cciss: export resettable host attribute
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / block / cciss.c
blobeeed7aeb0b83accea60232cd62b2bd0680a7ae26
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/delay.h>
30 #include <linux/major.h>
31 #include <linux/fs.h>
32 #include <linux/bio.h>
33 #include <linux/blkpg.h>
34 #include <linux/timer.h>
35 #include <linux/proc_fs.h>
36 #include <linux/seq_file.h>
37 #include <linux/init.h>
38 #include <linux/jiffies.h>
39 #include <linux/hdreg.h>
40 #include <linux/spinlock.h>
41 #include <linux/compat.h>
42 #include <linux/mutex.h>
43 #include <asm/uaccess.h>
44 #include <asm/io.h>
46 #include <linux/dma-mapping.h>
47 #include <linux/blkdev.h>
48 #include <linux/genhd.h>
49 #include <linux/completion.h>
50 #include <scsi/scsi.h>
51 #include <scsi/sg.h>
52 #include <scsi/scsi_ioctl.h>
53 #include <linux/cdrom.h>
54 #include <linux/scatterlist.h>
55 #include <linux/kthread.h>
57 #define CCISS_DRIVER_VERSION(maj,min,submin) ((maj<<16)|(min<<8)|(submin))
58 #define DRIVER_NAME "HP CISS Driver (v 3.6.26)"
59 #define DRIVER_VERSION CCISS_DRIVER_VERSION(3, 6, 26)
61 /* Embedded module documentation macros - see modules.h */
62 MODULE_AUTHOR("Hewlett-Packard Company");
63 MODULE_DESCRIPTION("Driver for HP Smart Array Controllers");
64 MODULE_SUPPORTED_DEVICE("HP Smart Array Controllers");
65 MODULE_VERSION("3.6.26");
66 MODULE_LICENSE("GPL");
68 static DEFINE_MUTEX(cciss_mutex);
69 static struct proc_dir_entry *proc_cciss;
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 {0,}
100 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
102 /* board_id = Subsystem Device ID & Vendor ID
103 * product = Marketing Name for the board
104 * access = Address of the struct of function pointers
106 static struct board_type products[] = {
107 {0x40700E11, "Smart Array 5300", &SA5_access},
108 {0x40800E11, "Smart Array 5i", &SA5B_access},
109 {0x40820E11, "Smart Array 532", &SA5B_access},
110 {0x40830E11, "Smart Array 5312", &SA5B_access},
111 {0x409A0E11, "Smart Array 641", &SA5_access},
112 {0x409B0E11, "Smart Array 642", &SA5_access},
113 {0x409C0E11, "Smart Array 6400", &SA5_access},
114 {0x409D0E11, "Smart Array 6400 EM", &SA5_access},
115 {0x40910E11, "Smart Array 6i", &SA5_access},
116 {0x3225103C, "Smart Array P600", &SA5_access},
117 {0x3223103C, "Smart Array P800", &SA5_access},
118 {0x3234103C, "Smart Array P400", &SA5_access},
119 {0x3235103C, "Smart Array P400i", &SA5_access},
120 {0x3211103C, "Smart Array E200i", &SA5_access},
121 {0x3212103C, "Smart Array E200", &SA5_access},
122 {0x3213103C, "Smart Array E200i", &SA5_access},
123 {0x3214103C, "Smart Array E200i", &SA5_access},
124 {0x3215103C, "Smart Array E200i", &SA5_access},
125 {0x3237103C, "Smart Array E500", &SA5_access},
126 {0x3223103C, "Smart Array P800", &SA5_access},
127 {0x3234103C, "Smart Array P400", &SA5_access},
128 {0x323D103C, "Smart Array P700m", &SA5_access},
131 /* How long to wait (in milliseconds) for board to go into simple mode */
132 #define MAX_CONFIG_WAIT 30000
133 #define MAX_IOCTL_CONFIG_WAIT 1000
135 /*define how many times we will try a command because of bus resets */
136 #define MAX_CMD_RETRIES 3
138 #define MAX_CTLR 32
140 /* Originally cciss driver only supports 8 major numbers */
141 #define MAX_CTLR_ORIG 8
143 static ctlr_info_t *hba[MAX_CTLR];
145 static struct task_struct *cciss_scan_thread;
146 static DEFINE_MUTEX(scan_mutex);
147 static LIST_HEAD(scan_q);
149 static void do_cciss_request(struct request_queue *q);
150 static irqreturn_t do_cciss_intx(int irq, void *dev_id);
151 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id);
152 static int cciss_open(struct block_device *bdev, fmode_t mode);
153 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode);
154 static int cciss_release(struct gendisk *disk, fmode_t mode);
155 static int do_ioctl(struct block_device *bdev, fmode_t mode,
156 unsigned int cmd, unsigned long arg);
157 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
158 unsigned int cmd, unsigned long arg);
159 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
161 static int cciss_revalidate(struct gendisk *disk);
162 static int rebuild_lun_table(ctlr_info_t *h, int first_time, int via_ioctl);
163 static int deregister_disk(ctlr_info_t *h, int drv_index,
164 int clear_all, int via_ioctl);
166 static void cciss_read_capacity(ctlr_info_t *h, int logvol,
167 sector_t *total_size, unsigned int *block_size);
168 static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
169 sector_t *total_size, unsigned int *block_size);
170 static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
171 sector_t total_size,
172 unsigned int block_size, InquiryData_struct *inq_buff,
173 drive_info_struct *drv);
174 static void __devinit cciss_interrupt_mode(ctlr_info_t *);
175 static void start_io(ctlr_info_t *h);
176 static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
177 __u8 page_code, unsigned char scsi3addr[],
178 int cmd_type);
179 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
180 int attempt_retry);
181 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c);
183 static int add_to_scan_list(struct ctlr_info *h);
184 static int scan_thread(void *data);
185 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c);
186 static void cciss_hba_release(struct device *dev);
187 static void cciss_device_release(struct device *dev);
188 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index);
189 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index);
190 static inline u32 next_command(ctlr_info_t *h);
191 static int __devinit cciss_find_cfg_addrs(struct pci_dev *pdev,
192 void __iomem *vaddr, u32 *cfg_base_addr, u64 *cfg_base_addr_index,
193 u64 *cfg_offset);
194 static int __devinit cciss_pci_find_memory_BAR(struct pci_dev *pdev,
195 unsigned long *memory_bar);
198 /* performant mode helper functions */
199 static void calc_bucket_map(int *bucket, int num_buckets, int nsgs,
200 int *bucket_map);
201 static void cciss_put_controller_into_performant_mode(ctlr_info_t *h);
203 #ifdef CONFIG_PROC_FS
204 static void cciss_procinit(ctlr_info_t *h);
205 #else
206 static void cciss_procinit(ctlr_info_t *h)
209 #endif /* CONFIG_PROC_FS */
211 #ifdef CONFIG_COMPAT
212 static int cciss_compat_ioctl(struct block_device *, fmode_t,
213 unsigned, unsigned long);
214 #endif
216 static const struct block_device_operations cciss_fops = {
217 .owner = THIS_MODULE,
218 .open = cciss_unlocked_open,
219 .release = cciss_release,
220 .ioctl = do_ioctl,
221 .getgeo = cciss_getgeo,
222 #ifdef CONFIG_COMPAT
223 .compat_ioctl = cciss_compat_ioctl,
224 #endif
225 .revalidate_disk = cciss_revalidate,
228 /* set_performant_mode: Modify the tag for cciss performant
229 * set bit 0 for pull model, bits 3-1 for block fetch
230 * register number
232 static void set_performant_mode(ctlr_info_t *h, CommandList_struct *c)
234 if (likely(h->transMethod == CFGTBL_Trans_Performant))
235 c->busaddr |= 1 | (h->blockFetchTable[c->Header.SGList] << 1);
239 * Enqueuing and dequeuing functions for cmdlists.
241 static inline void addQ(struct list_head *list, CommandList_struct *c)
243 list_add_tail(&c->list, list);
246 static inline void removeQ(CommandList_struct *c)
249 * After kexec/dump some commands might still
250 * be in flight, which the firmware will try
251 * to complete. Resetting the firmware doesn't work
252 * with old fw revisions, so we have to mark
253 * them off as 'stale' to prevent the driver from
254 * falling over.
256 if (WARN_ON(list_empty(&c->list))) {
257 c->cmd_type = CMD_MSG_STALE;
258 return;
261 list_del_init(&c->list);
264 static void enqueue_cmd_and_start_io(ctlr_info_t *h,
265 CommandList_struct *c)
267 unsigned long flags;
268 set_performant_mode(h, c);
269 spin_lock_irqsave(&h->lock, flags);
270 addQ(&h->reqQ, c);
271 h->Qdepth++;
272 if (h->Qdepth > h->maxQsinceinit)
273 h->maxQsinceinit = h->Qdepth;
274 start_io(h);
275 spin_unlock_irqrestore(&h->lock, flags);
278 static void cciss_free_sg_chain_blocks(SGDescriptor_struct **cmd_sg_list,
279 int nr_cmds)
281 int i;
283 if (!cmd_sg_list)
284 return;
285 for (i = 0; i < nr_cmds; i++) {
286 kfree(cmd_sg_list[i]);
287 cmd_sg_list[i] = NULL;
289 kfree(cmd_sg_list);
292 static SGDescriptor_struct **cciss_allocate_sg_chain_blocks(
293 ctlr_info_t *h, int chainsize, int nr_cmds)
295 int j;
296 SGDescriptor_struct **cmd_sg_list;
298 if (chainsize <= 0)
299 return NULL;
301 cmd_sg_list = kmalloc(sizeof(*cmd_sg_list) * nr_cmds, GFP_KERNEL);
302 if (!cmd_sg_list)
303 return NULL;
305 /* Build up chain blocks for each command */
306 for (j = 0; j < nr_cmds; j++) {
307 /* Need a block of chainsized s/g elements. */
308 cmd_sg_list[j] = kmalloc((chainsize *
309 sizeof(*cmd_sg_list[j])), GFP_KERNEL);
310 if (!cmd_sg_list[j]) {
311 dev_err(&h->pdev->dev, "Cannot get memory "
312 "for s/g chains.\n");
313 goto clean;
316 return cmd_sg_list;
317 clean:
318 cciss_free_sg_chain_blocks(cmd_sg_list, nr_cmds);
319 return NULL;
322 static void cciss_unmap_sg_chain_block(ctlr_info_t *h, CommandList_struct *c)
324 SGDescriptor_struct *chain_sg;
325 u64bit temp64;
327 if (c->Header.SGTotal <= h->max_cmd_sgentries)
328 return;
330 chain_sg = &c->SG[h->max_cmd_sgentries - 1];
331 temp64.val32.lower = chain_sg->Addr.lower;
332 temp64.val32.upper = chain_sg->Addr.upper;
333 pci_unmap_single(h->pdev, temp64.val, chain_sg->Len, PCI_DMA_TODEVICE);
336 static void cciss_map_sg_chain_block(ctlr_info_t *h, CommandList_struct *c,
337 SGDescriptor_struct *chain_block, int len)
339 SGDescriptor_struct *chain_sg;
340 u64bit temp64;
342 chain_sg = &c->SG[h->max_cmd_sgentries - 1];
343 chain_sg->Ext = CCISS_SG_CHAIN;
344 chain_sg->Len = len;
345 temp64.val = pci_map_single(h->pdev, chain_block, len,
346 PCI_DMA_TODEVICE);
347 chain_sg->Addr.lower = temp64.val32.lower;
348 chain_sg->Addr.upper = temp64.val32.upper;
351 #include "cciss_scsi.c" /* For SCSI tape support */
353 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
354 "UNKNOWN"
356 #define RAID_UNKNOWN (ARRAY_SIZE(raid_label)-1)
358 #ifdef CONFIG_PROC_FS
361 * Report information about this controller.
363 #define ENG_GIG 1000000000
364 #define ENG_GIG_FACTOR (ENG_GIG/512)
365 #define ENGAGE_SCSI "engage scsi"
367 static void cciss_seq_show_header(struct seq_file *seq)
369 ctlr_info_t *h = seq->private;
371 seq_printf(seq, "%s: HP %s Controller\n"
372 "Board ID: 0x%08lx\n"
373 "Firmware Version: %c%c%c%c\n"
374 "IRQ: %d\n"
375 "Logical drives: %d\n"
376 "Current Q depth: %d\n"
377 "Current # commands on controller: %d\n"
378 "Max Q depth since init: %d\n"
379 "Max # commands on controller since init: %d\n"
380 "Max SG entries since init: %d\n",
381 h->devname,
382 h->product_name,
383 (unsigned long)h->board_id,
384 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
385 h->firm_ver[3], (unsigned int)h->intr[PERF_MODE_INT],
386 h->num_luns,
387 h->Qdepth, h->commands_outstanding,
388 h->maxQsinceinit, h->max_outstanding, h->maxSG);
390 #ifdef CONFIG_CISS_SCSI_TAPE
391 cciss_seq_tape_report(seq, h);
392 #endif /* CONFIG_CISS_SCSI_TAPE */
395 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
397 ctlr_info_t *h = seq->private;
398 unsigned long flags;
400 /* prevent displaying bogus info during configuration
401 * or deconfiguration of a logical volume
403 spin_lock_irqsave(&h->lock, flags);
404 if (h->busy_configuring) {
405 spin_unlock_irqrestore(&h->lock, flags);
406 return ERR_PTR(-EBUSY);
408 h->busy_configuring = 1;
409 spin_unlock_irqrestore(&h->lock, flags);
411 if (*pos == 0)
412 cciss_seq_show_header(seq);
414 return pos;
417 static int cciss_seq_show(struct seq_file *seq, void *v)
419 sector_t vol_sz, vol_sz_frac;
420 ctlr_info_t *h = seq->private;
421 unsigned ctlr = h->ctlr;
422 loff_t *pos = v;
423 drive_info_struct *drv = h->drv[*pos];
425 if (*pos > h->highest_lun)
426 return 0;
428 if (drv == NULL) /* it's possible for h->drv[] to have holes. */
429 return 0;
431 if (drv->heads == 0)
432 return 0;
434 vol_sz = drv->nr_blocks;
435 vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
436 vol_sz_frac *= 100;
437 sector_div(vol_sz_frac, ENG_GIG_FACTOR);
439 if (drv->raid_level < 0 || drv->raid_level > RAID_UNKNOWN)
440 drv->raid_level = RAID_UNKNOWN;
441 seq_printf(seq, "cciss/c%dd%d:"
442 "\t%4u.%02uGB\tRAID %s\n",
443 ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
444 raid_label[drv->raid_level]);
445 return 0;
448 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
450 ctlr_info_t *h = seq->private;
452 if (*pos > h->highest_lun)
453 return NULL;
454 *pos += 1;
456 return pos;
459 static void cciss_seq_stop(struct seq_file *seq, void *v)
461 ctlr_info_t *h = seq->private;
463 /* Only reset h->busy_configuring if we succeeded in setting
464 * it during cciss_seq_start. */
465 if (v == ERR_PTR(-EBUSY))
466 return;
468 h->busy_configuring = 0;
471 static const struct seq_operations cciss_seq_ops = {
472 .start = cciss_seq_start,
473 .show = cciss_seq_show,
474 .next = cciss_seq_next,
475 .stop = cciss_seq_stop,
478 static int cciss_seq_open(struct inode *inode, struct file *file)
480 int ret = seq_open(file, &cciss_seq_ops);
481 struct seq_file *seq = file->private_data;
483 if (!ret)
484 seq->private = PDE(inode)->data;
486 return ret;
489 static ssize_t
490 cciss_proc_write(struct file *file, const char __user *buf,
491 size_t length, loff_t *ppos)
493 int err;
494 char *buffer;
496 #ifndef CONFIG_CISS_SCSI_TAPE
497 return -EINVAL;
498 #endif
500 if (!buf || length > PAGE_SIZE - 1)
501 return -EINVAL;
503 buffer = (char *)__get_free_page(GFP_KERNEL);
504 if (!buffer)
505 return -ENOMEM;
507 err = -EFAULT;
508 if (copy_from_user(buffer, buf, length))
509 goto out;
510 buffer[length] = '\0';
512 #ifdef CONFIG_CISS_SCSI_TAPE
513 if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
514 struct seq_file *seq = file->private_data;
515 ctlr_info_t *h = seq->private;
517 err = cciss_engage_scsi(h);
518 if (err == 0)
519 err = length;
520 } else
521 #endif /* CONFIG_CISS_SCSI_TAPE */
522 err = -EINVAL;
523 /* might be nice to have "disengage" too, but it's not
524 safely possible. (only 1 module use count, lock issues.) */
526 out:
527 free_page((unsigned long)buffer);
528 return err;
531 static const struct file_operations cciss_proc_fops = {
532 .owner = THIS_MODULE,
533 .open = cciss_seq_open,
534 .read = seq_read,
535 .llseek = seq_lseek,
536 .release = seq_release,
537 .write = cciss_proc_write,
540 static void __devinit cciss_procinit(ctlr_info_t *h)
542 struct proc_dir_entry *pde;
544 if (proc_cciss == NULL)
545 proc_cciss = proc_mkdir("driver/cciss", NULL);
546 if (!proc_cciss)
547 return;
548 pde = proc_create_data(h->devname, S_IWUSR | S_IRUSR | S_IRGRP |
549 S_IROTH, proc_cciss,
550 &cciss_proc_fops, h);
552 #endif /* CONFIG_PROC_FS */
554 #define MAX_PRODUCT_NAME_LEN 19
556 #define to_hba(n) container_of(n, struct ctlr_info, dev)
557 #define to_drv(n) container_of(n, drive_info_struct, dev)
559 /* List of controllers which cannot be reset on kexec with reset_devices */
560 static u32 unresettable_controller[] = {
561 0x324a103C, /* Smart Array P712m */
562 0x324b103C, /* SmartArray P711m */
563 0x3223103C, /* Smart Array P800 */
564 0x3234103C, /* Smart Array P400 */
565 0x3235103C, /* Smart Array P400i */
566 0x3211103C, /* Smart Array E200i */
567 0x3212103C, /* Smart Array E200 */
568 0x3213103C, /* Smart Array E200i */
569 0x3214103C, /* Smart Array E200i */
570 0x3215103C, /* Smart Array E200i */
571 0x3237103C, /* Smart Array E500 */
572 0x323D103C, /* Smart Array P700m */
573 0x409C0E11, /* Smart Array 6400 */
574 0x409D0E11, /* Smart Array 6400 EM */
577 static int ctlr_is_resettable(struct ctlr_info *h)
579 int i;
581 for (i = 0; i < ARRAY_SIZE(unresettable_controller); i++)
582 if (unresettable_controller[i] == h->board_id)
583 return 0;
584 return 1;
587 static ssize_t host_show_resettable(struct device *dev,
588 struct device_attribute *attr,
589 char *buf)
591 struct ctlr_info *h = to_hba(dev);
593 return snprintf(buf, 20, "%d\n", ctlr_is_resettable(h));
595 static DEVICE_ATTR(resettable, S_IRUGO, host_show_resettable, NULL);
597 static ssize_t host_store_rescan(struct device *dev,
598 struct device_attribute *attr,
599 const char *buf, size_t count)
601 struct ctlr_info *h = to_hba(dev);
603 add_to_scan_list(h);
604 wake_up_process(cciss_scan_thread);
605 wait_for_completion_interruptible(&h->scan_wait);
607 return count;
609 static DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
611 static ssize_t dev_show_unique_id(struct device *dev,
612 struct device_attribute *attr,
613 char *buf)
615 drive_info_struct *drv = to_drv(dev);
616 struct ctlr_info *h = to_hba(drv->dev.parent);
617 __u8 sn[16];
618 unsigned long flags;
619 int ret = 0;
621 spin_lock_irqsave(&h->lock, flags);
622 if (h->busy_configuring)
623 ret = -EBUSY;
624 else
625 memcpy(sn, drv->serial_no, sizeof(sn));
626 spin_unlock_irqrestore(&h->lock, flags);
628 if (ret)
629 return ret;
630 else
631 return snprintf(buf, 16 * 2 + 2,
632 "%02X%02X%02X%02X%02X%02X%02X%02X"
633 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
634 sn[0], sn[1], sn[2], sn[3],
635 sn[4], sn[5], sn[6], sn[7],
636 sn[8], sn[9], sn[10], sn[11],
637 sn[12], sn[13], sn[14], sn[15]);
639 static DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
641 static ssize_t dev_show_vendor(struct device *dev,
642 struct device_attribute *attr,
643 char *buf)
645 drive_info_struct *drv = to_drv(dev);
646 struct ctlr_info *h = to_hba(drv->dev.parent);
647 char vendor[VENDOR_LEN + 1];
648 unsigned long flags;
649 int ret = 0;
651 spin_lock_irqsave(&h->lock, flags);
652 if (h->busy_configuring)
653 ret = -EBUSY;
654 else
655 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
656 spin_unlock_irqrestore(&h->lock, flags);
658 if (ret)
659 return ret;
660 else
661 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
663 static DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
665 static ssize_t dev_show_model(struct device *dev,
666 struct device_attribute *attr,
667 char *buf)
669 drive_info_struct *drv = to_drv(dev);
670 struct ctlr_info *h = to_hba(drv->dev.parent);
671 char model[MODEL_LEN + 1];
672 unsigned long flags;
673 int ret = 0;
675 spin_lock_irqsave(&h->lock, flags);
676 if (h->busy_configuring)
677 ret = -EBUSY;
678 else
679 memcpy(model, drv->model, MODEL_LEN + 1);
680 spin_unlock_irqrestore(&h->lock, flags);
682 if (ret)
683 return ret;
684 else
685 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
687 static DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
689 static ssize_t dev_show_rev(struct device *dev,
690 struct device_attribute *attr,
691 char *buf)
693 drive_info_struct *drv = to_drv(dev);
694 struct ctlr_info *h = to_hba(drv->dev.parent);
695 char rev[REV_LEN + 1];
696 unsigned long flags;
697 int ret = 0;
699 spin_lock_irqsave(&h->lock, flags);
700 if (h->busy_configuring)
701 ret = -EBUSY;
702 else
703 memcpy(rev, drv->rev, REV_LEN + 1);
704 spin_unlock_irqrestore(&h->lock, flags);
706 if (ret)
707 return ret;
708 else
709 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
711 static DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
713 static ssize_t cciss_show_lunid(struct device *dev,
714 struct device_attribute *attr, char *buf)
716 drive_info_struct *drv = to_drv(dev);
717 struct ctlr_info *h = to_hba(drv->dev.parent);
718 unsigned long flags;
719 unsigned char lunid[8];
721 spin_lock_irqsave(&h->lock, flags);
722 if (h->busy_configuring) {
723 spin_unlock_irqrestore(&h->lock, flags);
724 return -EBUSY;
726 if (!drv->heads) {
727 spin_unlock_irqrestore(&h->lock, flags);
728 return -ENOTTY;
730 memcpy(lunid, drv->LunID, sizeof(lunid));
731 spin_unlock_irqrestore(&h->lock, flags);
732 return snprintf(buf, 20, "0x%02x%02x%02x%02x%02x%02x%02x%02x\n",
733 lunid[0], lunid[1], lunid[2], lunid[3],
734 lunid[4], lunid[5], lunid[6], lunid[7]);
736 static DEVICE_ATTR(lunid, S_IRUGO, cciss_show_lunid, NULL);
738 static ssize_t cciss_show_raid_level(struct device *dev,
739 struct device_attribute *attr, char *buf)
741 drive_info_struct *drv = to_drv(dev);
742 struct ctlr_info *h = to_hba(drv->dev.parent);
743 int raid;
744 unsigned long flags;
746 spin_lock_irqsave(&h->lock, flags);
747 if (h->busy_configuring) {
748 spin_unlock_irqrestore(&h->lock, flags);
749 return -EBUSY;
751 raid = drv->raid_level;
752 spin_unlock_irqrestore(&h->lock, flags);
753 if (raid < 0 || raid > RAID_UNKNOWN)
754 raid = RAID_UNKNOWN;
756 return snprintf(buf, strlen(raid_label[raid]) + 7, "RAID %s\n",
757 raid_label[raid]);
759 static DEVICE_ATTR(raid_level, S_IRUGO, cciss_show_raid_level, NULL);
761 static ssize_t cciss_show_usage_count(struct device *dev,
762 struct device_attribute *attr, char *buf)
764 drive_info_struct *drv = to_drv(dev);
765 struct ctlr_info *h = to_hba(drv->dev.parent);
766 unsigned long flags;
767 int count;
769 spin_lock_irqsave(&h->lock, flags);
770 if (h->busy_configuring) {
771 spin_unlock_irqrestore(&h->lock, flags);
772 return -EBUSY;
774 count = drv->usage_count;
775 spin_unlock_irqrestore(&h->lock, flags);
776 return snprintf(buf, 20, "%d\n", count);
778 static DEVICE_ATTR(usage_count, S_IRUGO, cciss_show_usage_count, NULL);
780 static struct attribute *cciss_host_attrs[] = {
781 &dev_attr_rescan.attr,
782 &dev_attr_resettable.attr,
783 NULL
786 static struct attribute_group cciss_host_attr_group = {
787 .attrs = cciss_host_attrs,
790 static const struct attribute_group *cciss_host_attr_groups[] = {
791 &cciss_host_attr_group,
792 NULL
795 static struct device_type cciss_host_type = {
796 .name = "cciss_host",
797 .groups = cciss_host_attr_groups,
798 .release = cciss_hba_release,
801 static struct attribute *cciss_dev_attrs[] = {
802 &dev_attr_unique_id.attr,
803 &dev_attr_model.attr,
804 &dev_attr_vendor.attr,
805 &dev_attr_rev.attr,
806 &dev_attr_lunid.attr,
807 &dev_attr_raid_level.attr,
808 &dev_attr_usage_count.attr,
809 NULL
812 static struct attribute_group cciss_dev_attr_group = {
813 .attrs = cciss_dev_attrs,
816 static const struct attribute_group *cciss_dev_attr_groups[] = {
817 &cciss_dev_attr_group,
818 NULL
821 static struct device_type cciss_dev_type = {
822 .name = "cciss_device",
823 .groups = cciss_dev_attr_groups,
824 .release = cciss_device_release,
827 static struct bus_type cciss_bus_type = {
828 .name = "cciss",
832 * cciss_hba_release is called when the reference count
833 * of h->dev goes to zero.
835 static void cciss_hba_release(struct device *dev)
838 * nothing to do, but need this to avoid a warning
839 * about not having a release handler from lib/kref.c.
844 * Initialize sysfs entry for each controller. This sets up and registers
845 * the 'cciss#' directory for each individual controller under
846 * /sys/bus/pci/devices/<dev>/.
848 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
850 device_initialize(&h->dev);
851 h->dev.type = &cciss_host_type;
852 h->dev.bus = &cciss_bus_type;
853 dev_set_name(&h->dev, "%s", h->devname);
854 h->dev.parent = &h->pdev->dev;
856 return device_add(&h->dev);
860 * Remove sysfs entries for an hba.
862 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
864 device_del(&h->dev);
865 put_device(&h->dev); /* final put. */
868 /* cciss_device_release is called when the reference count
869 * of h->drv[x]dev goes to zero.
871 static void cciss_device_release(struct device *dev)
873 drive_info_struct *drv = to_drv(dev);
874 kfree(drv);
878 * Initialize sysfs for each logical drive. This sets up and registers
879 * the 'c#d#' directory for each individual logical drive under
880 * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
881 * /sys/block/cciss!c#d# to this entry.
883 static long cciss_create_ld_sysfs_entry(struct ctlr_info *h,
884 int drv_index)
886 struct device *dev;
888 if (h->drv[drv_index]->device_initialized)
889 return 0;
891 dev = &h->drv[drv_index]->dev;
892 device_initialize(dev);
893 dev->type = &cciss_dev_type;
894 dev->bus = &cciss_bus_type;
895 dev_set_name(dev, "c%dd%d", h->ctlr, drv_index);
896 dev->parent = &h->dev;
897 h->drv[drv_index]->device_initialized = 1;
898 return device_add(dev);
902 * Remove sysfs entries for a logical drive.
904 static void cciss_destroy_ld_sysfs_entry(struct ctlr_info *h, int drv_index,
905 int ctlr_exiting)
907 struct device *dev = &h->drv[drv_index]->dev;
909 /* special case for c*d0, we only destroy it on controller exit */
910 if (drv_index == 0 && !ctlr_exiting)
911 return;
913 device_del(dev);
914 put_device(dev); /* the "final" put. */
915 h->drv[drv_index] = NULL;
919 * For operations that cannot sleep, a command block is allocated at init,
920 * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
921 * which ones are free or in use.
923 static CommandList_struct *cmd_alloc(ctlr_info_t *h)
925 CommandList_struct *c;
926 int i;
927 u64bit temp64;
928 dma_addr_t cmd_dma_handle, err_dma_handle;
930 do {
931 i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
932 if (i == h->nr_cmds)
933 return NULL;
934 } while (test_and_set_bit(i & (BITS_PER_LONG - 1),
935 h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
936 c = h->cmd_pool + i;
937 memset(c, 0, sizeof(CommandList_struct));
938 cmd_dma_handle = h->cmd_pool_dhandle + i * sizeof(CommandList_struct);
939 c->err_info = h->errinfo_pool + i;
940 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
941 err_dma_handle = h->errinfo_pool_dhandle
942 + i * sizeof(ErrorInfo_struct);
943 h->nr_allocs++;
945 c->cmdindex = i;
947 INIT_LIST_HEAD(&c->list);
948 c->busaddr = (__u32) cmd_dma_handle;
949 temp64.val = (__u64) err_dma_handle;
950 c->ErrDesc.Addr.lower = temp64.val32.lower;
951 c->ErrDesc.Addr.upper = temp64.val32.upper;
952 c->ErrDesc.Len = sizeof(ErrorInfo_struct);
954 c->ctlr = h->ctlr;
955 return c;
958 /* allocate a command using pci_alloc_consistent, used for ioctls,
959 * etc., not for the main i/o path.
961 static CommandList_struct *cmd_special_alloc(ctlr_info_t *h)
963 CommandList_struct *c;
964 u64bit temp64;
965 dma_addr_t cmd_dma_handle, err_dma_handle;
967 c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
968 sizeof(CommandList_struct), &cmd_dma_handle);
969 if (c == NULL)
970 return NULL;
971 memset(c, 0, sizeof(CommandList_struct));
973 c->cmdindex = -1;
975 c->err_info = (ErrorInfo_struct *)
976 pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
977 &err_dma_handle);
979 if (c->err_info == NULL) {
980 pci_free_consistent(h->pdev,
981 sizeof(CommandList_struct), c, cmd_dma_handle);
982 return NULL;
984 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
986 INIT_LIST_HEAD(&c->list);
987 c->busaddr = (__u32) cmd_dma_handle;
988 temp64.val = (__u64) err_dma_handle;
989 c->ErrDesc.Addr.lower = temp64.val32.lower;
990 c->ErrDesc.Addr.upper = temp64.val32.upper;
991 c->ErrDesc.Len = sizeof(ErrorInfo_struct);
993 c->ctlr = h->ctlr;
994 return c;
997 static void cmd_free(ctlr_info_t *h, CommandList_struct *c)
999 int i;
1001 i = c - h->cmd_pool;
1002 clear_bit(i & (BITS_PER_LONG - 1),
1003 h->cmd_pool_bits + (i / BITS_PER_LONG));
1004 h->nr_frees++;
1007 static void cmd_special_free(ctlr_info_t *h, CommandList_struct *c)
1009 u64bit temp64;
1011 temp64.val32.lower = c->ErrDesc.Addr.lower;
1012 temp64.val32.upper = c->ErrDesc.Addr.upper;
1013 pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
1014 c->err_info, (dma_addr_t) temp64.val);
1015 pci_free_consistent(h->pdev, sizeof(CommandList_struct),
1016 c, (dma_addr_t) c->busaddr);
1019 static inline ctlr_info_t *get_host(struct gendisk *disk)
1021 return disk->queue->queuedata;
1024 static inline drive_info_struct *get_drv(struct gendisk *disk)
1026 return disk->private_data;
1030 * Open. Make sure the device is really there.
1032 static int cciss_open(struct block_device *bdev, fmode_t mode)
1034 ctlr_info_t *h = get_host(bdev->bd_disk);
1035 drive_info_struct *drv = get_drv(bdev->bd_disk);
1037 dev_dbg(&h->pdev->dev, "cciss_open %s\n", bdev->bd_disk->disk_name);
1038 if (drv->busy_configuring)
1039 return -EBUSY;
1041 * Root is allowed to open raw volume zero even if it's not configured
1042 * so array config can still work. Root is also allowed to open any
1043 * volume that has a LUN ID, so it can issue IOCTL to reread the
1044 * disk information. I don't think I really like this
1045 * but I'm already using way to many device nodes to claim another one
1046 * for "raw controller".
1048 if (drv->heads == 0) {
1049 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
1050 /* if not node 0 make sure it is a partition = 0 */
1051 if (MINOR(bdev->bd_dev) & 0x0f) {
1052 return -ENXIO;
1053 /* if it is, make sure we have a LUN ID */
1054 } else if (memcmp(drv->LunID, CTLR_LUNID,
1055 sizeof(drv->LunID))) {
1056 return -ENXIO;
1059 if (!capable(CAP_SYS_ADMIN))
1060 return -EPERM;
1062 drv->usage_count++;
1063 h->usage_count++;
1064 return 0;
1067 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode)
1069 int ret;
1071 mutex_lock(&cciss_mutex);
1072 ret = cciss_open(bdev, mode);
1073 mutex_unlock(&cciss_mutex);
1075 return ret;
1079 * Close. Sync first.
1081 static int cciss_release(struct gendisk *disk, fmode_t mode)
1083 ctlr_info_t *h;
1084 drive_info_struct *drv;
1086 mutex_lock(&cciss_mutex);
1087 h = get_host(disk);
1088 drv = get_drv(disk);
1089 dev_dbg(&h->pdev->dev, "cciss_release %s\n", disk->disk_name);
1090 drv->usage_count--;
1091 h->usage_count--;
1092 mutex_unlock(&cciss_mutex);
1093 return 0;
1096 static int do_ioctl(struct block_device *bdev, fmode_t mode,
1097 unsigned cmd, unsigned long arg)
1099 int ret;
1100 mutex_lock(&cciss_mutex);
1101 ret = cciss_ioctl(bdev, mode, cmd, arg);
1102 mutex_unlock(&cciss_mutex);
1103 return ret;
1106 #ifdef CONFIG_COMPAT
1108 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1109 unsigned cmd, unsigned long arg);
1110 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1111 unsigned cmd, unsigned long arg);
1113 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
1114 unsigned cmd, unsigned long arg)
1116 switch (cmd) {
1117 case CCISS_GETPCIINFO:
1118 case CCISS_GETINTINFO:
1119 case CCISS_SETINTINFO:
1120 case CCISS_GETNODENAME:
1121 case CCISS_SETNODENAME:
1122 case CCISS_GETHEARTBEAT:
1123 case CCISS_GETBUSTYPES:
1124 case CCISS_GETFIRMVER:
1125 case CCISS_GETDRIVVER:
1126 case CCISS_REVALIDVOLS:
1127 case CCISS_DEREGDISK:
1128 case CCISS_REGNEWDISK:
1129 case CCISS_REGNEWD:
1130 case CCISS_RESCANDISK:
1131 case CCISS_GETLUNINFO:
1132 return do_ioctl(bdev, mode, cmd, arg);
1134 case CCISS_PASSTHRU32:
1135 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
1136 case CCISS_BIG_PASSTHRU32:
1137 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
1139 default:
1140 return -ENOIOCTLCMD;
1144 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1145 unsigned cmd, unsigned long arg)
1147 IOCTL32_Command_struct __user *arg32 =
1148 (IOCTL32_Command_struct __user *) arg;
1149 IOCTL_Command_struct arg64;
1150 IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
1151 int err;
1152 u32 cp;
1154 err = 0;
1155 err |=
1156 copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1157 sizeof(arg64.LUN_info));
1158 err |=
1159 copy_from_user(&arg64.Request, &arg32->Request,
1160 sizeof(arg64.Request));
1161 err |=
1162 copy_from_user(&arg64.error_info, &arg32->error_info,
1163 sizeof(arg64.error_info));
1164 err |= get_user(arg64.buf_size, &arg32->buf_size);
1165 err |= get_user(cp, &arg32->buf);
1166 arg64.buf = compat_ptr(cp);
1167 err |= copy_to_user(p, &arg64, sizeof(arg64));
1169 if (err)
1170 return -EFAULT;
1172 err = do_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
1173 if (err)
1174 return err;
1175 err |=
1176 copy_in_user(&arg32->error_info, &p->error_info,
1177 sizeof(arg32->error_info));
1178 if (err)
1179 return -EFAULT;
1180 return err;
1183 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1184 unsigned cmd, unsigned long arg)
1186 BIG_IOCTL32_Command_struct __user *arg32 =
1187 (BIG_IOCTL32_Command_struct __user *) arg;
1188 BIG_IOCTL_Command_struct arg64;
1189 BIG_IOCTL_Command_struct __user *p =
1190 compat_alloc_user_space(sizeof(arg64));
1191 int err;
1192 u32 cp;
1194 memset(&arg64, 0, sizeof(arg64));
1195 err = 0;
1196 err |=
1197 copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1198 sizeof(arg64.LUN_info));
1199 err |=
1200 copy_from_user(&arg64.Request, &arg32->Request,
1201 sizeof(arg64.Request));
1202 err |=
1203 copy_from_user(&arg64.error_info, &arg32->error_info,
1204 sizeof(arg64.error_info));
1205 err |= get_user(arg64.buf_size, &arg32->buf_size);
1206 err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1207 err |= get_user(cp, &arg32->buf);
1208 arg64.buf = compat_ptr(cp);
1209 err |= copy_to_user(p, &arg64, sizeof(arg64));
1211 if (err)
1212 return -EFAULT;
1214 err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1215 if (err)
1216 return err;
1217 err |=
1218 copy_in_user(&arg32->error_info, &p->error_info,
1219 sizeof(arg32->error_info));
1220 if (err)
1221 return -EFAULT;
1222 return err;
1224 #endif
1226 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1228 drive_info_struct *drv = get_drv(bdev->bd_disk);
1230 if (!drv->cylinders)
1231 return -ENXIO;
1233 geo->heads = drv->heads;
1234 geo->sectors = drv->sectors;
1235 geo->cylinders = drv->cylinders;
1236 return 0;
1239 static void check_ioctl_unit_attention(ctlr_info_t *h, CommandList_struct *c)
1241 if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1242 c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1243 (void)check_for_unit_attention(h, c);
1246 static int cciss_getpciinfo(ctlr_info_t *h, void __user *argp)
1248 cciss_pci_info_struct pciinfo;
1250 if (!argp)
1251 return -EINVAL;
1252 pciinfo.domain = pci_domain_nr(h->pdev->bus);
1253 pciinfo.bus = h->pdev->bus->number;
1254 pciinfo.dev_fn = h->pdev->devfn;
1255 pciinfo.board_id = h->board_id;
1256 if (copy_to_user(argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1257 return -EFAULT;
1258 return 0;
1261 static int cciss_getintinfo(ctlr_info_t *h, void __user *argp)
1263 cciss_coalint_struct intinfo;
1265 if (!argp)
1266 return -EINVAL;
1267 intinfo.delay = readl(&h->cfgtable->HostWrite.CoalIntDelay);
1268 intinfo.count = readl(&h->cfgtable->HostWrite.CoalIntCount);
1269 if (copy_to_user
1270 (argp, &intinfo, sizeof(cciss_coalint_struct)))
1271 return -EFAULT;
1272 return 0;
1275 static int cciss_setintinfo(ctlr_info_t *h, void __user *argp)
1277 cciss_coalint_struct intinfo;
1278 unsigned long flags;
1279 int i;
1281 if (!argp)
1282 return -EINVAL;
1283 if (!capable(CAP_SYS_ADMIN))
1284 return -EPERM;
1285 if (copy_from_user(&intinfo, argp, sizeof(intinfo)))
1286 return -EFAULT;
1287 if ((intinfo.delay == 0) && (intinfo.count == 0))
1288 return -EINVAL;
1289 spin_lock_irqsave(&h->lock, flags);
1290 /* Update the field, and then ring the doorbell */
1291 writel(intinfo.delay, &(h->cfgtable->HostWrite.CoalIntDelay));
1292 writel(intinfo.count, &(h->cfgtable->HostWrite.CoalIntCount));
1293 writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1295 for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1296 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
1297 break;
1298 udelay(1000); /* delay and try again */
1300 spin_unlock_irqrestore(&h->lock, flags);
1301 if (i >= MAX_IOCTL_CONFIG_WAIT)
1302 return -EAGAIN;
1303 return 0;
1306 static int cciss_getnodename(ctlr_info_t *h, void __user *argp)
1308 NodeName_type NodeName;
1309 int i;
1311 if (!argp)
1312 return -EINVAL;
1313 for (i = 0; i < 16; i++)
1314 NodeName[i] = readb(&h->cfgtable->ServerName[i]);
1315 if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1316 return -EFAULT;
1317 return 0;
1320 static int cciss_setnodename(ctlr_info_t *h, void __user *argp)
1322 NodeName_type NodeName;
1323 unsigned long flags;
1324 int i;
1326 if (!argp)
1327 return -EINVAL;
1328 if (!capable(CAP_SYS_ADMIN))
1329 return -EPERM;
1330 if (copy_from_user(NodeName, argp, sizeof(NodeName_type)))
1331 return -EFAULT;
1332 spin_lock_irqsave(&h->lock, flags);
1333 /* Update the field, and then ring the doorbell */
1334 for (i = 0; i < 16; i++)
1335 writeb(NodeName[i], &h->cfgtable->ServerName[i]);
1336 writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1337 for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1338 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
1339 break;
1340 udelay(1000); /* delay and try again */
1342 spin_unlock_irqrestore(&h->lock, flags);
1343 if (i >= MAX_IOCTL_CONFIG_WAIT)
1344 return -EAGAIN;
1345 return 0;
1348 static int cciss_getheartbeat(ctlr_info_t *h, void __user *argp)
1350 Heartbeat_type heartbeat;
1352 if (!argp)
1353 return -EINVAL;
1354 heartbeat = readl(&h->cfgtable->HeartBeat);
1355 if (copy_to_user(argp, &heartbeat, sizeof(Heartbeat_type)))
1356 return -EFAULT;
1357 return 0;
1360 static int cciss_getbustypes(ctlr_info_t *h, void __user *argp)
1362 BusTypes_type BusTypes;
1364 if (!argp)
1365 return -EINVAL;
1366 BusTypes = readl(&h->cfgtable->BusTypes);
1367 if (copy_to_user(argp, &BusTypes, sizeof(BusTypes_type)))
1368 return -EFAULT;
1369 return 0;
1372 static int cciss_getfirmver(ctlr_info_t *h, void __user *argp)
1374 FirmwareVer_type firmware;
1376 if (!argp)
1377 return -EINVAL;
1378 memcpy(firmware, h->firm_ver, 4);
1380 if (copy_to_user
1381 (argp, firmware, sizeof(FirmwareVer_type)))
1382 return -EFAULT;
1383 return 0;
1386 static int cciss_getdrivver(ctlr_info_t *h, void __user *argp)
1388 DriverVer_type DriverVer = DRIVER_VERSION;
1390 if (!argp)
1391 return -EINVAL;
1392 if (copy_to_user(argp, &DriverVer, sizeof(DriverVer_type)))
1393 return -EFAULT;
1394 return 0;
1397 static int cciss_getluninfo(ctlr_info_t *h,
1398 struct gendisk *disk, void __user *argp)
1400 LogvolInfo_struct luninfo;
1401 drive_info_struct *drv = get_drv(disk);
1403 if (!argp)
1404 return -EINVAL;
1405 memcpy(&luninfo.LunID, drv->LunID, sizeof(luninfo.LunID));
1406 luninfo.num_opens = drv->usage_count;
1407 luninfo.num_parts = 0;
1408 if (copy_to_user(argp, &luninfo, sizeof(LogvolInfo_struct)))
1409 return -EFAULT;
1410 return 0;
1413 static int cciss_passthru(ctlr_info_t *h, void __user *argp)
1415 IOCTL_Command_struct iocommand;
1416 CommandList_struct *c;
1417 char *buff = NULL;
1418 u64bit temp64;
1419 DECLARE_COMPLETION_ONSTACK(wait);
1421 if (!argp)
1422 return -EINVAL;
1424 if (!capable(CAP_SYS_RAWIO))
1425 return -EPERM;
1427 if (copy_from_user
1428 (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1429 return -EFAULT;
1430 if ((iocommand.buf_size < 1) &&
1431 (iocommand.Request.Type.Direction != XFER_NONE)) {
1432 return -EINVAL;
1434 if (iocommand.buf_size > 0) {
1435 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1436 if (buff == NULL)
1437 return -EFAULT;
1439 if (iocommand.Request.Type.Direction == XFER_WRITE) {
1440 /* Copy the data into the buffer we created */
1441 if (copy_from_user(buff, iocommand.buf, iocommand.buf_size)) {
1442 kfree(buff);
1443 return -EFAULT;
1445 } else {
1446 memset(buff, 0, iocommand.buf_size);
1448 c = cmd_special_alloc(h);
1449 if (!c) {
1450 kfree(buff);
1451 return -ENOMEM;
1453 /* Fill in the command type */
1454 c->cmd_type = CMD_IOCTL_PEND;
1455 /* Fill in Command Header */
1456 c->Header.ReplyQueue = 0; /* unused in simple mode */
1457 if (iocommand.buf_size > 0) { /* buffer to fill */
1458 c->Header.SGList = 1;
1459 c->Header.SGTotal = 1;
1460 } else { /* no buffers to fill */
1461 c->Header.SGList = 0;
1462 c->Header.SGTotal = 0;
1464 c->Header.LUN = iocommand.LUN_info;
1465 /* use the kernel address the cmd block for tag */
1466 c->Header.Tag.lower = c->busaddr;
1468 /* Fill in Request block */
1469 c->Request = iocommand.Request;
1471 /* Fill in the scatter gather information */
1472 if (iocommand.buf_size > 0) {
1473 temp64.val = pci_map_single(h->pdev, buff,
1474 iocommand.buf_size, PCI_DMA_BIDIRECTIONAL);
1475 c->SG[0].Addr.lower = temp64.val32.lower;
1476 c->SG[0].Addr.upper = temp64.val32.upper;
1477 c->SG[0].Len = iocommand.buf_size;
1478 c->SG[0].Ext = 0; /* we are not chaining */
1480 c->waiting = &wait;
1482 enqueue_cmd_and_start_io(h, c);
1483 wait_for_completion(&wait);
1485 /* unlock the buffers from DMA */
1486 temp64.val32.lower = c->SG[0].Addr.lower;
1487 temp64.val32.upper = c->SG[0].Addr.upper;
1488 pci_unmap_single(h->pdev, (dma_addr_t) temp64.val, iocommand.buf_size,
1489 PCI_DMA_BIDIRECTIONAL);
1490 check_ioctl_unit_attention(h, c);
1492 /* Copy the error information out */
1493 iocommand.error_info = *(c->err_info);
1494 if (copy_to_user(argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1495 kfree(buff);
1496 cmd_special_free(h, c);
1497 return -EFAULT;
1500 if (iocommand.Request.Type.Direction == XFER_READ) {
1501 /* Copy the data out of the buffer we created */
1502 if (copy_to_user(iocommand.buf, buff, iocommand.buf_size)) {
1503 kfree(buff);
1504 cmd_special_free(h, c);
1505 return -EFAULT;
1508 kfree(buff);
1509 cmd_special_free(h, c);
1510 return 0;
1513 static int cciss_bigpassthru(ctlr_info_t *h, void __user *argp)
1515 BIG_IOCTL_Command_struct *ioc;
1516 CommandList_struct *c;
1517 unsigned char **buff = NULL;
1518 int *buff_size = NULL;
1519 u64bit temp64;
1520 BYTE sg_used = 0;
1521 int status = 0;
1522 int i;
1523 DECLARE_COMPLETION_ONSTACK(wait);
1524 __u32 left;
1525 __u32 sz;
1526 BYTE __user *data_ptr;
1528 if (!argp)
1529 return -EINVAL;
1530 if (!capable(CAP_SYS_RAWIO))
1531 return -EPERM;
1532 ioc = (BIG_IOCTL_Command_struct *)
1533 kmalloc(sizeof(*ioc), GFP_KERNEL);
1534 if (!ioc) {
1535 status = -ENOMEM;
1536 goto cleanup1;
1538 if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1539 status = -EFAULT;
1540 goto cleanup1;
1542 if ((ioc->buf_size < 1) &&
1543 (ioc->Request.Type.Direction != XFER_NONE)) {
1544 status = -EINVAL;
1545 goto cleanup1;
1547 /* Check kmalloc limits using all SGs */
1548 if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1549 status = -EINVAL;
1550 goto cleanup1;
1552 if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1553 status = -EINVAL;
1554 goto cleanup1;
1556 buff = kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1557 if (!buff) {
1558 status = -ENOMEM;
1559 goto cleanup1;
1561 buff_size = kmalloc(MAXSGENTRIES * sizeof(int), GFP_KERNEL);
1562 if (!buff_size) {
1563 status = -ENOMEM;
1564 goto cleanup1;
1566 left = ioc->buf_size;
1567 data_ptr = ioc->buf;
1568 while (left) {
1569 sz = (left > ioc->malloc_size) ? ioc->malloc_size : left;
1570 buff_size[sg_used] = sz;
1571 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1572 if (buff[sg_used] == NULL) {
1573 status = -ENOMEM;
1574 goto cleanup1;
1576 if (ioc->Request.Type.Direction == XFER_WRITE) {
1577 if (copy_from_user(buff[sg_used], data_ptr, sz)) {
1578 status = -EFAULT;
1579 goto cleanup1;
1581 } else {
1582 memset(buff[sg_used], 0, sz);
1584 left -= sz;
1585 data_ptr += sz;
1586 sg_used++;
1588 c = cmd_special_alloc(h);
1589 if (!c) {
1590 status = -ENOMEM;
1591 goto cleanup1;
1593 c->cmd_type = CMD_IOCTL_PEND;
1594 c->Header.ReplyQueue = 0;
1595 c->Header.SGList = sg_used;
1596 c->Header.SGTotal = sg_used;
1597 c->Header.LUN = ioc->LUN_info;
1598 c->Header.Tag.lower = c->busaddr;
1600 c->Request = ioc->Request;
1601 for (i = 0; i < sg_used; i++) {
1602 temp64.val = pci_map_single(h->pdev, buff[i], buff_size[i],
1603 PCI_DMA_BIDIRECTIONAL);
1604 c->SG[i].Addr.lower = temp64.val32.lower;
1605 c->SG[i].Addr.upper = temp64.val32.upper;
1606 c->SG[i].Len = buff_size[i];
1607 c->SG[i].Ext = 0; /* we are not chaining */
1609 c->waiting = &wait;
1610 enqueue_cmd_and_start_io(h, c);
1611 wait_for_completion(&wait);
1612 /* unlock the buffers from DMA */
1613 for (i = 0; i < sg_used; i++) {
1614 temp64.val32.lower = c->SG[i].Addr.lower;
1615 temp64.val32.upper = c->SG[i].Addr.upper;
1616 pci_unmap_single(h->pdev,
1617 (dma_addr_t) temp64.val, buff_size[i],
1618 PCI_DMA_BIDIRECTIONAL);
1620 check_ioctl_unit_attention(h, c);
1621 /* Copy the error information out */
1622 ioc->error_info = *(c->err_info);
1623 if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1624 cmd_special_free(h, c);
1625 status = -EFAULT;
1626 goto cleanup1;
1628 if (ioc->Request.Type.Direction == XFER_READ) {
1629 /* Copy the data out of the buffer we created */
1630 BYTE __user *ptr = ioc->buf;
1631 for (i = 0; i < sg_used; i++) {
1632 if (copy_to_user(ptr, buff[i], buff_size[i])) {
1633 cmd_special_free(h, c);
1634 status = -EFAULT;
1635 goto cleanup1;
1637 ptr += buff_size[i];
1640 cmd_special_free(h, c);
1641 status = 0;
1642 cleanup1:
1643 if (buff) {
1644 for (i = 0; i < sg_used; i++)
1645 kfree(buff[i]);
1646 kfree(buff);
1648 kfree(buff_size);
1649 kfree(ioc);
1650 return status;
1653 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1654 unsigned int cmd, unsigned long arg)
1656 struct gendisk *disk = bdev->bd_disk;
1657 ctlr_info_t *h = get_host(disk);
1658 void __user *argp = (void __user *)arg;
1660 dev_dbg(&h->pdev->dev, "cciss_ioctl: Called with cmd=%x %lx\n",
1661 cmd, arg);
1662 switch (cmd) {
1663 case CCISS_GETPCIINFO:
1664 return cciss_getpciinfo(h, argp);
1665 case CCISS_GETINTINFO:
1666 return cciss_getintinfo(h, argp);
1667 case CCISS_SETINTINFO:
1668 return cciss_setintinfo(h, argp);
1669 case CCISS_GETNODENAME:
1670 return cciss_getnodename(h, argp);
1671 case CCISS_SETNODENAME:
1672 return cciss_setnodename(h, argp);
1673 case CCISS_GETHEARTBEAT:
1674 return cciss_getheartbeat(h, argp);
1675 case CCISS_GETBUSTYPES:
1676 return cciss_getbustypes(h, argp);
1677 case CCISS_GETFIRMVER:
1678 return cciss_getfirmver(h, argp);
1679 case CCISS_GETDRIVVER:
1680 return cciss_getdrivver(h, argp);
1681 case CCISS_DEREGDISK:
1682 case CCISS_REGNEWD:
1683 case CCISS_REVALIDVOLS:
1684 return rebuild_lun_table(h, 0, 1);
1685 case CCISS_GETLUNINFO:
1686 return cciss_getluninfo(h, disk, argp);
1687 case CCISS_PASSTHRU:
1688 return cciss_passthru(h, argp);
1689 case CCISS_BIG_PASSTHRU:
1690 return cciss_bigpassthru(h, argp);
1692 /* scsi_cmd_ioctl handles these, below, though some are not */
1693 /* very meaningful for cciss. SG_IO is the main one people want. */
1695 case SG_GET_VERSION_NUM:
1696 case SG_SET_TIMEOUT:
1697 case SG_GET_TIMEOUT:
1698 case SG_GET_RESERVED_SIZE:
1699 case SG_SET_RESERVED_SIZE:
1700 case SG_EMULATED_HOST:
1701 case SG_IO:
1702 case SCSI_IOCTL_SEND_COMMAND:
1703 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
1705 /* scsi_cmd_ioctl would normally handle these, below, but */
1706 /* they aren't a good fit for cciss, as CD-ROMs are */
1707 /* not supported, and we don't have any bus/target/lun */
1708 /* which we present to the kernel. */
1710 case CDROM_SEND_PACKET:
1711 case CDROMCLOSETRAY:
1712 case CDROMEJECT:
1713 case SCSI_IOCTL_GET_IDLUN:
1714 case SCSI_IOCTL_GET_BUS_NUMBER:
1715 default:
1716 return -ENOTTY;
1720 static void cciss_check_queues(ctlr_info_t *h)
1722 int start_queue = h->next_to_run;
1723 int i;
1725 /* check to see if we have maxed out the number of commands that can
1726 * be placed on the queue. If so then exit. We do this check here
1727 * in case the interrupt we serviced was from an ioctl and did not
1728 * free any new commands.
1730 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1731 return;
1733 /* We have room on the queue for more commands. Now we need to queue
1734 * them up. We will also keep track of the next queue to run so
1735 * that every queue gets a chance to be started first.
1737 for (i = 0; i < h->highest_lun + 1; i++) {
1738 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1739 /* make sure the disk has been added and the drive is real
1740 * because this can be called from the middle of init_one.
1742 if (!h->drv[curr_queue])
1743 continue;
1744 if (!(h->drv[curr_queue]->queue) ||
1745 !(h->drv[curr_queue]->heads))
1746 continue;
1747 blk_start_queue(h->gendisk[curr_queue]->queue);
1749 /* check to see if we have maxed out the number of commands
1750 * that can be placed on the queue.
1752 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1753 if (curr_queue == start_queue) {
1754 h->next_to_run =
1755 (start_queue + 1) % (h->highest_lun + 1);
1756 break;
1757 } else {
1758 h->next_to_run = curr_queue;
1759 break;
1765 static void cciss_softirq_done(struct request *rq)
1767 CommandList_struct *c = rq->completion_data;
1768 ctlr_info_t *h = hba[c->ctlr];
1769 SGDescriptor_struct *curr_sg = c->SG;
1770 u64bit temp64;
1771 unsigned long flags;
1772 int i, ddir;
1773 int sg_index = 0;
1775 if (c->Request.Type.Direction == XFER_READ)
1776 ddir = PCI_DMA_FROMDEVICE;
1777 else
1778 ddir = PCI_DMA_TODEVICE;
1780 /* command did not need to be retried */
1781 /* unmap the DMA mapping for all the scatter gather elements */
1782 for (i = 0; i < c->Header.SGList; i++) {
1783 if (curr_sg[sg_index].Ext == CCISS_SG_CHAIN) {
1784 cciss_unmap_sg_chain_block(h, c);
1785 /* Point to the next block */
1786 curr_sg = h->cmd_sg_list[c->cmdindex];
1787 sg_index = 0;
1789 temp64.val32.lower = curr_sg[sg_index].Addr.lower;
1790 temp64.val32.upper = curr_sg[sg_index].Addr.upper;
1791 pci_unmap_page(h->pdev, temp64.val, curr_sg[sg_index].Len,
1792 ddir);
1793 ++sg_index;
1796 dev_dbg(&h->pdev->dev, "Done with %p\n", rq);
1798 /* set the residual count for pc requests */
1799 if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
1800 rq->resid_len = c->err_info->ResidualCnt;
1802 blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1804 spin_lock_irqsave(&h->lock, flags);
1805 cmd_free(h, c);
1806 cciss_check_queues(h);
1807 spin_unlock_irqrestore(&h->lock, flags);
1810 static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1811 unsigned char scsi3addr[], uint32_t log_unit)
1813 memcpy(scsi3addr, h->drv[log_unit]->LunID,
1814 sizeof(h->drv[log_unit]->LunID));
1817 /* This function gets the SCSI vendor, model, and revision of a logical drive
1818 * via the inquiry page 0. Model, vendor, and rev are set to empty strings if
1819 * they cannot be read.
1821 static void cciss_get_device_descr(ctlr_info_t *h, int logvol,
1822 char *vendor, char *model, char *rev)
1824 int rc;
1825 InquiryData_struct *inq_buf;
1826 unsigned char scsi3addr[8];
1828 *vendor = '\0';
1829 *model = '\0';
1830 *rev = '\0';
1832 inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1833 if (!inq_buf)
1834 return;
1836 log_unit_to_scsi3addr(h, scsi3addr, logvol);
1837 rc = sendcmd_withirq(h, CISS_INQUIRY, inq_buf, sizeof(*inq_buf), 0,
1838 scsi3addr, TYPE_CMD);
1839 if (rc == IO_OK) {
1840 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1841 vendor[VENDOR_LEN] = '\0';
1842 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1843 model[MODEL_LEN] = '\0';
1844 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1845 rev[REV_LEN] = '\0';
1848 kfree(inq_buf);
1849 return;
1852 /* This function gets the serial number of a logical drive via
1853 * inquiry page 0x83. Serial no. is 16 bytes. If the serial
1854 * number cannot be had, for whatever reason, 16 bytes of 0xff
1855 * are returned instead.
1857 static void cciss_get_serial_no(ctlr_info_t *h, int logvol,
1858 unsigned char *serial_no, int buflen)
1860 #define PAGE_83_INQ_BYTES 64
1861 int rc;
1862 unsigned char *buf;
1863 unsigned char scsi3addr[8];
1865 if (buflen > 16)
1866 buflen = 16;
1867 memset(serial_no, 0xff, buflen);
1868 buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1869 if (!buf)
1870 return;
1871 memset(serial_no, 0, buflen);
1872 log_unit_to_scsi3addr(h, scsi3addr, logvol);
1873 rc = sendcmd_withirq(h, CISS_INQUIRY, buf,
1874 PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1875 if (rc == IO_OK)
1876 memcpy(serial_no, &buf[8], buflen);
1877 kfree(buf);
1878 return;
1882 * cciss_add_disk sets up the block device queue for a logical drive
1884 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1885 int drv_index)
1887 disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1888 if (!disk->queue)
1889 goto init_queue_failure;
1890 sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1891 disk->major = h->major;
1892 disk->first_minor = drv_index << NWD_SHIFT;
1893 disk->fops = &cciss_fops;
1894 if (cciss_create_ld_sysfs_entry(h, drv_index))
1895 goto cleanup_queue;
1896 disk->private_data = h->drv[drv_index];
1897 disk->driverfs_dev = &h->drv[drv_index]->dev;
1899 /* Set up queue information */
1900 blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1902 /* This is a hardware imposed limit. */
1903 blk_queue_max_segments(disk->queue, h->maxsgentries);
1905 blk_queue_max_hw_sectors(disk->queue, h->cciss_max_sectors);
1907 blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1909 disk->queue->queuedata = h;
1911 blk_queue_logical_block_size(disk->queue,
1912 h->drv[drv_index]->block_size);
1914 /* Make sure all queue data is written out before */
1915 /* setting h->drv[drv_index]->queue, as setting this */
1916 /* allows the interrupt handler to start the queue */
1917 wmb();
1918 h->drv[drv_index]->queue = disk->queue;
1919 add_disk(disk);
1920 return 0;
1922 cleanup_queue:
1923 blk_cleanup_queue(disk->queue);
1924 disk->queue = NULL;
1925 init_queue_failure:
1926 return -1;
1929 /* This function will check the usage_count of the drive to be updated/added.
1930 * If the usage_count is zero and it is a heretofore unknown drive, or,
1931 * the drive's capacity, geometry, or serial number has changed,
1932 * then the drive information will be updated and the disk will be
1933 * re-registered with the kernel. If these conditions don't hold,
1934 * then it will be left alone for the next reboot. The exception to this
1935 * is disk 0 which will always be left registered with the kernel since it
1936 * is also the controller node. Any changes to disk 0 will show up on
1937 * the next reboot.
1939 static void cciss_update_drive_info(ctlr_info_t *h, int drv_index,
1940 int first_time, int via_ioctl)
1942 struct gendisk *disk;
1943 InquiryData_struct *inq_buff = NULL;
1944 unsigned int block_size;
1945 sector_t total_size;
1946 unsigned long flags = 0;
1947 int ret = 0;
1948 drive_info_struct *drvinfo;
1950 /* Get information about the disk and modify the driver structure */
1951 inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1952 drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
1953 if (inq_buff == NULL || drvinfo == NULL)
1954 goto mem_msg;
1956 /* testing to see if 16-byte CDBs are already being used */
1957 if (h->cciss_read == CCISS_READ_16) {
1958 cciss_read_capacity_16(h, drv_index,
1959 &total_size, &block_size);
1961 } else {
1962 cciss_read_capacity(h, drv_index, &total_size, &block_size);
1963 /* if read_capacity returns all F's this volume is >2TB */
1964 /* in size so we switch to 16-byte CDB's for all */
1965 /* read/write ops */
1966 if (total_size == 0xFFFFFFFFULL) {
1967 cciss_read_capacity_16(h, drv_index,
1968 &total_size, &block_size);
1969 h->cciss_read = CCISS_READ_16;
1970 h->cciss_write = CCISS_WRITE_16;
1971 } else {
1972 h->cciss_read = CCISS_READ_10;
1973 h->cciss_write = CCISS_WRITE_10;
1977 cciss_geometry_inquiry(h, drv_index, total_size, block_size,
1978 inq_buff, drvinfo);
1979 drvinfo->block_size = block_size;
1980 drvinfo->nr_blocks = total_size + 1;
1982 cciss_get_device_descr(h, drv_index, drvinfo->vendor,
1983 drvinfo->model, drvinfo->rev);
1984 cciss_get_serial_no(h, drv_index, drvinfo->serial_no,
1985 sizeof(drvinfo->serial_no));
1986 /* Save the lunid in case we deregister the disk, below. */
1987 memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
1988 sizeof(drvinfo->LunID));
1990 /* Is it the same disk we already know, and nothing's changed? */
1991 if (h->drv[drv_index]->raid_level != -1 &&
1992 ((memcmp(drvinfo->serial_no,
1993 h->drv[drv_index]->serial_no, 16) == 0) &&
1994 drvinfo->block_size == h->drv[drv_index]->block_size &&
1995 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
1996 drvinfo->heads == h->drv[drv_index]->heads &&
1997 drvinfo->sectors == h->drv[drv_index]->sectors &&
1998 drvinfo->cylinders == h->drv[drv_index]->cylinders))
1999 /* The disk is unchanged, nothing to update */
2000 goto freeret;
2002 /* If we get here it's not the same disk, or something's changed,
2003 * so we need to * deregister it, and re-register it, if it's not
2004 * in use.
2005 * If the disk already exists then deregister it before proceeding
2006 * (unless it's the first disk (for the controller node).
2008 if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
2009 dev_warn(&h->pdev->dev, "disk %d has changed.\n", drv_index);
2010 spin_lock_irqsave(&h->lock, flags);
2011 h->drv[drv_index]->busy_configuring = 1;
2012 spin_unlock_irqrestore(&h->lock, flags);
2014 /* deregister_disk sets h->drv[drv_index]->queue = NULL
2015 * which keeps the interrupt handler from starting
2016 * the queue.
2018 ret = deregister_disk(h, drv_index, 0, via_ioctl);
2021 /* If the disk is in use return */
2022 if (ret)
2023 goto freeret;
2025 /* Save the new information from cciss_geometry_inquiry
2026 * and serial number inquiry. If the disk was deregistered
2027 * above, then h->drv[drv_index] will be NULL.
2029 if (h->drv[drv_index] == NULL) {
2030 drvinfo->device_initialized = 0;
2031 h->drv[drv_index] = drvinfo;
2032 drvinfo = NULL; /* so it won't be freed below. */
2033 } else {
2034 /* special case for cxd0 */
2035 h->drv[drv_index]->block_size = drvinfo->block_size;
2036 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
2037 h->drv[drv_index]->heads = drvinfo->heads;
2038 h->drv[drv_index]->sectors = drvinfo->sectors;
2039 h->drv[drv_index]->cylinders = drvinfo->cylinders;
2040 h->drv[drv_index]->raid_level = drvinfo->raid_level;
2041 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
2042 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
2043 VENDOR_LEN + 1);
2044 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
2045 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
2048 ++h->num_luns;
2049 disk = h->gendisk[drv_index];
2050 set_capacity(disk, h->drv[drv_index]->nr_blocks);
2052 /* If it's not disk 0 (drv_index != 0)
2053 * or if it was disk 0, but there was previously
2054 * no actual corresponding configured logical drive
2055 * (raid_leve == -1) then we want to update the
2056 * logical drive's information.
2058 if (drv_index || first_time) {
2059 if (cciss_add_disk(h, disk, drv_index) != 0) {
2060 cciss_free_gendisk(h, drv_index);
2061 cciss_free_drive_info(h, drv_index);
2062 dev_warn(&h->pdev->dev, "could not update disk %d\n",
2063 drv_index);
2064 --h->num_luns;
2068 freeret:
2069 kfree(inq_buff);
2070 kfree(drvinfo);
2071 return;
2072 mem_msg:
2073 dev_err(&h->pdev->dev, "out of memory\n");
2074 goto freeret;
2077 /* This function will find the first index of the controllers drive array
2078 * that has a null drv pointer and allocate the drive info struct and
2079 * will return that index This is where new drives will be added.
2080 * If the index to be returned is greater than the highest_lun index for
2081 * the controller then highest_lun is set * to this new index.
2082 * If there are no available indexes or if tha allocation fails, then -1
2083 * is returned. * "controller_node" is used to know if this is a real
2084 * logical drive, or just the controller node, which determines if this
2085 * counts towards highest_lun.
2087 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
2089 int i;
2090 drive_info_struct *drv;
2092 /* Search for an empty slot for our drive info */
2093 for (i = 0; i < CISS_MAX_LUN; i++) {
2095 /* if not cxd0 case, and it's occupied, skip it. */
2096 if (h->drv[i] && i != 0)
2097 continue;
2099 * If it's cxd0 case, and drv is alloc'ed already, and a
2100 * disk is configured there, skip it.
2102 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2103 continue;
2106 * We've found an empty slot. Update highest_lun
2107 * provided this isn't just the fake cxd0 controller node.
2109 if (i > h->highest_lun && !controller_node)
2110 h->highest_lun = i;
2112 /* If adding a real disk at cxd0, and it's already alloc'ed */
2113 if (i == 0 && h->drv[i] != NULL)
2114 return i;
2117 * Found an empty slot, not already alloc'ed. Allocate it.
2118 * Mark it with raid_level == -1, so we know it's new later on.
2120 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2121 if (!drv)
2122 return -1;
2123 drv->raid_level = -1; /* so we know it's new */
2124 h->drv[i] = drv;
2125 return i;
2127 return -1;
2130 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2132 kfree(h->drv[drv_index]);
2133 h->drv[drv_index] = NULL;
2136 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2138 put_disk(h->gendisk[drv_index]);
2139 h->gendisk[drv_index] = NULL;
2142 /* cciss_add_gendisk finds a free hba[]->drv structure
2143 * and allocates a gendisk if needed, and sets the lunid
2144 * in the drvinfo structure. It returns the index into
2145 * the ->drv[] array, or -1 if none are free.
2146 * is_controller_node indicates whether highest_lun should
2147 * count this disk, or if it's only being added to provide
2148 * a means to talk to the controller in case no logical
2149 * drives have yet been configured.
2151 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2152 int controller_node)
2154 int drv_index;
2156 drv_index = cciss_alloc_drive_info(h, controller_node);
2157 if (drv_index == -1)
2158 return -1;
2160 /*Check if the gendisk needs to be allocated */
2161 if (!h->gendisk[drv_index]) {
2162 h->gendisk[drv_index] =
2163 alloc_disk(1 << NWD_SHIFT);
2164 if (!h->gendisk[drv_index]) {
2165 dev_err(&h->pdev->dev,
2166 "could not allocate a new disk %d\n",
2167 drv_index);
2168 goto err_free_drive_info;
2171 memcpy(h->drv[drv_index]->LunID, lunid,
2172 sizeof(h->drv[drv_index]->LunID));
2173 if (cciss_create_ld_sysfs_entry(h, drv_index))
2174 goto err_free_disk;
2175 /* Don't need to mark this busy because nobody */
2176 /* else knows about this disk yet to contend */
2177 /* for access to it. */
2178 h->drv[drv_index]->busy_configuring = 0;
2179 wmb();
2180 return drv_index;
2182 err_free_disk:
2183 cciss_free_gendisk(h, drv_index);
2184 err_free_drive_info:
2185 cciss_free_drive_info(h, drv_index);
2186 return -1;
2189 /* This is for the special case of a controller which
2190 * has no logical drives. In this case, we still need
2191 * to register a disk so the controller can be accessed
2192 * by the Array Config Utility.
2194 static void cciss_add_controller_node(ctlr_info_t *h)
2196 struct gendisk *disk;
2197 int drv_index;
2199 if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2200 return;
2202 drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2203 if (drv_index == -1)
2204 goto error;
2205 h->drv[drv_index]->block_size = 512;
2206 h->drv[drv_index]->nr_blocks = 0;
2207 h->drv[drv_index]->heads = 0;
2208 h->drv[drv_index]->sectors = 0;
2209 h->drv[drv_index]->cylinders = 0;
2210 h->drv[drv_index]->raid_level = -1;
2211 memset(h->drv[drv_index]->serial_no, 0, 16);
2212 disk = h->gendisk[drv_index];
2213 if (cciss_add_disk(h, disk, drv_index) == 0)
2214 return;
2215 cciss_free_gendisk(h, drv_index);
2216 cciss_free_drive_info(h, drv_index);
2217 error:
2218 dev_warn(&h->pdev->dev, "could not add disk 0.\n");
2219 return;
2222 /* This function will add and remove logical drives from the Logical
2223 * drive array of the controller and maintain persistency of ordering
2224 * so that mount points are preserved until the next reboot. This allows
2225 * for the removal of logical drives in the middle of the drive array
2226 * without a re-ordering of those drives.
2227 * INPUT
2228 * h = The controller to perform the operations on
2230 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2231 int via_ioctl)
2233 int num_luns;
2234 ReportLunData_struct *ld_buff = NULL;
2235 int return_code;
2236 int listlength = 0;
2237 int i;
2238 int drv_found;
2239 int drv_index = 0;
2240 unsigned char lunid[8] = CTLR_LUNID;
2241 unsigned long flags;
2243 if (!capable(CAP_SYS_RAWIO))
2244 return -EPERM;
2246 /* Set busy_configuring flag for this operation */
2247 spin_lock_irqsave(&h->lock, flags);
2248 if (h->busy_configuring) {
2249 spin_unlock_irqrestore(&h->lock, flags);
2250 return -EBUSY;
2252 h->busy_configuring = 1;
2253 spin_unlock_irqrestore(&h->lock, flags);
2255 ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2256 if (ld_buff == NULL)
2257 goto mem_msg;
2259 return_code = sendcmd_withirq(h, CISS_REPORT_LOG, ld_buff,
2260 sizeof(ReportLunData_struct),
2261 0, CTLR_LUNID, TYPE_CMD);
2263 if (return_code == IO_OK)
2264 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2265 else { /* reading number of logical volumes failed */
2266 dev_warn(&h->pdev->dev,
2267 "report logical volume command failed\n");
2268 listlength = 0;
2269 goto freeret;
2272 num_luns = listlength / 8; /* 8 bytes per entry */
2273 if (num_luns > CISS_MAX_LUN) {
2274 num_luns = CISS_MAX_LUN;
2275 dev_warn(&h->pdev->dev, "more luns configured"
2276 " on controller than can be handled by"
2277 " this driver.\n");
2280 if (num_luns == 0)
2281 cciss_add_controller_node(h);
2283 /* Compare controller drive array to driver's drive array
2284 * to see if any drives are missing on the controller due
2285 * to action of Array Config Utility (user deletes drive)
2286 * and deregister logical drives which have disappeared.
2288 for (i = 0; i <= h->highest_lun; i++) {
2289 int j;
2290 drv_found = 0;
2292 /* skip holes in the array from already deleted drives */
2293 if (h->drv[i] == NULL)
2294 continue;
2296 for (j = 0; j < num_luns; j++) {
2297 memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2298 if (memcmp(h->drv[i]->LunID, lunid,
2299 sizeof(lunid)) == 0) {
2300 drv_found = 1;
2301 break;
2304 if (!drv_found) {
2305 /* Deregister it from the OS, it's gone. */
2306 spin_lock_irqsave(&h->lock, flags);
2307 h->drv[i]->busy_configuring = 1;
2308 spin_unlock_irqrestore(&h->lock, flags);
2309 return_code = deregister_disk(h, i, 1, via_ioctl);
2310 if (h->drv[i] != NULL)
2311 h->drv[i]->busy_configuring = 0;
2315 /* Compare controller drive array to driver's drive array.
2316 * Check for updates in the drive information and any new drives
2317 * on the controller due to ACU adding logical drives, or changing
2318 * a logical drive's size, etc. Reregister any new/changed drives
2320 for (i = 0; i < num_luns; i++) {
2321 int j;
2323 drv_found = 0;
2325 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2326 /* Find if the LUN is already in the drive array
2327 * of the driver. If so then update its info
2328 * if not in use. If it does not exist then find
2329 * the first free index and add it.
2331 for (j = 0; j <= h->highest_lun; j++) {
2332 if (h->drv[j] != NULL &&
2333 memcmp(h->drv[j]->LunID, lunid,
2334 sizeof(h->drv[j]->LunID)) == 0) {
2335 drv_index = j;
2336 drv_found = 1;
2337 break;
2341 /* check if the drive was found already in the array */
2342 if (!drv_found) {
2343 drv_index = cciss_add_gendisk(h, lunid, 0);
2344 if (drv_index == -1)
2345 goto freeret;
2347 cciss_update_drive_info(h, drv_index, first_time, via_ioctl);
2348 } /* end for */
2350 freeret:
2351 kfree(ld_buff);
2352 h->busy_configuring = 0;
2353 /* We return -1 here to tell the ACU that we have registered/updated
2354 * all of the drives that we can and to keep it from calling us
2355 * additional times.
2357 return -1;
2358 mem_msg:
2359 dev_err(&h->pdev->dev, "out of memory\n");
2360 h->busy_configuring = 0;
2361 goto freeret;
2364 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2366 /* zero out the disk size info */
2367 drive_info->nr_blocks = 0;
2368 drive_info->block_size = 0;
2369 drive_info->heads = 0;
2370 drive_info->sectors = 0;
2371 drive_info->cylinders = 0;
2372 drive_info->raid_level = -1;
2373 memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2374 memset(drive_info->model, 0, sizeof(drive_info->model));
2375 memset(drive_info->rev, 0, sizeof(drive_info->rev));
2376 memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2378 * don't clear the LUNID though, we need to remember which
2379 * one this one is.
2383 /* This function will deregister the disk and it's queue from the
2384 * kernel. It must be called with the controller lock held and the
2385 * drv structures busy_configuring flag set. It's parameters are:
2387 * disk = This is the disk to be deregistered
2388 * drv = This is the drive_info_struct associated with the disk to be
2389 * deregistered. It contains information about the disk used
2390 * by the driver.
2391 * clear_all = This flag determines whether or not the disk information
2392 * is going to be completely cleared out and the highest_lun
2393 * reset. Sometimes we want to clear out information about
2394 * the disk in preparation for re-adding it. In this case
2395 * the highest_lun should be left unchanged and the LunID
2396 * should not be cleared.
2397 * via_ioctl
2398 * This indicates whether we've reached this path via ioctl.
2399 * This affects the maximum usage count allowed for c0d0 to be messed with.
2400 * If this path is reached via ioctl(), then the max_usage_count will
2401 * be 1, as the process calling ioctl() has got to have the device open.
2402 * If we get here via sysfs, then the max usage count will be zero.
2404 static int deregister_disk(ctlr_info_t *h, int drv_index,
2405 int clear_all, int via_ioctl)
2407 int i;
2408 struct gendisk *disk;
2409 drive_info_struct *drv;
2410 int recalculate_highest_lun;
2412 if (!capable(CAP_SYS_RAWIO))
2413 return -EPERM;
2415 drv = h->drv[drv_index];
2416 disk = h->gendisk[drv_index];
2418 /* make sure logical volume is NOT is use */
2419 if (clear_all || (h->gendisk[0] == disk)) {
2420 if (drv->usage_count > via_ioctl)
2421 return -EBUSY;
2422 } else if (drv->usage_count > 0)
2423 return -EBUSY;
2425 recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2427 /* invalidate the devices and deregister the disk. If it is disk
2428 * zero do not deregister it but just zero out it's values. This
2429 * allows us to delete disk zero but keep the controller registered.
2431 if (h->gendisk[0] != disk) {
2432 struct request_queue *q = disk->queue;
2433 if (disk->flags & GENHD_FL_UP) {
2434 cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2435 del_gendisk(disk);
2437 if (q)
2438 blk_cleanup_queue(q);
2439 /* If clear_all is set then we are deleting the logical
2440 * drive, not just refreshing its info. For drives
2441 * other than disk 0 we will call put_disk. We do not
2442 * do this for disk 0 as we need it to be able to
2443 * configure the controller.
2445 if (clear_all){
2446 /* This isn't pretty, but we need to find the
2447 * disk in our array and NULL our the pointer.
2448 * This is so that we will call alloc_disk if
2449 * this index is used again later.
2451 for (i=0; i < CISS_MAX_LUN; i++){
2452 if (h->gendisk[i] == disk) {
2453 h->gendisk[i] = NULL;
2454 break;
2457 put_disk(disk);
2459 } else {
2460 set_capacity(disk, 0);
2461 cciss_clear_drive_info(drv);
2464 --h->num_luns;
2466 /* if it was the last disk, find the new hightest lun */
2467 if (clear_all && recalculate_highest_lun) {
2468 int newhighest = -1;
2469 for (i = 0; i <= h->highest_lun; i++) {
2470 /* if the disk has size > 0, it is available */
2471 if (h->drv[i] && h->drv[i]->heads)
2472 newhighest = i;
2474 h->highest_lun = newhighest;
2476 return 0;
2479 static int fill_cmd(ctlr_info_t *h, CommandList_struct *c, __u8 cmd, void *buff,
2480 size_t size, __u8 page_code, unsigned char *scsi3addr,
2481 int cmd_type)
2483 u64bit buff_dma_handle;
2484 int status = IO_OK;
2486 c->cmd_type = CMD_IOCTL_PEND;
2487 c->Header.ReplyQueue = 0;
2488 if (buff != NULL) {
2489 c->Header.SGList = 1;
2490 c->Header.SGTotal = 1;
2491 } else {
2492 c->Header.SGList = 0;
2493 c->Header.SGTotal = 0;
2495 c->Header.Tag.lower = c->busaddr;
2496 memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2498 c->Request.Type.Type = cmd_type;
2499 if (cmd_type == TYPE_CMD) {
2500 switch (cmd) {
2501 case CISS_INQUIRY:
2502 /* are we trying to read a vital product page */
2503 if (page_code != 0) {
2504 c->Request.CDB[1] = 0x01;
2505 c->Request.CDB[2] = page_code;
2507 c->Request.CDBLen = 6;
2508 c->Request.Type.Attribute = ATTR_SIMPLE;
2509 c->Request.Type.Direction = XFER_READ;
2510 c->Request.Timeout = 0;
2511 c->Request.CDB[0] = CISS_INQUIRY;
2512 c->Request.CDB[4] = size & 0xFF;
2513 break;
2514 case CISS_REPORT_LOG:
2515 case CISS_REPORT_PHYS:
2516 /* Talking to controller so It's a physical command
2517 mode = 00 target = 0. Nothing to write.
2519 c->Request.CDBLen = 12;
2520 c->Request.Type.Attribute = ATTR_SIMPLE;
2521 c->Request.Type.Direction = XFER_READ;
2522 c->Request.Timeout = 0;
2523 c->Request.CDB[0] = cmd;
2524 c->Request.CDB[6] = (size >> 24) & 0xFF; /* MSB */
2525 c->Request.CDB[7] = (size >> 16) & 0xFF;
2526 c->Request.CDB[8] = (size >> 8) & 0xFF;
2527 c->Request.CDB[9] = size & 0xFF;
2528 break;
2530 case CCISS_READ_CAPACITY:
2531 c->Request.CDBLen = 10;
2532 c->Request.Type.Attribute = ATTR_SIMPLE;
2533 c->Request.Type.Direction = XFER_READ;
2534 c->Request.Timeout = 0;
2535 c->Request.CDB[0] = cmd;
2536 break;
2537 case CCISS_READ_CAPACITY_16:
2538 c->Request.CDBLen = 16;
2539 c->Request.Type.Attribute = ATTR_SIMPLE;
2540 c->Request.Type.Direction = XFER_READ;
2541 c->Request.Timeout = 0;
2542 c->Request.CDB[0] = cmd;
2543 c->Request.CDB[1] = 0x10;
2544 c->Request.CDB[10] = (size >> 24) & 0xFF;
2545 c->Request.CDB[11] = (size >> 16) & 0xFF;
2546 c->Request.CDB[12] = (size >> 8) & 0xFF;
2547 c->Request.CDB[13] = size & 0xFF;
2548 c->Request.Timeout = 0;
2549 c->Request.CDB[0] = cmd;
2550 break;
2551 case CCISS_CACHE_FLUSH:
2552 c->Request.CDBLen = 12;
2553 c->Request.Type.Attribute = ATTR_SIMPLE;
2554 c->Request.Type.Direction = XFER_WRITE;
2555 c->Request.Timeout = 0;
2556 c->Request.CDB[0] = BMIC_WRITE;
2557 c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2558 break;
2559 case TEST_UNIT_READY:
2560 c->Request.CDBLen = 6;
2561 c->Request.Type.Attribute = ATTR_SIMPLE;
2562 c->Request.Type.Direction = XFER_NONE;
2563 c->Request.Timeout = 0;
2564 break;
2565 default:
2566 dev_warn(&h->pdev->dev, "Unknown Command 0x%c\n", cmd);
2567 return IO_ERROR;
2569 } else if (cmd_type == TYPE_MSG) {
2570 switch (cmd) {
2571 case 0: /* ABORT message */
2572 c->Request.CDBLen = 12;
2573 c->Request.Type.Attribute = ATTR_SIMPLE;
2574 c->Request.Type.Direction = XFER_WRITE;
2575 c->Request.Timeout = 0;
2576 c->Request.CDB[0] = cmd; /* abort */
2577 c->Request.CDB[1] = 0; /* abort a command */
2578 /* buff contains the tag of the command to abort */
2579 memcpy(&c->Request.CDB[4], buff, 8);
2580 break;
2581 case 1: /* RESET message */
2582 c->Request.CDBLen = 16;
2583 c->Request.Type.Attribute = ATTR_SIMPLE;
2584 c->Request.Type.Direction = XFER_NONE;
2585 c->Request.Timeout = 0;
2586 memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2587 c->Request.CDB[0] = cmd; /* reset */
2588 c->Request.CDB[1] = 0x03; /* reset a target */
2589 break;
2590 case 3: /* No-Op message */
2591 c->Request.CDBLen = 1;
2592 c->Request.Type.Attribute = ATTR_SIMPLE;
2593 c->Request.Type.Direction = XFER_WRITE;
2594 c->Request.Timeout = 0;
2595 c->Request.CDB[0] = cmd;
2596 break;
2597 default:
2598 dev_warn(&h->pdev->dev,
2599 "unknown message type %d\n", cmd);
2600 return IO_ERROR;
2602 } else {
2603 dev_warn(&h->pdev->dev, "unknown command type %d\n", cmd_type);
2604 return IO_ERROR;
2606 /* Fill in the scatter gather information */
2607 if (size > 0) {
2608 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2609 buff, size,
2610 PCI_DMA_BIDIRECTIONAL);
2611 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2612 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2613 c->SG[0].Len = size;
2614 c->SG[0].Ext = 0; /* we are not chaining */
2616 return status;
2619 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2621 switch (c->err_info->ScsiStatus) {
2622 case SAM_STAT_GOOD:
2623 return IO_OK;
2624 case SAM_STAT_CHECK_CONDITION:
2625 switch (0xf & c->err_info->SenseInfo[2]) {
2626 case 0: return IO_OK; /* no sense */
2627 case 1: return IO_OK; /* recovered error */
2628 default:
2629 if (check_for_unit_attention(h, c))
2630 return IO_NEEDS_RETRY;
2631 dev_warn(&h->pdev->dev, "cmd 0x%02x "
2632 "check condition, sense key = 0x%02x\n",
2633 c->Request.CDB[0], c->err_info->SenseInfo[2]);
2635 break;
2636 default:
2637 dev_warn(&h->pdev->dev, "cmd 0x%02x"
2638 "scsi status = 0x%02x\n",
2639 c->Request.CDB[0], c->err_info->ScsiStatus);
2640 break;
2642 return IO_ERROR;
2645 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2647 int return_status = IO_OK;
2649 if (c->err_info->CommandStatus == CMD_SUCCESS)
2650 return IO_OK;
2652 switch (c->err_info->CommandStatus) {
2653 case CMD_TARGET_STATUS:
2654 return_status = check_target_status(h, c);
2655 break;
2656 case CMD_DATA_UNDERRUN:
2657 case CMD_DATA_OVERRUN:
2658 /* expected for inquiry and report lun commands */
2659 break;
2660 case CMD_INVALID:
2661 dev_warn(&h->pdev->dev, "cmd 0x%02x is "
2662 "reported invalid\n", c->Request.CDB[0]);
2663 return_status = IO_ERROR;
2664 break;
2665 case CMD_PROTOCOL_ERR:
2666 dev_warn(&h->pdev->dev, "cmd 0x%02x has "
2667 "protocol error\n", c->Request.CDB[0]);
2668 return_status = IO_ERROR;
2669 break;
2670 case CMD_HARDWARE_ERR:
2671 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2672 " hardware error\n", c->Request.CDB[0]);
2673 return_status = IO_ERROR;
2674 break;
2675 case CMD_CONNECTION_LOST:
2676 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2677 "connection lost\n", c->Request.CDB[0]);
2678 return_status = IO_ERROR;
2679 break;
2680 case CMD_ABORTED:
2681 dev_warn(&h->pdev->dev, "cmd 0x%02x was "
2682 "aborted\n", c->Request.CDB[0]);
2683 return_status = IO_ERROR;
2684 break;
2685 case CMD_ABORT_FAILED:
2686 dev_warn(&h->pdev->dev, "cmd 0x%02x reports "
2687 "abort failed\n", c->Request.CDB[0]);
2688 return_status = IO_ERROR;
2689 break;
2690 case CMD_UNSOLICITED_ABORT:
2691 dev_warn(&h->pdev->dev, "unsolicited abort 0x%02x\n",
2692 c->Request.CDB[0]);
2693 return_status = IO_NEEDS_RETRY;
2694 break;
2695 default:
2696 dev_warn(&h->pdev->dev, "cmd 0x%02x returned "
2697 "unknown status %x\n", c->Request.CDB[0],
2698 c->err_info->CommandStatus);
2699 return_status = IO_ERROR;
2701 return return_status;
2704 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2705 int attempt_retry)
2707 DECLARE_COMPLETION_ONSTACK(wait);
2708 u64bit buff_dma_handle;
2709 int return_status = IO_OK;
2711 resend_cmd2:
2712 c->waiting = &wait;
2713 enqueue_cmd_and_start_io(h, c);
2715 wait_for_completion(&wait);
2717 if (c->err_info->CommandStatus == 0 || !attempt_retry)
2718 goto command_done;
2720 return_status = process_sendcmd_error(h, c);
2722 if (return_status == IO_NEEDS_RETRY &&
2723 c->retry_count < MAX_CMD_RETRIES) {
2724 dev_warn(&h->pdev->dev, "retrying 0x%02x\n",
2725 c->Request.CDB[0]);
2726 c->retry_count++;
2727 /* erase the old error information */
2728 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2729 return_status = IO_OK;
2730 INIT_COMPLETION(wait);
2731 goto resend_cmd2;
2734 command_done:
2735 /* unlock the buffers from DMA */
2736 buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2737 buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2738 pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2739 c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2740 return return_status;
2743 static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
2744 __u8 page_code, unsigned char scsi3addr[],
2745 int cmd_type)
2747 CommandList_struct *c;
2748 int return_status;
2750 c = cmd_special_alloc(h);
2751 if (!c)
2752 return -ENOMEM;
2753 return_status = fill_cmd(h, c, cmd, buff, size, page_code,
2754 scsi3addr, cmd_type);
2755 if (return_status == IO_OK)
2756 return_status = sendcmd_withirq_core(h, c, 1);
2758 cmd_special_free(h, c);
2759 return return_status;
2762 static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
2763 sector_t total_size,
2764 unsigned int block_size,
2765 InquiryData_struct *inq_buff,
2766 drive_info_struct *drv)
2768 int return_code;
2769 unsigned long t;
2770 unsigned char scsi3addr[8];
2772 memset(inq_buff, 0, sizeof(InquiryData_struct));
2773 log_unit_to_scsi3addr(h, scsi3addr, logvol);
2774 return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
2775 sizeof(*inq_buff), 0xC1, scsi3addr, TYPE_CMD);
2776 if (return_code == IO_OK) {
2777 if (inq_buff->data_byte[8] == 0xFF) {
2778 dev_warn(&h->pdev->dev,
2779 "reading geometry failed, volume "
2780 "does not support reading geometry\n");
2781 drv->heads = 255;
2782 drv->sectors = 32; /* Sectors per track */
2783 drv->cylinders = total_size + 1;
2784 drv->raid_level = RAID_UNKNOWN;
2785 } else {
2786 drv->heads = inq_buff->data_byte[6];
2787 drv->sectors = inq_buff->data_byte[7];
2788 drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2789 drv->cylinders += inq_buff->data_byte[5];
2790 drv->raid_level = inq_buff->data_byte[8];
2792 drv->block_size = block_size;
2793 drv->nr_blocks = total_size + 1;
2794 t = drv->heads * drv->sectors;
2795 if (t > 1) {
2796 sector_t real_size = total_size + 1;
2797 unsigned long rem = sector_div(real_size, t);
2798 if (rem)
2799 real_size++;
2800 drv->cylinders = real_size;
2802 } else { /* Get geometry failed */
2803 dev_warn(&h->pdev->dev, "reading geometry failed\n");
2807 static void
2808 cciss_read_capacity(ctlr_info_t *h, int logvol, sector_t *total_size,
2809 unsigned int *block_size)
2811 ReadCapdata_struct *buf;
2812 int return_code;
2813 unsigned char scsi3addr[8];
2815 buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2816 if (!buf) {
2817 dev_warn(&h->pdev->dev, "out of memory\n");
2818 return;
2821 log_unit_to_scsi3addr(h, scsi3addr, logvol);
2822 return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY, buf,
2823 sizeof(ReadCapdata_struct), 0, scsi3addr, TYPE_CMD);
2824 if (return_code == IO_OK) {
2825 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2826 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2827 } else { /* read capacity command failed */
2828 dev_warn(&h->pdev->dev, "read capacity failed\n");
2829 *total_size = 0;
2830 *block_size = BLOCK_SIZE;
2832 kfree(buf);
2835 static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
2836 sector_t *total_size, unsigned int *block_size)
2838 ReadCapdata_struct_16 *buf;
2839 int return_code;
2840 unsigned char scsi3addr[8];
2842 buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2843 if (!buf) {
2844 dev_warn(&h->pdev->dev, "out of memory\n");
2845 return;
2848 log_unit_to_scsi3addr(h, scsi3addr, logvol);
2849 return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY_16,
2850 buf, sizeof(ReadCapdata_struct_16),
2851 0, scsi3addr, TYPE_CMD);
2852 if (return_code == IO_OK) {
2853 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2854 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2855 } else { /* read capacity command failed */
2856 dev_warn(&h->pdev->dev, "read capacity failed\n");
2857 *total_size = 0;
2858 *block_size = BLOCK_SIZE;
2860 dev_info(&h->pdev->dev, " blocks= %llu block_size= %d\n",
2861 (unsigned long long)*total_size+1, *block_size);
2862 kfree(buf);
2865 static int cciss_revalidate(struct gendisk *disk)
2867 ctlr_info_t *h = get_host(disk);
2868 drive_info_struct *drv = get_drv(disk);
2869 int logvol;
2870 int FOUND = 0;
2871 unsigned int block_size;
2872 sector_t total_size;
2873 InquiryData_struct *inq_buff = NULL;
2875 for (logvol = 0; logvol <= h->highest_lun; logvol++) {
2876 if (!h->drv[logvol])
2877 continue;
2878 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2879 sizeof(drv->LunID)) == 0) {
2880 FOUND = 1;
2881 break;
2885 if (!FOUND)
2886 return 1;
2888 inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2889 if (inq_buff == NULL) {
2890 dev_warn(&h->pdev->dev, "out of memory\n");
2891 return 1;
2893 if (h->cciss_read == CCISS_READ_10) {
2894 cciss_read_capacity(h, logvol,
2895 &total_size, &block_size);
2896 } else {
2897 cciss_read_capacity_16(h, logvol,
2898 &total_size, &block_size);
2900 cciss_geometry_inquiry(h, logvol, total_size, block_size,
2901 inq_buff, drv);
2903 blk_queue_logical_block_size(drv->queue, drv->block_size);
2904 set_capacity(disk, drv->nr_blocks);
2906 kfree(inq_buff);
2907 return 0;
2911 * Map (physical) PCI mem into (virtual) kernel space
2913 static void __iomem *remap_pci_mem(ulong base, ulong size)
2915 ulong page_base = ((ulong) base) & PAGE_MASK;
2916 ulong page_offs = ((ulong) base) - page_base;
2917 void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2919 return page_remapped ? (page_remapped + page_offs) : NULL;
2923 * Takes jobs of the Q and sends them to the hardware, then puts it on
2924 * the Q to wait for completion.
2926 static void start_io(ctlr_info_t *h)
2928 CommandList_struct *c;
2930 while (!list_empty(&h->reqQ)) {
2931 c = list_entry(h->reqQ.next, CommandList_struct, list);
2932 /* can't do anything if fifo is full */
2933 if ((h->access.fifo_full(h))) {
2934 dev_warn(&h->pdev->dev, "fifo full\n");
2935 break;
2938 /* Get the first entry from the Request Q */
2939 removeQ(c);
2940 h->Qdepth--;
2942 /* Tell the controller execute command */
2943 h->access.submit_command(h, c);
2945 /* Put job onto the completed Q */
2946 addQ(&h->cmpQ, c);
2950 /* Assumes that h->lock is held. */
2951 /* Zeros out the error record and then resends the command back */
2952 /* to the controller */
2953 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2955 /* erase the old error information */
2956 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2958 /* add it to software queue and then send it to the controller */
2959 addQ(&h->reqQ, c);
2960 h->Qdepth++;
2961 if (h->Qdepth > h->maxQsinceinit)
2962 h->maxQsinceinit = h->Qdepth;
2964 start_io(h);
2967 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
2968 unsigned int msg_byte, unsigned int host_byte,
2969 unsigned int driver_byte)
2971 /* inverse of macros in scsi.h */
2972 return (scsi_status_byte & 0xff) |
2973 ((msg_byte & 0xff) << 8) |
2974 ((host_byte & 0xff) << 16) |
2975 ((driver_byte & 0xff) << 24);
2978 static inline int evaluate_target_status(ctlr_info_t *h,
2979 CommandList_struct *cmd, int *retry_cmd)
2981 unsigned char sense_key;
2982 unsigned char status_byte, msg_byte, host_byte, driver_byte;
2983 int error_value;
2985 *retry_cmd = 0;
2986 /* If we get in here, it means we got "target status", that is, scsi status */
2987 status_byte = cmd->err_info->ScsiStatus;
2988 driver_byte = DRIVER_OK;
2989 msg_byte = cmd->err_info->CommandStatus; /* correct? seems too device specific */
2991 if (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC)
2992 host_byte = DID_PASSTHROUGH;
2993 else
2994 host_byte = DID_OK;
2996 error_value = make_status_bytes(status_byte, msg_byte,
2997 host_byte, driver_byte);
2999 if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
3000 if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC)
3001 dev_warn(&h->pdev->dev, "cmd %p "
3002 "has SCSI Status 0x%x\n",
3003 cmd, cmd->err_info->ScsiStatus);
3004 return error_value;
3007 /* check the sense key */
3008 sense_key = 0xf & cmd->err_info->SenseInfo[2];
3009 /* no status or recovered error */
3010 if (((sense_key == 0x0) || (sense_key == 0x1)) &&
3011 (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC))
3012 error_value = 0;
3014 if (check_for_unit_attention(h, cmd)) {
3015 *retry_cmd = !(cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC);
3016 return 0;
3019 /* Not SG_IO or similar? */
3020 if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC) {
3021 if (error_value != 0)
3022 dev_warn(&h->pdev->dev, "cmd %p has CHECK CONDITION"
3023 " sense key = 0x%x\n", cmd, sense_key);
3024 return error_value;
3027 /* SG_IO or similar, copy sense data back */
3028 if (cmd->rq->sense) {
3029 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
3030 cmd->rq->sense_len = cmd->err_info->SenseLen;
3031 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
3032 cmd->rq->sense_len);
3033 } else
3034 cmd->rq->sense_len = 0;
3036 return error_value;
3039 /* checks the status of the job and calls complete buffers to mark all
3040 * buffers for the completed job. Note that this function does not need
3041 * to hold the hba/queue lock.
3043 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
3044 int timeout)
3046 int retry_cmd = 0;
3047 struct request *rq = cmd->rq;
3049 rq->errors = 0;
3051 if (timeout)
3052 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
3054 if (cmd->err_info->CommandStatus == 0) /* no error has occurred */
3055 goto after_error_processing;
3057 switch (cmd->err_info->CommandStatus) {
3058 case CMD_TARGET_STATUS:
3059 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
3060 break;
3061 case CMD_DATA_UNDERRUN:
3062 if (cmd->rq->cmd_type == REQ_TYPE_FS) {
3063 dev_warn(&h->pdev->dev, "cmd %p has"
3064 " completed with data underrun "
3065 "reported\n", cmd);
3066 cmd->rq->resid_len = cmd->err_info->ResidualCnt;
3068 break;
3069 case CMD_DATA_OVERRUN:
3070 if (cmd->rq->cmd_type == REQ_TYPE_FS)
3071 dev_warn(&h->pdev->dev, "cciss: cmd %p has"
3072 " completed with data overrun "
3073 "reported\n", cmd);
3074 break;
3075 case CMD_INVALID:
3076 dev_warn(&h->pdev->dev, "cciss: cmd %p is "
3077 "reported invalid\n", cmd);
3078 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3079 cmd->err_info->CommandStatus, DRIVER_OK,
3080 (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3081 DID_PASSTHROUGH : DID_ERROR);
3082 break;
3083 case CMD_PROTOCOL_ERR:
3084 dev_warn(&h->pdev->dev, "cciss: cmd %p has "
3085 "protocol error\n", cmd);
3086 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3087 cmd->err_info->CommandStatus, DRIVER_OK,
3088 (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3089 DID_PASSTHROUGH : DID_ERROR);
3090 break;
3091 case CMD_HARDWARE_ERR:
3092 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3093 " hardware error\n", cmd);
3094 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3095 cmd->err_info->CommandStatus, DRIVER_OK,
3096 (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3097 DID_PASSTHROUGH : DID_ERROR);
3098 break;
3099 case CMD_CONNECTION_LOST:
3100 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3101 "connection lost\n", cmd);
3102 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3103 cmd->err_info->CommandStatus, DRIVER_OK,
3104 (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3105 DID_PASSTHROUGH : DID_ERROR);
3106 break;
3107 case CMD_ABORTED:
3108 dev_warn(&h->pdev->dev, "cciss: cmd %p was "
3109 "aborted\n", cmd);
3110 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3111 cmd->err_info->CommandStatus, DRIVER_OK,
3112 (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3113 DID_PASSTHROUGH : DID_ABORT);
3114 break;
3115 case CMD_ABORT_FAILED:
3116 dev_warn(&h->pdev->dev, "cciss: cmd %p reports "
3117 "abort failed\n", cmd);
3118 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3119 cmd->err_info->CommandStatus, DRIVER_OK,
3120 (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3121 DID_PASSTHROUGH : DID_ERROR);
3122 break;
3123 case CMD_UNSOLICITED_ABORT:
3124 dev_warn(&h->pdev->dev, "cciss%d: unsolicited "
3125 "abort %p\n", h->ctlr, cmd);
3126 if (cmd->retry_count < MAX_CMD_RETRIES) {
3127 retry_cmd = 1;
3128 dev_warn(&h->pdev->dev, "retrying %p\n", cmd);
3129 cmd->retry_count++;
3130 } else
3131 dev_warn(&h->pdev->dev,
3132 "%p retried too many times\n", cmd);
3133 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3134 cmd->err_info->CommandStatus, DRIVER_OK,
3135 (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3136 DID_PASSTHROUGH : DID_ABORT);
3137 break;
3138 case CMD_TIMEOUT:
3139 dev_warn(&h->pdev->dev, "cmd %p timedout\n", cmd);
3140 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3141 cmd->err_info->CommandStatus, DRIVER_OK,
3142 (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3143 DID_PASSTHROUGH : DID_ERROR);
3144 break;
3145 default:
3146 dev_warn(&h->pdev->dev, "cmd %p returned "
3147 "unknown status %x\n", cmd,
3148 cmd->err_info->CommandStatus);
3149 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3150 cmd->err_info->CommandStatus, DRIVER_OK,
3151 (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3152 DID_PASSTHROUGH : DID_ERROR);
3155 after_error_processing:
3157 /* We need to return this command */
3158 if (retry_cmd) {
3159 resend_cciss_cmd(h, cmd);
3160 return;
3162 cmd->rq->completion_data = cmd;
3163 blk_complete_request(cmd->rq);
3166 static inline u32 cciss_tag_contains_index(u32 tag)
3168 #define DIRECT_LOOKUP_BIT 0x10
3169 return tag & DIRECT_LOOKUP_BIT;
3172 static inline u32 cciss_tag_to_index(u32 tag)
3174 #define DIRECT_LOOKUP_SHIFT 5
3175 return tag >> DIRECT_LOOKUP_SHIFT;
3178 static inline u32 cciss_tag_discard_error_bits(u32 tag)
3180 #define CCISS_ERROR_BITS 0x03
3181 return tag & ~CCISS_ERROR_BITS;
3184 static inline void cciss_mark_tag_indexed(u32 *tag)
3186 *tag |= DIRECT_LOOKUP_BIT;
3189 static inline void cciss_set_tag_index(u32 *tag, u32 index)
3191 *tag |= (index << DIRECT_LOOKUP_SHIFT);
3195 * Get a request and submit it to the controller.
3197 static void do_cciss_request(struct request_queue *q)
3199 ctlr_info_t *h = q->queuedata;
3200 CommandList_struct *c;
3201 sector_t start_blk;
3202 int seg;
3203 struct request *creq;
3204 u64bit temp64;
3205 struct scatterlist *tmp_sg;
3206 SGDescriptor_struct *curr_sg;
3207 drive_info_struct *drv;
3208 int i, dir;
3209 int sg_index = 0;
3210 int chained = 0;
3212 queue:
3213 creq = blk_peek_request(q);
3214 if (!creq)
3215 goto startio;
3217 BUG_ON(creq->nr_phys_segments > h->maxsgentries);
3219 c = cmd_alloc(h);
3220 if (!c)
3221 goto full;
3223 blk_start_request(creq);
3225 tmp_sg = h->scatter_list[c->cmdindex];
3226 spin_unlock_irq(q->queue_lock);
3228 c->cmd_type = CMD_RWREQ;
3229 c->rq = creq;
3231 /* fill in the request */
3232 drv = creq->rq_disk->private_data;
3233 c->Header.ReplyQueue = 0; /* unused in simple mode */
3234 /* got command from pool, so use the command block index instead */
3235 /* for direct lookups. */
3236 /* The first 2 bits are reserved for controller error reporting. */
3237 cciss_set_tag_index(&c->Header.Tag.lower, c->cmdindex);
3238 cciss_mark_tag_indexed(&c->Header.Tag.lower);
3239 memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3240 c->Request.CDBLen = 10; /* 12 byte commands not in FW yet; */
3241 c->Request.Type.Type = TYPE_CMD; /* It is a command. */
3242 c->Request.Type.Attribute = ATTR_SIMPLE;
3243 c->Request.Type.Direction =
3244 (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3245 c->Request.Timeout = 0; /* Don't time out */
3246 c->Request.CDB[0] =
3247 (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3248 start_blk = blk_rq_pos(creq);
3249 dev_dbg(&h->pdev->dev, "sector =%d nr_sectors=%d\n",
3250 (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3251 sg_init_table(tmp_sg, h->maxsgentries);
3252 seg = blk_rq_map_sg(q, creq, tmp_sg);
3254 /* get the DMA records for the setup */
3255 if (c->Request.Type.Direction == XFER_READ)
3256 dir = PCI_DMA_FROMDEVICE;
3257 else
3258 dir = PCI_DMA_TODEVICE;
3260 curr_sg = c->SG;
3261 sg_index = 0;
3262 chained = 0;
3264 for (i = 0; i < seg; i++) {
3265 if (((sg_index+1) == (h->max_cmd_sgentries)) &&
3266 !chained && ((seg - i) > 1)) {
3267 /* Point to next chain block. */
3268 curr_sg = h->cmd_sg_list[c->cmdindex];
3269 sg_index = 0;
3270 chained = 1;
3272 curr_sg[sg_index].Len = tmp_sg[i].length;
3273 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3274 tmp_sg[i].offset,
3275 tmp_sg[i].length, dir);
3276 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3277 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3278 curr_sg[sg_index].Ext = 0; /* we are not chaining */
3279 ++sg_index;
3281 if (chained)
3282 cciss_map_sg_chain_block(h, c, h->cmd_sg_list[c->cmdindex],
3283 (seg - (h->max_cmd_sgentries - 1)) *
3284 sizeof(SGDescriptor_struct));
3286 /* track how many SG entries we are using */
3287 if (seg > h->maxSG)
3288 h->maxSG = seg;
3290 dev_dbg(&h->pdev->dev, "Submitting %u sectors in %d segments "
3291 "chained[%d]\n",
3292 blk_rq_sectors(creq), seg, chained);
3294 c->Header.SGTotal = seg + chained;
3295 if (seg <= h->max_cmd_sgentries)
3296 c->Header.SGList = c->Header.SGTotal;
3297 else
3298 c->Header.SGList = h->max_cmd_sgentries;
3299 set_performant_mode(h, c);
3301 if (likely(creq->cmd_type == REQ_TYPE_FS)) {
3302 if(h->cciss_read == CCISS_READ_10) {
3303 c->Request.CDB[1] = 0;
3304 c->Request.CDB[2] = (start_blk >> 24) & 0xff; /* MSB */
3305 c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3306 c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3307 c->Request.CDB[5] = start_blk & 0xff;
3308 c->Request.CDB[6] = 0; /* (sect >> 24) & 0xff; MSB */
3309 c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3310 c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3311 c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3312 } else {
3313 u32 upper32 = upper_32_bits(start_blk);
3315 c->Request.CDBLen = 16;
3316 c->Request.CDB[1]= 0;
3317 c->Request.CDB[2]= (upper32 >> 24) & 0xff; /* MSB */
3318 c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3319 c->Request.CDB[4]= (upper32 >> 8) & 0xff;
3320 c->Request.CDB[5]= upper32 & 0xff;
3321 c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3322 c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3323 c->Request.CDB[8]= (start_blk >> 8) & 0xff;
3324 c->Request.CDB[9]= start_blk & 0xff;
3325 c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3326 c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3327 c->Request.CDB[12]= (blk_rq_sectors(creq) >> 8) & 0xff;
3328 c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3329 c->Request.CDB[14] = c->Request.CDB[15] = 0;
3331 } else if (creq->cmd_type == REQ_TYPE_BLOCK_PC) {
3332 c->Request.CDBLen = creq->cmd_len;
3333 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3334 } else {
3335 dev_warn(&h->pdev->dev, "bad request type %d\n",
3336 creq->cmd_type);
3337 BUG();
3340 spin_lock_irq(q->queue_lock);
3342 addQ(&h->reqQ, c);
3343 h->Qdepth++;
3344 if (h->Qdepth > h->maxQsinceinit)
3345 h->maxQsinceinit = h->Qdepth;
3347 goto queue;
3348 full:
3349 blk_stop_queue(q);
3350 startio:
3351 /* We will already have the driver lock here so not need
3352 * to lock it.
3354 start_io(h);
3357 static inline unsigned long get_next_completion(ctlr_info_t *h)
3359 return h->access.command_completed(h);
3362 static inline int interrupt_pending(ctlr_info_t *h)
3364 return h->access.intr_pending(h);
3367 static inline long interrupt_not_for_us(ctlr_info_t *h)
3369 return ((h->access.intr_pending(h) == 0) ||
3370 (h->interrupts_enabled == 0));
3373 static inline int bad_tag(ctlr_info_t *h, u32 tag_index,
3374 u32 raw_tag)
3376 if (unlikely(tag_index >= h->nr_cmds)) {
3377 dev_warn(&h->pdev->dev, "bad tag 0x%08x ignored.\n", raw_tag);
3378 return 1;
3380 return 0;
3383 static inline void finish_cmd(ctlr_info_t *h, CommandList_struct *c,
3384 u32 raw_tag)
3386 removeQ(c);
3387 if (likely(c->cmd_type == CMD_RWREQ))
3388 complete_command(h, c, 0);
3389 else if (c->cmd_type == CMD_IOCTL_PEND)
3390 complete(c->waiting);
3391 #ifdef CONFIG_CISS_SCSI_TAPE
3392 else if (c->cmd_type == CMD_SCSI)
3393 complete_scsi_command(c, 0, raw_tag);
3394 #endif
3397 static inline u32 next_command(ctlr_info_t *h)
3399 u32 a;
3401 if (unlikely(h->transMethod != CFGTBL_Trans_Performant))
3402 return h->access.command_completed(h);
3404 if ((*(h->reply_pool_head) & 1) == (h->reply_pool_wraparound)) {
3405 a = *(h->reply_pool_head); /* Next cmd in ring buffer */
3406 (h->reply_pool_head)++;
3407 h->commands_outstanding--;
3408 } else {
3409 a = FIFO_EMPTY;
3411 /* Check for wraparound */
3412 if (h->reply_pool_head == (h->reply_pool + h->max_commands)) {
3413 h->reply_pool_head = h->reply_pool;
3414 h->reply_pool_wraparound ^= 1;
3416 return a;
3419 /* process completion of an indexed ("direct lookup") command */
3420 static inline u32 process_indexed_cmd(ctlr_info_t *h, u32 raw_tag)
3422 u32 tag_index;
3423 CommandList_struct *c;
3425 tag_index = cciss_tag_to_index(raw_tag);
3426 if (bad_tag(h, tag_index, raw_tag))
3427 return next_command(h);
3428 c = h->cmd_pool + tag_index;
3429 finish_cmd(h, c, raw_tag);
3430 return next_command(h);
3433 /* process completion of a non-indexed command */
3434 static inline u32 process_nonindexed_cmd(ctlr_info_t *h, u32 raw_tag)
3436 u32 tag;
3437 CommandList_struct *c = NULL;
3438 __u32 busaddr_masked, tag_masked;
3440 tag = cciss_tag_discard_error_bits(raw_tag);
3441 list_for_each_entry(c, &h->cmpQ, list) {
3442 busaddr_masked = cciss_tag_discard_error_bits(c->busaddr);
3443 tag_masked = cciss_tag_discard_error_bits(tag);
3444 if (busaddr_masked == tag_masked) {
3445 finish_cmd(h, c, raw_tag);
3446 return next_command(h);
3449 bad_tag(h, h->nr_cmds + 1, raw_tag);
3450 return next_command(h);
3453 static irqreturn_t do_cciss_intx(int irq, void *dev_id)
3455 ctlr_info_t *h = dev_id;
3456 unsigned long flags;
3457 u32 raw_tag;
3459 if (interrupt_not_for_us(h))
3460 return IRQ_NONE;
3461 spin_lock_irqsave(&h->lock, flags);
3462 while (interrupt_pending(h)) {
3463 raw_tag = get_next_completion(h);
3464 while (raw_tag != FIFO_EMPTY) {
3465 if (cciss_tag_contains_index(raw_tag))
3466 raw_tag = process_indexed_cmd(h, raw_tag);
3467 else
3468 raw_tag = process_nonindexed_cmd(h, raw_tag);
3471 spin_unlock_irqrestore(&h->lock, flags);
3472 return IRQ_HANDLED;
3475 /* Add a second interrupt handler for MSI/MSI-X mode. In this mode we never
3476 * check the interrupt pending register because it is not set.
3478 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id)
3480 ctlr_info_t *h = dev_id;
3481 unsigned long flags;
3482 u32 raw_tag;
3484 spin_lock_irqsave(&h->lock, flags);
3485 raw_tag = get_next_completion(h);
3486 while (raw_tag != FIFO_EMPTY) {
3487 if (cciss_tag_contains_index(raw_tag))
3488 raw_tag = process_indexed_cmd(h, raw_tag);
3489 else
3490 raw_tag = process_nonindexed_cmd(h, raw_tag);
3492 spin_unlock_irqrestore(&h->lock, flags);
3493 return IRQ_HANDLED;
3497 * add_to_scan_list() - add controller to rescan queue
3498 * @h: Pointer to the controller.
3500 * Adds the controller to the rescan queue if not already on the queue.
3502 * returns 1 if added to the queue, 0 if skipped (could be on the
3503 * queue already, or the controller could be initializing or shutting
3504 * down).
3506 static int add_to_scan_list(struct ctlr_info *h)
3508 struct ctlr_info *test_h;
3509 int found = 0;
3510 int ret = 0;
3512 if (h->busy_initializing)
3513 return 0;
3515 if (!mutex_trylock(&h->busy_shutting_down))
3516 return 0;
3518 mutex_lock(&scan_mutex);
3519 list_for_each_entry(test_h, &scan_q, scan_list) {
3520 if (test_h == h) {
3521 found = 1;
3522 break;
3525 if (!found && !h->busy_scanning) {
3526 INIT_COMPLETION(h->scan_wait);
3527 list_add_tail(&h->scan_list, &scan_q);
3528 ret = 1;
3530 mutex_unlock(&scan_mutex);
3531 mutex_unlock(&h->busy_shutting_down);
3533 return ret;
3537 * remove_from_scan_list() - remove controller from rescan queue
3538 * @h: Pointer to the controller.
3540 * Removes the controller from the rescan queue if present. Blocks if
3541 * the controller is currently conducting a rescan. The controller
3542 * can be in one of three states:
3543 * 1. Doesn't need a scan
3544 * 2. On the scan list, but not scanning yet (we remove it)
3545 * 3. Busy scanning (and not on the list). In this case we want to wait for
3546 * the scan to complete to make sure the scanning thread for this
3547 * controller is completely idle.
3549 static void remove_from_scan_list(struct ctlr_info *h)
3551 struct ctlr_info *test_h, *tmp_h;
3553 mutex_lock(&scan_mutex);
3554 list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3555 if (test_h == h) { /* state 2. */
3556 list_del(&h->scan_list);
3557 complete_all(&h->scan_wait);
3558 mutex_unlock(&scan_mutex);
3559 return;
3562 if (h->busy_scanning) { /* state 3. */
3563 mutex_unlock(&scan_mutex);
3564 wait_for_completion(&h->scan_wait);
3565 } else { /* state 1, nothing to do. */
3566 mutex_unlock(&scan_mutex);
3571 * scan_thread() - kernel thread used to rescan controllers
3572 * @data: Ignored.
3574 * A kernel thread used scan for drive topology changes on
3575 * controllers. The thread processes only one controller at a time
3576 * using a queue. Controllers are added to the queue using
3577 * add_to_scan_list() and removed from the queue either after done
3578 * processing or using remove_from_scan_list().
3580 * returns 0.
3582 static int scan_thread(void *data)
3584 struct ctlr_info *h;
3586 while (1) {
3587 set_current_state(TASK_INTERRUPTIBLE);
3588 schedule();
3589 if (kthread_should_stop())
3590 break;
3592 while (1) {
3593 mutex_lock(&scan_mutex);
3594 if (list_empty(&scan_q)) {
3595 mutex_unlock(&scan_mutex);
3596 break;
3599 h = list_entry(scan_q.next,
3600 struct ctlr_info,
3601 scan_list);
3602 list_del(&h->scan_list);
3603 h->busy_scanning = 1;
3604 mutex_unlock(&scan_mutex);
3606 rebuild_lun_table(h, 0, 0);
3607 complete_all(&h->scan_wait);
3608 mutex_lock(&scan_mutex);
3609 h->busy_scanning = 0;
3610 mutex_unlock(&scan_mutex);
3614 return 0;
3617 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3619 if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3620 return 0;
3622 switch (c->err_info->SenseInfo[12]) {
3623 case STATE_CHANGED:
3624 dev_warn(&h->pdev->dev, "a state change "
3625 "detected, command retried\n");
3626 return 1;
3627 break;
3628 case LUN_FAILED:
3629 dev_warn(&h->pdev->dev, "LUN failure "
3630 "detected, action required\n");
3631 return 1;
3632 break;
3633 case REPORT_LUNS_CHANGED:
3634 dev_warn(&h->pdev->dev, "report LUN data changed\n");
3636 * Here, we could call add_to_scan_list and wake up the scan thread,
3637 * except that it's quite likely that we will get more than one
3638 * REPORT_LUNS_CHANGED condition in quick succession, which means
3639 * that those which occur after the first one will likely happen
3640 * *during* the scan_thread's rescan. And the rescan code is not
3641 * robust enough to restart in the middle, undoing what it has already
3642 * done, and it's not clear that it's even possible to do this, since
3643 * part of what it does is notify the block layer, which starts
3644 * doing it's own i/o to read partition tables and so on, and the
3645 * driver doesn't have visibility to know what might need undoing.
3646 * In any event, if possible, it is horribly complicated to get right
3647 * so we just don't do it for now.
3649 * Note: this REPORT_LUNS_CHANGED condition only occurs on the MSA2012.
3651 return 1;
3652 break;
3653 case POWER_OR_RESET:
3654 dev_warn(&h->pdev->dev,
3655 "a power on or device reset detected\n");
3656 return 1;
3657 break;
3658 case UNIT_ATTENTION_CLEARED:
3659 dev_warn(&h->pdev->dev,
3660 "unit attention cleared by another initiator\n");
3661 return 1;
3662 break;
3663 default:
3664 dev_warn(&h->pdev->dev, "unknown unit attention detected\n");
3665 return 1;
3670 * We cannot read the structure directly, for portability we must use
3671 * the io functions.
3672 * This is for debug only.
3674 static void print_cfg_table(ctlr_info_t *h)
3676 int i;
3677 char temp_name[17];
3678 CfgTable_struct *tb = h->cfgtable;
3680 dev_dbg(&h->pdev->dev, "Controller Configuration information\n");
3681 dev_dbg(&h->pdev->dev, "------------------------------------\n");
3682 for (i = 0; i < 4; i++)
3683 temp_name[i] = readb(&(tb->Signature[i]));
3684 temp_name[4] = '\0';
3685 dev_dbg(&h->pdev->dev, " Signature = %s\n", temp_name);
3686 dev_dbg(&h->pdev->dev, " Spec Number = %d\n",
3687 readl(&(tb->SpecValence)));
3688 dev_dbg(&h->pdev->dev, " Transport methods supported = 0x%x\n",
3689 readl(&(tb->TransportSupport)));
3690 dev_dbg(&h->pdev->dev, " Transport methods active = 0x%x\n",
3691 readl(&(tb->TransportActive)));
3692 dev_dbg(&h->pdev->dev, " Requested transport Method = 0x%x\n",
3693 readl(&(tb->HostWrite.TransportRequest)));
3694 dev_dbg(&h->pdev->dev, " Coalesce Interrupt Delay = 0x%x\n",
3695 readl(&(tb->HostWrite.CoalIntDelay)));
3696 dev_dbg(&h->pdev->dev, " Coalesce Interrupt Count = 0x%x\n",
3697 readl(&(tb->HostWrite.CoalIntCount)));
3698 dev_dbg(&h->pdev->dev, " Max outstanding commands = 0x%d\n",
3699 readl(&(tb->CmdsOutMax)));
3700 dev_dbg(&h->pdev->dev, " Bus Types = 0x%x\n",
3701 readl(&(tb->BusTypes)));
3702 for (i = 0; i < 16; i++)
3703 temp_name[i] = readb(&(tb->ServerName[i]));
3704 temp_name[16] = '\0';
3705 dev_dbg(&h->pdev->dev, " Server Name = %s\n", temp_name);
3706 dev_dbg(&h->pdev->dev, " Heartbeat Counter = 0x%x\n\n\n",
3707 readl(&(tb->HeartBeat)));
3710 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3712 int i, offset, mem_type, bar_type;
3713 if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3714 return 0;
3715 offset = 0;
3716 for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3717 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3718 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3719 offset += 4;
3720 else {
3721 mem_type = pci_resource_flags(pdev, i) &
3722 PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3723 switch (mem_type) {
3724 case PCI_BASE_ADDRESS_MEM_TYPE_32:
3725 case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3726 offset += 4; /* 32 bit */
3727 break;
3728 case PCI_BASE_ADDRESS_MEM_TYPE_64:
3729 offset += 8;
3730 break;
3731 default: /* reserved in PCI 2.2 */
3732 dev_warn(&pdev->dev,
3733 "Base address is invalid\n");
3734 return -1;
3735 break;
3738 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3739 return i + 1;
3741 return -1;
3744 /* Fill in bucket_map[], given nsgs (the max number of
3745 * scatter gather elements supported) and bucket[],
3746 * which is an array of 8 integers. The bucket[] array
3747 * contains 8 different DMA transfer sizes (in 16
3748 * byte increments) which the controller uses to fetch
3749 * commands. This function fills in bucket_map[], which
3750 * maps a given number of scatter gather elements to one of
3751 * the 8 DMA transfer sizes. The point of it is to allow the
3752 * controller to only do as much DMA as needed to fetch the
3753 * command, with the DMA transfer size encoded in the lower
3754 * bits of the command address.
3756 static void calc_bucket_map(int bucket[], int num_buckets,
3757 int nsgs, int *bucket_map)
3759 int i, j, b, size;
3761 /* even a command with 0 SGs requires 4 blocks */
3762 #define MINIMUM_TRANSFER_BLOCKS 4
3763 #define NUM_BUCKETS 8
3764 /* Note, bucket_map must have nsgs+1 entries. */
3765 for (i = 0; i <= nsgs; i++) {
3766 /* Compute size of a command with i SG entries */
3767 size = i + MINIMUM_TRANSFER_BLOCKS;
3768 b = num_buckets; /* Assume the biggest bucket */
3769 /* Find the bucket that is just big enough */
3770 for (j = 0; j < 8; j++) {
3771 if (bucket[j] >= size) {
3772 b = j;
3773 break;
3776 /* for a command with i SG entries, use bucket b. */
3777 bucket_map[i] = b;
3781 static void __devinit cciss_wait_for_mode_change_ack(ctlr_info_t *h)
3783 int i;
3785 /* under certain very rare conditions, this can take awhile.
3786 * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3787 * as we enter this code.) */
3788 for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3789 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3790 break;
3791 usleep_range(10000, 20000);
3795 static __devinit void cciss_enter_performant_mode(ctlr_info_t *h)
3797 /* This is a bit complicated. There are 8 registers on
3798 * the controller which we write to to tell it 8 different
3799 * sizes of commands which there may be. It's a way of
3800 * reducing the DMA done to fetch each command. Encoded into
3801 * each command's tag are 3 bits which communicate to the controller
3802 * which of the eight sizes that command fits within. The size of
3803 * each command depends on how many scatter gather entries there are.
3804 * Each SG entry requires 16 bytes. The eight registers are programmed
3805 * with the number of 16-byte blocks a command of that size requires.
3806 * The smallest command possible requires 5 such 16 byte blocks.
3807 * the largest command possible requires MAXSGENTRIES + 4 16-byte
3808 * blocks. Note, this only extends to the SG entries contained
3809 * within the command block, and does not extend to chained blocks
3810 * of SG elements. bft[] contains the eight values we write to
3811 * the registers. They are not evenly distributed, but have more
3812 * sizes for small commands, and fewer sizes for larger commands.
3814 __u32 trans_offset;
3815 int bft[8] = { 5, 6, 8, 10, 12, 20, 28, MAXSGENTRIES + 4};
3817 * 5 = 1 s/g entry or 4k
3818 * 6 = 2 s/g entry or 8k
3819 * 8 = 4 s/g entry or 16k
3820 * 10 = 6 s/g entry or 24k
3822 unsigned long register_value;
3823 BUILD_BUG_ON(28 > MAXSGENTRIES + 4);
3825 h->reply_pool_wraparound = 1; /* spec: init to 1 */
3827 /* Controller spec: zero out this buffer. */
3828 memset(h->reply_pool, 0, h->max_commands * sizeof(__u64));
3829 h->reply_pool_head = h->reply_pool;
3831 trans_offset = readl(&(h->cfgtable->TransMethodOffset));
3832 calc_bucket_map(bft, ARRAY_SIZE(bft), h->maxsgentries,
3833 h->blockFetchTable);
3834 writel(bft[0], &h->transtable->BlockFetch0);
3835 writel(bft[1], &h->transtable->BlockFetch1);
3836 writel(bft[2], &h->transtable->BlockFetch2);
3837 writel(bft[3], &h->transtable->BlockFetch3);
3838 writel(bft[4], &h->transtable->BlockFetch4);
3839 writel(bft[5], &h->transtable->BlockFetch5);
3840 writel(bft[6], &h->transtable->BlockFetch6);
3841 writel(bft[7], &h->transtable->BlockFetch7);
3843 /* size of controller ring buffer */
3844 writel(h->max_commands, &h->transtable->RepQSize);
3845 writel(1, &h->transtable->RepQCount);
3846 writel(0, &h->transtable->RepQCtrAddrLow32);
3847 writel(0, &h->transtable->RepQCtrAddrHigh32);
3848 writel(h->reply_pool_dhandle, &h->transtable->RepQAddr0Low32);
3849 writel(0, &h->transtable->RepQAddr0High32);
3850 writel(CFGTBL_Trans_Performant,
3851 &(h->cfgtable->HostWrite.TransportRequest));
3853 writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
3854 cciss_wait_for_mode_change_ack(h);
3855 register_value = readl(&(h->cfgtable->TransportActive));
3856 if (!(register_value & CFGTBL_Trans_Performant))
3857 dev_warn(&h->pdev->dev, "cciss: unable to get board into"
3858 " performant mode\n");
3861 static void __devinit cciss_put_controller_into_performant_mode(ctlr_info_t *h)
3863 __u32 trans_support;
3865 dev_dbg(&h->pdev->dev, "Trying to put board into Performant mode\n");
3866 /* Attempt to put controller into performant mode if supported */
3867 /* Does board support performant mode? */
3868 trans_support = readl(&(h->cfgtable->TransportSupport));
3869 if (!(trans_support & PERFORMANT_MODE))
3870 return;
3872 dev_dbg(&h->pdev->dev, "Placing controller into performant mode\n");
3873 /* Performant mode demands commands on a 32 byte boundary
3874 * pci_alloc_consistent aligns on page boundarys already.
3875 * Just need to check if divisible by 32
3877 if ((sizeof(CommandList_struct) % 32) != 0) {
3878 dev_warn(&h->pdev->dev, "%s %d %s\n",
3879 "cciss info: command size[",
3880 (int)sizeof(CommandList_struct),
3881 "] not divisible by 32, no performant mode..\n");
3882 return;
3885 /* Performant mode ring buffer and supporting data structures */
3886 h->reply_pool = (__u64 *)pci_alloc_consistent(
3887 h->pdev, h->max_commands * sizeof(__u64),
3888 &(h->reply_pool_dhandle));
3890 /* Need a block fetch table for performant mode */
3891 h->blockFetchTable = kmalloc(((h->maxsgentries+1) *
3892 sizeof(__u32)), GFP_KERNEL);
3894 if ((h->reply_pool == NULL) || (h->blockFetchTable == NULL))
3895 goto clean_up;
3897 cciss_enter_performant_mode(h);
3899 /* Change the access methods to the performant access methods */
3900 h->access = SA5_performant_access;
3901 h->transMethod = CFGTBL_Trans_Performant;
3903 return;
3904 clean_up:
3905 kfree(h->blockFetchTable);
3906 if (h->reply_pool)
3907 pci_free_consistent(h->pdev,
3908 h->max_commands * sizeof(__u64),
3909 h->reply_pool,
3910 h->reply_pool_dhandle);
3911 return;
3913 } /* cciss_put_controller_into_performant_mode */
3915 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
3916 * controllers that are capable. If not, we use IO-APIC mode.
3919 static void __devinit cciss_interrupt_mode(ctlr_info_t *h)
3921 #ifdef CONFIG_PCI_MSI
3922 int err;
3923 struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
3924 {0, 2}, {0, 3}
3927 /* Some boards advertise MSI but don't really support it */
3928 if ((h->board_id == 0x40700E11) || (h->board_id == 0x40800E11) ||
3929 (h->board_id == 0x40820E11) || (h->board_id == 0x40830E11))
3930 goto default_int_mode;
3932 if (pci_find_capability(h->pdev, PCI_CAP_ID_MSIX)) {
3933 err = pci_enable_msix(h->pdev, cciss_msix_entries, 4);
3934 if (!err) {
3935 h->intr[0] = cciss_msix_entries[0].vector;
3936 h->intr[1] = cciss_msix_entries[1].vector;
3937 h->intr[2] = cciss_msix_entries[2].vector;
3938 h->intr[3] = cciss_msix_entries[3].vector;
3939 h->msix_vector = 1;
3940 return;
3942 if (err > 0) {
3943 dev_warn(&h->pdev->dev,
3944 "only %d MSI-X vectors available\n", err);
3945 goto default_int_mode;
3946 } else {
3947 dev_warn(&h->pdev->dev,
3948 "MSI-X init failed %d\n", err);
3949 goto default_int_mode;
3952 if (pci_find_capability(h->pdev, PCI_CAP_ID_MSI)) {
3953 if (!pci_enable_msi(h->pdev))
3954 h->msi_vector = 1;
3955 else
3956 dev_warn(&h->pdev->dev, "MSI init failed\n");
3958 default_int_mode:
3959 #endif /* CONFIG_PCI_MSI */
3960 /* if we get here we're going to use the default interrupt mode */
3961 h->intr[PERF_MODE_INT] = h->pdev->irq;
3962 return;
3965 static int __devinit cciss_lookup_board_id(struct pci_dev *pdev, u32 *board_id)
3967 int i;
3968 u32 subsystem_vendor_id, subsystem_device_id;
3970 subsystem_vendor_id = pdev->subsystem_vendor;
3971 subsystem_device_id = pdev->subsystem_device;
3972 *board_id = ((subsystem_device_id << 16) & 0xffff0000) |
3973 subsystem_vendor_id;
3975 for (i = 0; i < ARRAY_SIZE(products); i++)
3976 if (*board_id == products[i].board_id)
3977 return i;
3978 dev_warn(&pdev->dev, "unrecognized board ID: 0x%08x, ignoring.\n",
3979 *board_id);
3980 return -ENODEV;
3983 static inline bool cciss_board_disabled(ctlr_info_t *h)
3985 u16 command;
3987 (void) pci_read_config_word(h->pdev, PCI_COMMAND, &command);
3988 return ((command & PCI_COMMAND_MEMORY) == 0);
3991 static int __devinit cciss_pci_find_memory_BAR(struct pci_dev *pdev,
3992 unsigned long *memory_bar)
3994 int i;
3996 for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
3997 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM) {
3998 /* addressing mode bits already removed */
3999 *memory_bar = pci_resource_start(pdev, i);
4000 dev_dbg(&pdev->dev, "memory BAR = %lx\n",
4001 *memory_bar);
4002 return 0;
4004 dev_warn(&pdev->dev, "no memory BAR found\n");
4005 return -ENODEV;
4008 static int __devinit cciss_wait_for_board_state(struct pci_dev *pdev,
4009 void __iomem *vaddr, int wait_for_ready)
4010 #define BOARD_READY 1
4011 #define BOARD_NOT_READY 0
4013 int i, iterations;
4014 u32 scratchpad;
4016 if (wait_for_ready)
4017 iterations = CCISS_BOARD_READY_ITERATIONS;
4018 else
4019 iterations = CCISS_BOARD_NOT_READY_ITERATIONS;
4021 for (i = 0; i < iterations; i++) {
4022 scratchpad = readl(vaddr + SA5_SCRATCHPAD_OFFSET);
4023 if (wait_for_ready) {
4024 if (scratchpad == CCISS_FIRMWARE_READY)
4025 return 0;
4026 } else {
4027 if (scratchpad != CCISS_FIRMWARE_READY)
4028 return 0;
4030 msleep(CCISS_BOARD_READY_POLL_INTERVAL_MSECS);
4032 dev_warn(&pdev->dev, "board not ready, timed out.\n");
4033 return -ENODEV;
4036 static int __devinit cciss_find_cfg_addrs(struct pci_dev *pdev,
4037 void __iomem *vaddr, u32 *cfg_base_addr, u64 *cfg_base_addr_index,
4038 u64 *cfg_offset)
4040 *cfg_base_addr = readl(vaddr + SA5_CTCFG_OFFSET);
4041 *cfg_offset = readl(vaddr + SA5_CTMEM_OFFSET);
4042 *cfg_base_addr &= (u32) 0x0000ffff;
4043 *cfg_base_addr_index = find_PCI_BAR_index(pdev, *cfg_base_addr);
4044 if (*cfg_base_addr_index == -1) {
4045 dev_warn(&pdev->dev, "cannot find cfg_base_addr_index, "
4046 "*cfg_base_addr = 0x%08x\n", *cfg_base_addr);
4047 return -ENODEV;
4049 return 0;
4052 static int __devinit cciss_find_cfgtables(ctlr_info_t *h)
4054 u64 cfg_offset;
4055 u32 cfg_base_addr;
4056 u64 cfg_base_addr_index;
4057 u32 trans_offset;
4058 int rc;
4060 rc = cciss_find_cfg_addrs(h->pdev, h->vaddr, &cfg_base_addr,
4061 &cfg_base_addr_index, &cfg_offset);
4062 if (rc)
4063 return rc;
4064 h->cfgtable = remap_pci_mem(pci_resource_start(h->pdev,
4065 cfg_base_addr_index) + cfg_offset, sizeof(h->cfgtable));
4066 if (!h->cfgtable)
4067 return -ENOMEM;
4068 /* Find performant mode table. */
4069 trans_offset = readl(&h->cfgtable->TransMethodOffset);
4070 h->transtable = remap_pci_mem(pci_resource_start(h->pdev,
4071 cfg_base_addr_index)+cfg_offset+trans_offset,
4072 sizeof(*h->transtable));
4073 if (!h->transtable)
4074 return -ENOMEM;
4075 return 0;
4078 static void __devinit cciss_get_max_perf_mode_cmds(struct ctlr_info *h)
4080 h->max_commands = readl(&(h->cfgtable->MaxPerformantModeCommands));
4082 /* Limit commands in memory limited kdump scenario. */
4083 if (reset_devices && h->max_commands > 32)
4084 h->max_commands = 32;
4086 if (h->max_commands < 16) {
4087 dev_warn(&h->pdev->dev, "Controller reports "
4088 "max supported commands of %d, an obvious lie. "
4089 "Using 16. Ensure that firmware is up to date.\n",
4090 h->max_commands);
4091 h->max_commands = 16;
4095 /* Interrogate the hardware for some limits:
4096 * max commands, max SG elements without chaining, and with chaining,
4097 * SG chain block size, etc.
4099 static void __devinit cciss_find_board_params(ctlr_info_t *h)
4101 cciss_get_max_perf_mode_cmds(h);
4102 h->nr_cmds = h->max_commands - 4; /* Allow room for some ioctls */
4103 h->maxsgentries = readl(&(h->cfgtable->MaxSGElements));
4105 * Limit in-command s/g elements to 32 save dma'able memory.
4106 * Howvever spec says if 0, use 31
4108 h->max_cmd_sgentries = 31;
4109 if (h->maxsgentries > 512) {
4110 h->max_cmd_sgentries = 32;
4111 h->chainsize = h->maxsgentries - h->max_cmd_sgentries + 1;
4112 h->maxsgentries--; /* save one for chain pointer */
4113 } else {
4114 h->maxsgentries = 31; /* default to traditional values */
4115 h->chainsize = 0;
4119 static inline bool CISS_signature_present(ctlr_info_t *h)
4121 if ((readb(&h->cfgtable->Signature[0]) != 'C') ||
4122 (readb(&h->cfgtable->Signature[1]) != 'I') ||
4123 (readb(&h->cfgtable->Signature[2]) != 'S') ||
4124 (readb(&h->cfgtable->Signature[3]) != 'S')) {
4125 dev_warn(&h->pdev->dev, "not a valid CISS config table\n");
4126 return false;
4128 return true;
4131 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
4132 static inline void cciss_enable_scsi_prefetch(ctlr_info_t *h)
4134 #ifdef CONFIG_X86
4135 u32 prefetch;
4137 prefetch = readl(&(h->cfgtable->SCSI_Prefetch));
4138 prefetch |= 0x100;
4139 writel(prefetch, &(h->cfgtable->SCSI_Prefetch));
4140 #endif
4143 /* Disable DMA prefetch for the P600. Otherwise an ASIC bug may result
4144 * in a prefetch beyond physical memory.
4146 static inline void cciss_p600_dma_prefetch_quirk(ctlr_info_t *h)
4148 u32 dma_prefetch;
4149 __u32 dma_refetch;
4151 if (h->board_id != 0x3225103C)
4152 return;
4153 dma_prefetch = readl(h->vaddr + I2O_DMA1_CFG);
4154 dma_prefetch |= 0x8000;
4155 writel(dma_prefetch, h->vaddr + I2O_DMA1_CFG);
4156 pci_read_config_dword(h->pdev, PCI_COMMAND_PARITY, &dma_refetch);
4157 dma_refetch |= 0x1;
4158 pci_write_config_dword(h->pdev, PCI_COMMAND_PARITY, dma_refetch);
4161 static int __devinit cciss_pci_init(ctlr_info_t *h)
4163 int prod_index, err;
4165 prod_index = cciss_lookup_board_id(h->pdev, &h->board_id);
4166 if (prod_index < 0)
4167 return -ENODEV;
4168 h->product_name = products[prod_index].product_name;
4169 h->access = *(products[prod_index].access);
4171 if (cciss_board_disabled(h)) {
4172 dev_warn(&h->pdev->dev, "controller appears to be disabled\n");
4173 return -ENODEV;
4175 err = pci_enable_device(h->pdev);
4176 if (err) {
4177 dev_warn(&h->pdev->dev, "Unable to Enable PCI device\n");
4178 return err;
4181 err = pci_request_regions(h->pdev, "cciss");
4182 if (err) {
4183 dev_warn(&h->pdev->dev,
4184 "Cannot obtain PCI resources, aborting\n");
4185 return err;
4188 dev_dbg(&h->pdev->dev, "irq = %x\n", h->pdev->irq);
4189 dev_dbg(&h->pdev->dev, "board_id = %x\n", h->board_id);
4191 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
4192 * else we use the IO-APIC interrupt assigned to us by system ROM.
4194 cciss_interrupt_mode(h);
4195 err = cciss_pci_find_memory_BAR(h->pdev, &h->paddr);
4196 if (err)
4197 goto err_out_free_res;
4198 h->vaddr = remap_pci_mem(h->paddr, 0x250);
4199 if (!h->vaddr) {
4200 err = -ENOMEM;
4201 goto err_out_free_res;
4203 err = cciss_wait_for_board_state(h->pdev, h->vaddr, BOARD_READY);
4204 if (err)
4205 goto err_out_free_res;
4206 err = cciss_find_cfgtables(h);
4207 if (err)
4208 goto err_out_free_res;
4209 print_cfg_table(h);
4210 cciss_find_board_params(h);
4212 if (!CISS_signature_present(h)) {
4213 err = -ENODEV;
4214 goto err_out_free_res;
4216 cciss_enable_scsi_prefetch(h);
4217 cciss_p600_dma_prefetch_quirk(h);
4218 cciss_put_controller_into_performant_mode(h);
4219 return 0;
4221 err_out_free_res:
4223 * Deliberately omit pci_disable_device(): it does something nasty to
4224 * Smart Array controllers that pci_enable_device does not undo
4226 if (h->transtable)
4227 iounmap(h->transtable);
4228 if (h->cfgtable)
4229 iounmap(h->cfgtable);
4230 if (h->vaddr)
4231 iounmap(h->vaddr);
4232 pci_release_regions(h->pdev);
4233 return err;
4236 /* Function to find the first free pointer into our hba[] array
4237 * Returns -1 if no free entries are left.
4239 static int alloc_cciss_hba(struct pci_dev *pdev)
4241 int i;
4243 for (i = 0; i < MAX_CTLR; i++) {
4244 if (!hba[i]) {
4245 ctlr_info_t *h;
4247 h = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
4248 if (!h)
4249 goto Enomem;
4250 hba[i] = h;
4251 return i;
4254 dev_warn(&pdev->dev, "This driver supports a maximum"
4255 " of %d controllers.\n", MAX_CTLR);
4256 return -1;
4257 Enomem:
4258 dev_warn(&pdev->dev, "out of memory.\n");
4259 return -1;
4262 static void free_hba(ctlr_info_t *h)
4264 int i;
4266 hba[h->ctlr] = NULL;
4267 for (i = 0; i < h->highest_lun + 1; i++)
4268 if (h->gendisk[i] != NULL)
4269 put_disk(h->gendisk[i]);
4270 kfree(h);
4273 /* Send a message CDB to the firmware. */
4274 static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
4276 typedef struct {
4277 CommandListHeader_struct CommandHeader;
4278 RequestBlock_struct Request;
4279 ErrDescriptor_struct ErrorDescriptor;
4280 } Command;
4281 static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
4282 Command *cmd;
4283 dma_addr_t paddr64;
4284 uint32_t paddr32, tag;
4285 void __iomem *vaddr;
4286 int i, err;
4288 vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
4289 if (vaddr == NULL)
4290 return -ENOMEM;
4292 /* The Inbound Post Queue only accepts 32-bit physical addresses for the
4293 CCISS commands, so they must be allocated from the lower 4GiB of
4294 memory. */
4295 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
4296 if (err) {
4297 iounmap(vaddr);
4298 return -ENOMEM;
4301 cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
4302 if (cmd == NULL) {
4303 iounmap(vaddr);
4304 return -ENOMEM;
4307 /* This must fit, because of the 32-bit consistent DMA mask. Also,
4308 although there's no guarantee, we assume that the address is at
4309 least 4-byte aligned (most likely, it's page-aligned). */
4310 paddr32 = paddr64;
4312 cmd->CommandHeader.ReplyQueue = 0;
4313 cmd->CommandHeader.SGList = 0;
4314 cmd->CommandHeader.SGTotal = 0;
4315 cmd->CommandHeader.Tag.lower = paddr32;
4316 cmd->CommandHeader.Tag.upper = 0;
4317 memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
4319 cmd->Request.CDBLen = 16;
4320 cmd->Request.Type.Type = TYPE_MSG;
4321 cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
4322 cmd->Request.Type.Direction = XFER_NONE;
4323 cmd->Request.Timeout = 0; /* Don't time out */
4324 cmd->Request.CDB[0] = opcode;
4325 cmd->Request.CDB[1] = type;
4326 memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
4328 cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
4329 cmd->ErrorDescriptor.Addr.upper = 0;
4330 cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
4332 writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
4334 for (i = 0; i < 10; i++) {
4335 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
4336 if ((tag & ~3) == paddr32)
4337 break;
4338 schedule_timeout_uninterruptible(HZ);
4341 iounmap(vaddr);
4343 /* we leak the DMA buffer here ... no choice since the controller could
4344 still complete the command. */
4345 if (i == 10) {
4346 dev_err(&pdev->dev,
4347 "controller message %02x:%02x timed out\n",
4348 opcode, type);
4349 return -ETIMEDOUT;
4352 pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
4354 if (tag & 2) {
4355 dev_err(&pdev->dev, "controller message %02x:%02x failed\n",
4356 opcode, type);
4357 return -EIO;
4360 dev_info(&pdev->dev, "controller message %02x:%02x succeeded\n",
4361 opcode, type);
4362 return 0;
4365 #define cciss_soft_reset_controller(p) cciss_message(p, 1, 0)
4366 #define cciss_noop(p) cciss_message(p, 3, 0)
4368 static int cciss_controller_hard_reset(struct pci_dev *pdev,
4369 void * __iomem vaddr, bool use_doorbell)
4371 u16 pmcsr;
4372 int pos;
4374 if (use_doorbell) {
4375 /* For everything after the P600, the PCI power state method
4376 * of resetting the controller doesn't work, so we have this
4377 * other way using the doorbell register.
4379 dev_info(&pdev->dev, "using doorbell to reset controller\n");
4380 writel(DOORBELL_CTLR_RESET, vaddr + SA5_DOORBELL);
4381 msleep(1000);
4382 } else { /* Try to do it the PCI power state way */
4384 /* Quoting from the Open CISS Specification: "The Power
4385 * Management Control/Status Register (CSR) controls the power
4386 * state of the device. The normal operating state is D0,
4387 * CSR=00h. The software off state is D3, CSR=03h. To reset
4388 * the controller, place the interface device in D3 then to D0,
4389 * this causes a secondary PCI reset which will reset the
4390 * controller." */
4392 pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4393 if (pos == 0) {
4394 dev_err(&pdev->dev,
4395 "cciss_controller_hard_reset: "
4396 "PCI PM not supported\n");
4397 return -ENODEV;
4399 dev_info(&pdev->dev, "using PCI PM to reset controller\n");
4400 /* enter the D3hot power management state */
4401 pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4402 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4403 pmcsr |= PCI_D3hot;
4404 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4406 msleep(500);
4408 /* enter the D0 power management state */
4409 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4410 pmcsr |= PCI_D0;
4411 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4413 msleep(500);
4415 return 0;
4418 /* This does a hard reset of the controller using PCI power management
4419 * states or using the doorbell register. */
4420 static __devinit int cciss_kdump_hard_reset_controller(struct pci_dev *pdev)
4422 u64 cfg_offset;
4423 u32 cfg_base_addr;
4424 u64 cfg_base_addr_index;
4425 void __iomem *vaddr;
4426 unsigned long paddr;
4427 u32 misc_fw_support, active_transport;
4428 int rc;
4429 CfgTable_struct __iomem *cfgtable;
4430 bool use_doorbell;
4431 u32 board_id;
4432 u16 command_register;
4434 /* For controllers as old a the p600, this is very nearly
4435 * the same thing as
4437 * pci_save_state(pci_dev);
4438 * pci_set_power_state(pci_dev, PCI_D3hot);
4439 * pci_set_power_state(pci_dev, PCI_D0);
4440 * pci_restore_state(pci_dev);
4442 * For controllers newer than the P600, the pci power state
4443 * method of resetting doesn't work so we have another way
4444 * using the doorbell register.
4447 /* Exclude 640x boards. These are two pci devices in one slot
4448 * which share a battery backed cache module. One controls the
4449 * cache, the other accesses the cache through the one that controls
4450 * it. If we reset the one controlling the cache, the other will
4451 * likely not be happy. Just forbid resetting this conjoined mess.
4453 cciss_lookup_board_id(pdev, &board_id);
4454 if (board_id == 0x409C0E11 || board_id == 0x409D0E11) {
4455 dev_warn(&pdev->dev, "Cannot reset Smart Array 640x "
4456 "due to shared cache module.");
4457 return -ENODEV;
4460 /* Save the PCI command register */
4461 pci_read_config_word(pdev, 4, &command_register);
4462 /* Turn the board off. This is so that later pci_restore_state()
4463 * won't turn the board on before the rest of config space is ready.
4465 pci_disable_device(pdev);
4466 pci_save_state(pdev);
4468 /* find the first memory BAR, so we can find the cfg table */
4469 rc = cciss_pci_find_memory_BAR(pdev, &paddr);
4470 if (rc)
4471 return rc;
4472 vaddr = remap_pci_mem(paddr, 0x250);
4473 if (!vaddr)
4474 return -ENOMEM;
4476 /* find cfgtable in order to check if reset via doorbell is supported */
4477 rc = cciss_find_cfg_addrs(pdev, vaddr, &cfg_base_addr,
4478 &cfg_base_addr_index, &cfg_offset);
4479 if (rc)
4480 goto unmap_vaddr;
4481 cfgtable = remap_pci_mem(pci_resource_start(pdev,
4482 cfg_base_addr_index) + cfg_offset, sizeof(*cfgtable));
4483 if (!cfgtable) {
4484 rc = -ENOMEM;
4485 goto unmap_vaddr;
4488 /* If reset via doorbell register is supported, use that. */
4489 misc_fw_support = readl(&cfgtable->misc_fw_support);
4490 use_doorbell = misc_fw_support & MISC_FW_DOORBELL_RESET;
4492 /* The doorbell reset seems to cause lockups on some Smart
4493 * Arrays (e.g. P410, P410i, maybe others). Until this is
4494 * fixed or at least isolated, avoid the doorbell reset.
4496 use_doorbell = 0;
4498 rc = cciss_controller_hard_reset(pdev, vaddr, use_doorbell);
4499 if (rc)
4500 goto unmap_cfgtable;
4501 pci_restore_state(pdev);
4502 rc = pci_enable_device(pdev);
4503 if (rc) {
4504 dev_warn(&pdev->dev, "failed to enable device.\n");
4505 goto unmap_cfgtable;
4507 pci_write_config_word(pdev, 4, command_register);
4509 /* Some devices (notably the HP Smart Array 5i Controller)
4510 need a little pause here */
4511 msleep(CCISS_POST_RESET_PAUSE_MSECS);
4513 /* Wait for board to become not ready, then ready. */
4514 dev_info(&pdev->dev, "Waiting for board to become ready.\n");
4515 rc = cciss_wait_for_board_state(pdev, vaddr, BOARD_NOT_READY);
4516 if (rc) /* Don't bail, might be E500, etc. which can't be reset */
4517 dev_warn(&pdev->dev,
4518 "failed waiting for board to become not ready\n");
4519 rc = cciss_wait_for_board_state(pdev, vaddr, BOARD_READY);
4520 if (rc) {
4521 dev_warn(&pdev->dev,
4522 "failed waiting for board to become ready\n");
4523 goto unmap_cfgtable;
4525 dev_info(&pdev->dev, "board ready.\n");
4527 /* Controller should be in simple mode at this point. If it's not,
4528 * It means we're on one of those controllers which doesn't support
4529 * the doorbell reset method and on which the PCI power management reset
4530 * method doesn't work (P800, for example.)
4531 * In those cases, don't try to proceed, as it generally doesn't work.
4533 active_transport = readl(&cfgtable->TransportActive);
4534 if (active_transport & PERFORMANT_MODE) {
4535 dev_warn(&pdev->dev, "Unable to successfully reset controller,"
4536 " Ignoring controller.\n");
4537 rc = -ENODEV;
4540 unmap_cfgtable:
4541 iounmap(cfgtable);
4543 unmap_vaddr:
4544 iounmap(vaddr);
4545 return rc;
4548 static __devinit int cciss_init_reset_devices(struct pci_dev *pdev)
4550 int rc, i;
4552 if (!reset_devices)
4553 return 0;
4555 /* Reset the controller with a PCI power-cycle or via doorbell */
4556 rc = cciss_kdump_hard_reset_controller(pdev);
4558 /* -ENOTSUPP here means we cannot reset the controller
4559 * but it's already (and still) up and running in
4560 * "performant mode". Or, it might be 640x, which can't reset
4561 * due to concerns about shared bbwc between 6402/6404 pair.
4563 if (rc == -ENOTSUPP)
4564 return 0; /* just try to do the kdump anyhow. */
4565 if (rc)
4566 return -ENODEV;
4568 /* Now try to get the controller to respond to a no-op */
4569 for (i = 0; i < CCISS_POST_RESET_NOOP_RETRIES; i++) {
4570 if (cciss_noop(pdev) == 0)
4571 break;
4572 else
4573 dev_warn(&pdev->dev, "no-op failed%s\n",
4574 (i < CCISS_POST_RESET_NOOP_RETRIES - 1 ?
4575 "; re-trying" : ""));
4576 msleep(CCISS_POST_RESET_NOOP_INTERVAL_MSECS);
4578 return 0;
4582 * This is it. Find all the controllers and register them. I really hate
4583 * stealing all these major device numbers.
4584 * returns the number of block devices registered.
4586 static int __devinit cciss_init_one(struct pci_dev *pdev,
4587 const struct pci_device_id *ent)
4589 int i;
4590 int j = 0;
4591 int k = 0;
4592 int rc;
4593 int dac, return_code;
4594 InquiryData_struct *inq_buff;
4595 ctlr_info_t *h;
4597 rc = cciss_init_reset_devices(pdev);
4598 if (rc)
4599 return rc;
4600 i = alloc_cciss_hba(pdev);
4601 if (i < 0)
4602 return -1;
4604 h = hba[i];
4605 h->pdev = pdev;
4606 h->busy_initializing = 1;
4607 INIT_LIST_HEAD(&h->cmpQ);
4608 INIT_LIST_HEAD(&h->reqQ);
4609 mutex_init(&h->busy_shutting_down);
4611 if (cciss_pci_init(h) != 0)
4612 goto clean_no_release_regions;
4614 sprintf(h->devname, "cciss%d", i);
4615 h->ctlr = i;
4617 init_completion(&h->scan_wait);
4619 if (cciss_create_hba_sysfs_entry(h))
4620 goto clean0;
4622 /* configure PCI DMA stuff */
4623 if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
4624 dac = 1;
4625 else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
4626 dac = 0;
4627 else {
4628 dev_err(&h->pdev->dev, "no suitable DMA available\n");
4629 goto clean1;
4633 * register with the major number, or get a dynamic major number
4634 * by passing 0 as argument. This is done for greater than
4635 * 8 controller support.
4637 if (i < MAX_CTLR_ORIG)
4638 h->major = COMPAQ_CISS_MAJOR + i;
4639 rc = register_blkdev(h->major, h->devname);
4640 if (rc == -EBUSY || rc == -EINVAL) {
4641 dev_err(&h->pdev->dev,
4642 "Unable to get major number %d for %s "
4643 "on hba %d\n", h->major, h->devname, i);
4644 goto clean1;
4645 } else {
4646 if (i >= MAX_CTLR_ORIG)
4647 h->major = rc;
4650 /* make sure the board interrupts are off */
4651 h->access.set_intr_mask(h, CCISS_INTR_OFF);
4652 if (h->msi_vector || h->msix_vector) {
4653 if (request_irq(h->intr[PERF_MODE_INT],
4654 do_cciss_msix_intr,
4655 IRQF_DISABLED, h->devname, h)) {
4656 dev_err(&h->pdev->dev, "Unable to get irq %d for %s\n",
4657 h->intr[PERF_MODE_INT], h->devname);
4658 goto clean2;
4660 } else {
4661 if (request_irq(h->intr[PERF_MODE_INT], do_cciss_intx,
4662 IRQF_DISABLED, h->devname, h)) {
4663 dev_err(&h->pdev->dev, "Unable to get irq %d for %s\n",
4664 h->intr[PERF_MODE_INT], h->devname);
4665 goto clean2;
4669 dev_info(&h->pdev->dev, "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
4670 h->devname, pdev->device, pci_name(pdev),
4671 h->intr[PERF_MODE_INT], dac ? "" : " not");
4673 h->cmd_pool_bits =
4674 kmalloc(DIV_ROUND_UP(h->nr_cmds, BITS_PER_LONG)
4675 * sizeof(unsigned long), GFP_KERNEL);
4676 h->cmd_pool = (CommandList_struct *)
4677 pci_alloc_consistent(h->pdev,
4678 h->nr_cmds * sizeof(CommandList_struct),
4679 &(h->cmd_pool_dhandle));
4680 h->errinfo_pool = (ErrorInfo_struct *)
4681 pci_alloc_consistent(h->pdev,
4682 h->nr_cmds * sizeof(ErrorInfo_struct),
4683 &(h->errinfo_pool_dhandle));
4684 if ((h->cmd_pool_bits == NULL)
4685 || (h->cmd_pool == NULL)
4686 || (h->errinfo_pool == NULL)) {
4687 dev_err(&h->pdev->dev, "out of memory");
4688 goto clean4;
4691 /* Need space for temp scatter list */
4692 h->scatter_list = kmalloc(h->max_commands *
4693 sizeof(struct scatterlist *),
4694 GFP_KERNEL);
4695 if (!h->scatter_list)
4696 goto clean4;
4698 for (k = 0; k < h->nr_cmds; k++) {
4699 h->scatter_list[k] = kmalloc(sizeof(struct scatterlist) *
4700 h->maxsgentries,
4701 GFP_KERNEL);
4702 if (h->scatter_list[k] == NULL) {
4703 dev_err(&h->pdev->dev,
4704 "could not allocate s/g lists\n");
4705 goto clean4;
4708 h->cmd_sg_list = cciss_allocate_sg_chain_blocks(h,
4709 h->chainsize, h->nr_cmds);
4710 if (!h->cmd_sg_list && h->chainsize > 0)
4711 goto clean4;
4713 spin_lock_init(&h->lock);
4715 /* Initialize the pdev driver private data.
4716 have it point to h. */
4717 pci_set_drvdata(pdev, h);
4718 /* command and error info recs zeroed out before
4719 they are used */
4720 memset(h->cmd_pool_bits, 0,
4721 DIV_ROUND_UP(h->nr_cmds, BITS_PER_LONG)
4722 * sizeof(unsigned long));
4724 h->num_luns = 0;
4725 h->highest_lun = -1;
4726 for (j = 0; j < CISS_MAX_LUN; j++) {
4727 h->drv[j] = NULL;
4728 h->gendisk[j] = NULL;
4731 cciss_scsi_setup(h);
4733 /* Turn the interrupts on so we can service requests */
4734 h->access.set_intr_mask(h, CCISS_INTR_ON);
4736 /* Get the firmware version */
4737 inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
4738 if (inq_buff == NULL) {
4739 dev_err(&h->pdev->dev, "out of memory\n");
4740 goto clean4;
4743 return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
4744 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
4745 if (return_code == IO_OK) {
4746 h->firm_ver[0] = inq_buff->data_byte[32];
4747 h->firm_ver[1] = inq_buff->data_byte[33];
4748 h->firm_ver[2] = inq_buff->data_byte[34];
4749 h->firm_ver[3] = inq_buff->data_byte[35];
4750 } else { /* send command failed */
4751 dev_warn(&h->pdev->dev, "unable to determine firmware"
4752 " version of controller\n");
4754 kfree(inq_buff);
4756 cciss_procinit(h);
4758 h->cciss_max_sectors = 8192;
4760 rebuild_lun_table(h, 1, 0);
4761 h->busy_initializing = 0;
4762 return 1;
4764 clean4:
4765 kfree(h->cmd_pool_bits);
4766 /* Free up sg elements */
4767 for (k-- ; k >= 0; k--)
4768 kfree(h->scatter_list[k]);
4769 kfree(h->scatter_list);
4770 cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
4771 if (h->cmd_pool)
4772 pci_free_consistent(h->pdev,
4773 h->nr_cmds * sizeof(CommandList_struct),
4774 h->cmd_pool, h->cmd_pool_dhandle);
4775 if (h->errinfo_pool)
4776 pci_free_consistent(h->pdev,
4777 h->nr_cmds * sizeof(ErrorInfo_struct),
4778 h->errinfo_pool,
4779 h->errinfo_pool_dhandle);
4780 free_irq(h->intr[PERF_MODE_INT], h);
4781 clean2:
4782 unregister_blkdev(h->major, h->devname);
4783 clean1:
4784 cciss_destroy_hba_sysfs_entry(h);
4785 clean0:
4786 pci_release_regions(pdev);
4787 clean_no_release_regions:
4788 h->busy_initializing = 0;
4791 * Deliberately omit pci_disable_device(): it does something nasty to
4792 * Smart Array controllers that pci_enable_device does not undo
4794 pci_set_drvdata(pdev, NULL);
4795 free_hba(h);
4796 return -1;
4799 static void cciss_shutdown(struct pci_dev *pdev)
4801 ctlr_info_t *h;
4802 char *flush_buf;
4803 int return_code;
4805 h = pci_get_drvdata(pdev);
4806 flush_buf = kzalloc(4, GFP_KERNEL);
4807 if (!flush_buf) {
4808 dev_warn(&h->pdev->dev, "cache not flushed, out of memory.\n");
4809 return;
4811 /* write all data in the battery backed cache to disk */
4812 memset(flush_buf, 0, 4);
4813 return_code = sendcmd_withirq(h, CCISS_CACHE_FLUSH, flush_buf,
4814 4, 0, CTLR_LUNID, TYPE_CMD);
4815 kfree(flush_buf);
4816 if (return_code != IO_OK)
4817 dev_warn(&h->pdev->dev, "Error flushing cache\n");
4818 h->access.set_intr_mask(h, CCISS_INTR_OFF);
4819 free_irq(h->intr[PERF_MODE_INT], h);
4822 static void __devexit cciss_remove_one(struct pci_dev *pdev)
4824 ctlr_info_t *h;
4825 int i, j;
4827 if (pci_get_drvdata(pdev) == NULL) {
4828 dev_err(&pdev->dev, "Unable to remove device\n");
4829 return;
4832 h = pci_get_drvdata(pdev);
4833 i = h->ctlr;
4834 if (hba[i] == NULL) {
4835 dev_err(&pdev->dev, "device appears to already be removed\n");
4836 return;
4839 mutex_lock(&h->busy_shutting_down);
4841 remove_from_scan_list(h);
4842 remove_proc_entry(h->devname, proc_cciss);
4843 unregister_blkdev(h->major, h->devname);
4845 /* remove it from the disk list */
4846 for (j = 0; j < CISS_MAX_LUN; j++) {
4847 struct gendisk *disk = h->gendisk[j];
4848 if (disk) {
4849 struct request_queue *q = disk->queue;
4851 if (disk->flags & GENHD_FL_UP) {
4852 cciss_destroy_ld_sysfs_entry(h, j, 1);
4853 del_gendisk(disk);
4855 if (q)
4856 blk_cleanup_queue(q);
4860 #ifdef CONFIG_CISS_SCSI_TAPE
4861 cciss_unregister_scsi(h); /* unhook from SCSI subsystem */
4862 #endif
4864 cciss_shutdown(pdev);
4866 #ifdef CONFIG_PCI_MSI
4867 if (h->msix_vector)
4868 pci_disable_msix(h->pdev);
4869 else if (h->msi_vector)
4870 pci_disable_msi(h->pdev);
4871 #endif /* CONFIG_PCI_MSI */
4873 iounmap(h->transtable);
4874 iounmap(h->cfgtable);
4875 iounmap(h->vaddr);
4877 pci_free_consistent(h->pdev, h->nr_cmds * sizeof(CommandList_struct),
4878 h->cmd_pool, h->cmd_pool_dhandle);
4879 pci_free_consistent(h->pdev, h->nr_cmds * sizeof(ErrorInfo_struct),
4880 h->errinfo_pool, h->errinfo_pool_dhandle);
4881 kfree(h->cmd_pool_bits);
4882 /* Free up sg elements */
4883 for (j = 0; j < h->nr_cmds; j++)
4884 kfree(h->scatter_list[j]);
4885 kfree(h->scatter_list);
4886 cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
4888 * Deliberately omit pci_disable_device(): it does something nasty to
4889 * Smart Array controllers that pci_enable_device does not undo
4891 pci_release_regions(pdev);
4892 pci_set_drvdata(pdev, NULL);
4893 cciss_destroy_hba_sysfs_entry(h);
4894 mutex_unlock(&h->busy_shutting_down);
4895 free_hba(h);
4898 static struct pci_driver cciss_pci_driver = {
4899 .name = "cciss",
4900 .probe = cciss_init_one,
4901 .remove = __devexit_p(cciss_remove_one),
4902 .id_table = cciss_pci_device_id, /* id_table */
4903 .shutdown = cciss_shutdown,
4907 * This is it. Register the PCI driver information for the cards we control
4908 * the OS will call our registered routines when it finds one of our cards.
4910 static int __init cciss_init(void)
4912 int err;
4915 * The hardware requires that commands are aligned on a 64-bit
4916 * boundary. Given that we use pci_alloc_consistent() to allocate an
4917 * array of them, the size must be a multiple of 8 bytes.
4919 BUILD_BUG_ON(sizeof(CommandList_struct) % COMMANDLIST_ALIGNMENT);
4920 printk(KERN_INFO DRIVER_NAME "\n");
4922 err = bus_register(&cciss_bus_type);
4923 if (err)
4924 return err;
4926 /* Start the scan thread */
4927 cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
4928 if (IS_ERR(cciss_scan_thread)) {
4929 err = PTR_ERR(cciss_scan_thread);
4930 goto err_bus_unregister;
4933 /* Register for our PCI devices */
4934 err = pci_register_driver(&cciss_pci_driver);
4935 if (err)
4936 goto err_thread_stop;
4938 return err;
4940 err_thread_stop:
4941 kthread_stop(cciss_scan_thread);
4942 err_bus_unregister:
4943 bus_unregister(&cciss_bus_type);
4945 return err;
4948 static void __exit cciss_cleanup(void)
4950 int i;
4952 pci_unregister_driver(&cciss_pci_driver);
4953 /* double check that all controller entrys have been removed */
4954 for (i = 0; i < MAX_CTLR; i++) {
4955 if (hba[i] != NULL) {
4956 dev_warn(&hba[i]->pdev->dev,
4957 "had to remove controller\n");
4958 cciss_remove_one(hba[i]->pdev);
4961 kthread_stop(cciss_scan_thread);
4962 if (proc_cciss)
4963 remove_proc_entry("driver/cciss", NULL);
4964 bus_unregister(&cciss_bus_type);
4967 module_init(cciss_init);
4968 module_exit(cciss_cleanup);