 return -ENOMEM;
3906 /* The Inbound Post Queue only accepts 32-bit physical addresses for the
3907 CCISS commands, so they must be allocated from the lower 4GiB of
3908 memory. */
3909 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
3910 if (err) {
3911 iounmap(vaddr);
3912 return -ENOMEM;
3915 cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
3916 if (cmd == NULL) {
3917 iounmap(vaddr);
3918 return -ENOMEM;
3921 /* This must fit, because of the 32-bit consistent DMA mask. Also,
3922 although there's no guarantee, we assume that the address is at
3923 least 4-byte aligned (most likely, it's page-aligned). */
3924 paddr32 = paddr64;
3926 cmd->CommandHeader.ReplyQueue = 0;
3927 cmd->CommandHeader.SGList = 0;
3928 cmd->CommandHeader.SGTotal = 0;
3929 cmd->CommandHeader.Tag.lower = paddr32;
3930 cmd->CommandHeader.Tag.upper = 0;
3931 memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
3933 cmd->Request.CDBLen = 16;
3934 cmd->Request.Type.Type = TYPE_MSG;
3935 cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
3936 cmd->Request.Type.Direction = XFER_NONE;
3937 cmd->Request.Timeout = 0; /* Don't time out */
3938 cmd->Request.CDB[0] = opcode;
3939 cmd->Request.CDB[1] = type;
3940 memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
3942 cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
3943 cmd->ErrorDescriptor.Addr.upper = 0;
3944 cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
3946 writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
3948 for (i = 0; i < 10; i++) {
3949 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
3950 if ((tag & ~3) == paddr32)
3951 break;
3952 schedule_timeout_uninterruptible(HZ);
3955 iounmap(vaddr);
3957 /* we leak the DMA buffer here ... no choice since the controller could
3958 still complete the command. */
3959 if (i == 10) {
3960 printk(KERN_ERR "cciss: controller message %02x:%02x timed out\n",
3961 opcode, type);
3962 return -ETIMEDOUT;
3965 pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
3967 if (tag & 2) {
3968 printk(KERN_ERR "cciss: controller message %02x:%02x failed\n",
3969 opcode, type);
3970 return -EIO;
3973 printk(KERN_INFO "cciss: controller message %02x:%02x succeeded\n",
3974 opcode, type);
3975 return 0;
3978 #define cciss_soft_reset_controller(p) cciss_message(p, 1, 0)
3979 #define cciss_noop(p) cciss_message(p, 3, 0)
3981 static __devinit int cciss_reset_msi(struct pci_dev *pdev)
3983 /* the #defines are stolen from drivers/pci/msi.h. */
3984 #define msi_control_reg(base) (base + PCI_MSI_FLAGS)
3985 #define PCI_MSIX_FLAGS_ENABLE (1 << 15)
3987 int pos;
3988 u16 control = 0;
3990 pos = pci_find_capability(pdev, PCI_CAP_ID_MSI);
3991 if (pos) {
3992 pci_read_config_word(pdev, msi_control_reg(pos), &control);
3993 if (control & PCI_MSI_FLAGS_ENABLE) {
3994 printk(KERN_INFO "cciss: resetting MSI\n");
3995 pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSI_FLAGS_ENABLE);
3999 pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
4000 if (pos) {
4001 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4002 if (control & PCI_MSIX_FLAGS_ENABLE) {
4003 printk(KERN_INFO "cciss: resetting MSI-X\n");
4004 pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSIX_FLAGS_ENABLE);
4008 return 0;
4011 /* This does a hard reset of the controller using PCI power management
4012 * states. */
4013 static __devinit int cciss_hard_reset_controller(struct pci_dev *pdev)
4015 u16 pmcsr, saved_config_space[32];
4016 int i, pos;
4018 printk(KERN_INFO "cciss: using PCI PM to reset controller\n");
4020 /* This is very nearly the same thing as
4022 pci_save_state(pci_dev);
4023 pci_set_power_state(pci_dev, PCI_D3hot);
4024 pci_set_power_state(pci_dev, PCI_D0);
4025 pci_restore_state(pci_dev);
4027 but we can't use these nice canned kernel routines on
4028 kexec, because they also check the MSI/MSI-X state in PCI
4029 configuration space and do the wrong thing when it is
4030 set/cleared. Also, the pci_save/restore_state functions
4031 violate the ordering requirements for restoring the
4032 configuration space from the CCISS document (see the
4033 comment below). So we roll our own .... */
4035 for (i = 0; i < 32; i++)
4036 pci_read_config_word(pdev, 2*i, &saved_config_space[i]);
4038 pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4039 if (pos == 0) {
4040 printk(KERN_ERR "cciss_reset_controller: PCI PM not supported\n");
4041 return -ENODEV;
4044 /* Quoting from the Open CISS Specification: "The Power
4045 * Management Control/Status Register (CSR) controls the power
4046 * state of the device. The normal operating state is D0,
4047 * CSR=00h. The software off state is D3, CSR=03h. To reset
4048 * the controller, place the interface device in D3 then to
4049 * D0, this causes a secondary PCI reset which will reset the
4050 * controller." */
4052 /* enter the D3hot power management state */
4053 pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4054 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4055 pmcsr |= PCI_D3hot;
4056 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4058 schedule_timeout_uninterruptible(HZ >> 1);
4060 /* enter the D0 power management state */
4061 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4062 pmcsr |= PCI_D0;
4063 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4065 schedule_timeout_uninterruptible(HZ >> 1);
4067 /* Restore the PCI configuration space. The Open CISS
4068 * Specification says, "Restore the PCI Configuration
4069 * Registers, offsets 00h through 60h. It is important to
4070 * restore the command register, 16-bits at offset 04h,
4071 * last. Do not restore the configuration status register,
4072 * 16-bits at offset 06h." Note that the offset is 2*i. */
4073 for (i = 0; i < 32; i++) {
4074 if (i == 2 || i == 3)
4075 continue;
4076 pci_write_config_word(pdev, 2*i, saved_config_space[i]);
4078 wmb();
4079 pci_write_config_word(pdev, 4, saved_config_space[2]);
4081 return 0;
4085 * This is it. Find all the controllers and register them. I really hate
4086 * stealing all these major device numbers.
4087 * returns the number of block devices registered.
4089 static int __devinit cciss_init_one(struct pci_dev *pdev,
4090 const struct pci_device_id *ent)
4092 int i;
4093 int j = 0;
4094 int rc;
4095 int dac, return_code;
4096 InquiryData_struct *inq_buff;
4098 if (reset_devices) {
4099 /* Reset the controller with a PCI power-cycle */
4100 if (cciss_hard_reset_controller(pdev) || cciss_reset_msi(pdev))
4101 return -ENODEV;
4103 /* Now try to get the controller to respond to a no-op. Some
4104 devices (notably the HP Smart Array 5i Controller) need
4105 up to 30 seconds to respond. */
4106 for (i=0; i<30; i++) {
4107 if (cciss_noop(pdev) == 0)
4108 break;
4110 schedule_timeout_uninterruptible(HZ);
4112 if (i == 30) {
4113 printk(KERN_ERR "cciss: controller seems dead\n");
4114 return -EBUSY;
4118 i = alloc_cciss_hba();
4119 if (i < 0)
4120 return -1;
4122 hba[i]->busy_initializing = 1;
4123 INIT_HLIST_HEAD(&hba[i]->cmpQ);
4124 INIT_HLIST_HEAD(&hba[i]->reqQ);
4125 mutex_init(&hba[i]->busy_shutting_down);
4127 if (cciss_pci_init(hba[i], pdev) != 0)
4128 goto clean0;
4130 sprintf(hba[i]->devname, "cciss%d", i);
4131 hba[i]->ctlr = i;
4132 hba[i]->pdev = pdev;
4134 init_completion(&hba[i]->scan_wait);
4136 if (cciss_create_hba_sysfs_entry(hba[i]))
4137 goto clean0;
4139 /* configure PCI DMA stuff */
4140 if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
4141 dac = 1;
4142 else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
4143 dac = 0;
4144 else {
4145 printk(KERN_ERR "cciss: no suitable DMA available\n");
4146 goto clean1;
4150 * register with the major number, or get a dynamic major number
4151 * by passing 0 as argument. This is done for greater than
4152 * 8 controller support.
4154 if (i < MAX_CTLR_ORIG)
4155 hba[i]->major = COMPAQ_CISS_MAJOR + i;
4156 rc = register_blkdev(hba[i]->major, hba[i]->devname);
4157 if (rc == -EBUSY || rc == -EINVAL) {
4158 printk(KERN_ERR
4159 "cciss: Unable to get major number %d for %s "
4160 "on hba %d\n", hba[i]->major, hba[i]->devname, i);
4161 goto clean1;
4162 } else {
4163 if (i >= MAX_CTLR_ORIG)
4164 hba[i]->major = rc;
4167 /* make sure the board interrupts are off */
4168 hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_OFF);
4169 if (request_irq(hba[i]->intr[SIMPLE_MODE_INT], do_cciss_intr,
4170 IRQF_DISABLED | IRQF_SHARED, hba[i]->devname, hba[i])) {
4171 printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
4172 hba[i]->intr[SIMPLE_MODE_INT], hba[i]->devname);
4173 goto clean2;
4176 printk(KERN_INFO "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
4177 hba[i]->devname, pdev->device, pci_name(pdev),
4178 hba[i]->intr[SIMPLE_MODE_INT], dac ? "" : " not");
4180 hba[i]->cmd_pool_bits =
4181 kmalloc(DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4182 * sizeof(unsigned long), GFP_KERNEL);
4183 hba[i]->cmd_pool = (CommandList_struct *)
4184 pci_alloc_consistent(hba[i]->pdev,
4185 hba[i]->nr_cmds * sizeof(CommandList_struct),
4186 &(hba[i]->cmd_pool_dhandle));
4187 hba[i]->errinfo_pool = (ErrorInfo_struct *)
4188 pci_alloc_consistent(hba[i]->pdev,
4189 hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4190 &(hba[i]->errinfo_pool_dhandle));
4191 if ((hba[i]->cmd_pool_bits == NULL)
4192 || (hba[i]->cmd_pool == NULL)
4193 || (hba[i]->errinfo_pool == NULL)) {
4194 printk(KERN_ERR "cciss: out of memory");
4195 goto clean4;
4197 spin_lock_init(&hba[i]->lock);
4199 /* Initialize the pdev driver private data.
4200 have it point to hba[i]. */
4201 pci_set_drvdata(pdev, hba[i]);
4202 /* command and error info recs zeroed out before
4203 they are used */
4204 memset(hba[i]->cmd_pool_bits, 0,
4205 DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4206 * sizeof(unsigned long));
4208 hba[i]->num_luns = 0;
4209 hba[i]->highest_lun = -1;
4210 for (j = 0; j < CISS_MAX_LUN; j++) {
4211 hba[i]->drv[j].raid_level = -1;
4212 hba[i]->drv[j].queue = NULL;
4213 hba[i]->gendisk[j] = NULL;
4216 cciss_scsi_setup(i);
4218 /* Turn the interrupts on so we can service requests */
4219 hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_ON);
4221 /* Get the firmware version */
4222 inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
4223 if (inq_buff == NULL) {
4224 printk(KERN_ERR "cciss: out of memory\n");
4225 goto clean4;
4228 return_code = sendcmd_withirq(CISS_INQUIRY, i, inq_buff,
4229 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
4230 if (return_code == IO_OK) {
4231 hba[i]->firm_ver[0] = inq_buff->data_byte[32];
4232 hba[i]->firm_ver[1] = inq_buff->data_byte[33];
4233 hba[i]->firm_ver[2] = inq_buff->data_byte[34];
4234 hba[i]->firm_ver[3] = inq_buff->data_byte[35];
4235 } else { /* send command failed */
4236 printk(KERN_WARNING "cciss: unable to determine firmware"
4237 " version of controller\n");
4239 kfree(inq_buff);
4241 cciss_procinit(i);
4243 hba[i]->cciss_max_sectors = 2048;
4245 rebuild_lun_table(hba[i], 1);
4246 hba[i]->busy_initializing = 0;
4247 return 1;
4249 clean4:
4250 kfree(hba[i]->cmd_pool_bits);
4251 if (hba[i]->cmd_pool)
4252 pci_free_consistent(hba[i]->pdev,
4253 hba[i]->nr_cmds * sizeof(CommandList_struct),
4254 hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4255 if (hba[i]->errinfo_pool)
4256 pci_free_consistent(hba[i]->pdev,
4257 hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4258 hba[i]->errinfo_pool,
4259 hba[i]->errinfo_pool_dhandle);
4260 free_irq(hba[i]->intr[SIMPLE_MODE_INT], hba[i]);
4261 clean2:
4262 unregister_blkdev(hba[i]->major, hba[i]->devname);
4263 clean1:
4264 cciss_destroy_hba_sysfs_entry(hba[i]);
4265 clean0:
4266 hba[i]->busy_initializing = 0;
4267 /* cleanup any queues that may have been initialized */
4268 for (j=0; j <= hba[i]->highest_lun; j++){
4269 drive_info_struct *drv = &(hba[i]->drv[j]);
4270 if (drv->queue)
4271 blk_cleanup_queue(drv->queue);
4274 * Deliberately omit pci_disable_device(): it does something nasty to
4275 * Smart Array controllers that pci_enable_device does not undo
4277 pci_release_regions(pdev);
4278 pci_set_drvdata(pdev, NULL);
4279 free_hba(i);
4280 return -1;
4283 static void cciss_shutdown(struct pci_dev *pdev)
4285 ctlr_info_t *tmp_ptr;
4286 int i;
4287 char flush_buf[4];
4288 int return_code;
4290 tmp_ptr = pci_get_drvdata(pdev);
4291 if (tmp_ptr == NULL)
4292 return;
4293 i = tmp_ptr->ctlr;
4294 if (hba[i] == NULL)
4295 return;
4297 /* Turn board interrupts off and send the flush cache command */
4298 /* sendcmd will turn off interrupt, and send the flush...
4299 * To write all data in the battery backed cache to disks */
4300 memset(flush_buf, 0, 4);
4301 return_code = sendcmd(CCISS_CACHE_FLUSH, i, flush_buf, 4, 0,
4302 CTLR_LUNID, TYPE_CMD);
4303 if (return_code == IO_OK) {
4304 printk(KERN_INFO "Completed flushing cache on controller %d\n", i);
4305 } else {
4306 printk(KERN_WARNING "Error flushing cache on controller %d\n", i);
4308 free_irq(hba[i]->intr[2], hba[i]);
4311 static void __devexit cciss_remove_one(struct pci_dev *pdev)
4313 ctlr_info_t *tmp_ptr;
4314 int i, j;
4316 if (pci_get_drvdata(pdev) == NULL) {
4317 printk(KERN_ERR "cciss: Unable to remove device \n");
4318 return;
4321 tmp_ptr = pci_get_drvdata(pdev);
4322 i = tmp_ptr->ctlr;
4323 if (hba[i] == NULL) {
4324 printk(KERN_ERR "cciss: device appears to "
4325 "already be removed \n");
4326 return;
4329 mutex_lock(&hba[i]->busy_shutting_down);
4331 remove_from_scan_list(hba[i]);
4332 remove_proc_entry(hba[i]->devname, proc_cciss);
4333 unregister_blkdev(hba[i]->major, hba[i]->devname);
4335 /* remove it from the disk list */
4336 for (j = 0; j < CISS_MAX_LUN; j++) {
4337 struct gendisk *disk = hba[i]->gendisk[j];
4338 if (disk) {
4339 struct request_queue *q = disk->queue;
4341 if (disk->flags & GENHD_FL_UP) {
4342 cciss_destroy_ld_sysfs_entry(hba[i], j, 1);
4343 del_gendisk(disk);
4345 if (q)
4346 blk_cleanup_queue(q);
4350 #ifdef CONFIG_CISS_SCSI_TAPE
4351 cciss_unregister_scsi(i); /* unhook from SCSI subsystem */
4352 #endif
4354 cciss_shutdown(pdev);
4356 #ifdef CONFIG_PCI_MSI
4357 if (hba[i]->msix_vector)
4358 pci_disable_msix(hba[i]->pdev);
4359 else if (hba[i]->msi_vector)
4360 pci_disable_msi(hba[i]->pdev);
4361 #endif /* CONFIG_PCI_MSI */
4363 iounmap(hba[i]->vaddr);
4365 pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(CommandList_struct),
4366 hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4367 pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4368 hba[i]->errinfo_pool, hba[i]->errinfo_pool_dhandle);
4369 kfree(hba[i]->cmd_pool_bits);
4371 * Deliberately omit pci_disable_device(): it does something nasty to
4372 * Smart Array controllers that pci_enable_device does not undo
4374 pci_release_regions(pdev);
4375 pci_set_drvdata(pdev, NULL);
4376 cciss_destroy_hba_sysfs_entry(hba[i]);
4377 mutex_unlock(&hba[i]->busy_shutting_down);
4378 free_hba(i);
4381 static struct pci_driver cciss_pci_driver = {
4382 .name = "cciss",
4383 .probe = cciss_init_one,
4384 .remove = __devexit_p(cciss_remove_one),
4385 .id_table = cciss_pci_device_id, /* id_table */
4386 .shutdown = cciss_shutdown,
4390 * This is it. Register the PCI driver information for the cards we control
4391 * the OS will call our registered routines when it finds one of our cards.
4393 static int __init cciss_init(void)
4395 int err;
4398 * The hardware requires that commands are aligned on a 64-bit
4399 * boundary. Given that we use pci_alloc_consistent() to allocate an
4400 * array of them, the size must be a multiple of 8 bytes.
4402 BUILD_BUG_ON(sizeof(CommandList_struct) % 8);
4404 printk(KERN_INFO DRIVER_NAME "\n");
4406 err = bus_register(&cciss_bus_type);
4407 if (err)
4408 return err;
4410 /* Start the scan thread */
4411 cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
4412 if (IS_ERR(cciss_scan_thread)) {
4413 err = PTR_ERR(cciss_scan_thread);
4414 goto err_bus_unregister;
4417 /* Register for our PCI devices */
4418 err = pci_register_driver(&cciss_pci_driver);
4419 if (err)
4420 goto err_thread_stop;
4422 return err;
4424 err_thread_stop:
4425 kthread_stop(cciss_scan_thread);
4426 err_bus_unregister:
4427 bus_unregister(&cciss_bus_type);
4429 return err;
4432 static void __exit cciss_cleanup(void)
4434 int i;
4436 pci_unregister_driver(&cciss_pci_driver);
4437 /* double check that all controller entrys have been removed */
4438 for (i = 0; i < MAX_CTLR; i++) {
4439 if (hba[i] != NULL) {
4440 printk(KERN_WARNING "cciss: had to remove"
4441 " controller %d\n", i);
4442 cciss_remove_one(hba[i]->pdev);
4445 kthread_stop(cciss_scan_thread);
4446 remove_proc_entry("driver/cciss", NULL);
4447 bus_unregister(&cciss_bus_type);
4450 static void fail_all_cmds(unsigned long ctlr)
4452 /* If we get here, the board is apparently dead. */
4453 ctlr_info_t *h = hba[ctlr];
4454 CommandList_struct *c;
4455 unsigned long flags;
4457 printk(KERN_WARNING "cciss%d: controller not responding.\n", h->ctlr);
4458 h->alive = 0; /* the controller apparently died... */
4460 spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
4462 pci_disable_device(h->pdev); /* Make sure it is really dead. */
4464 /* move everything off the request queue onto the completed queue */
4465 while (!hlist_empty(&h->reqQ)) {
4466 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
4467 removeQ(c);
4468 h->Qdepth--;
4469 addQ(&h->cmpQ, c);
4472 /* Now, fail everything on the completed queue with a HW error */
4473 while (!hlist_empty(&h->cmpQ)) {
4474 c = hlist_entry(h->cmpQ.first, CommandList_struct, list);
4475 removeQ(c);
4476 if (c->cmd_type != CMD_MSG_STALE)
4477 c->err_info->CommandStatus = CMD_HARDWARE_ERR;
4478 if (c->cmd_type == CMD_RWREQ) {
4479 complete_command(h, c, 0);
4480 } else if (c->cmd_type == CMD_IOCTL_PEND)
4481 complete(c->waiting);
4482 #ifdef CONFIG_CISS_SCSI_TAPE
4483 else if (c->cmd_type == CMD_SCSI)
4484 complete_scsi_command(c, 0, 0);
4485 #endif
4487 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
4488 return;
4491 module_init(cciss_init);
4492 module_exit(cciss_cleanup);