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
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>
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>
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>
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"
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},
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
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,
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[],
179 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
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,
194 static int __devinit cciss_pci_find_memory_BAR(struct pci_dev *pdev,
195 unsigned long *memory_bar);
196 static inline u32 cciss_tag_discard_error_bits(ctlr_info_t *h, u32 tag);
197 static __devinit int write_driver_ver_to_cfgtable(
198 CfgTable_struct __iomem *cfgtable);
200 /* performant mode helper functions */
201 static void calc_bucket_map(int *bucket, int num_buckets, int nsgs,
203 static void cciss_put_controller_into_performant_mode(ctlr_info_t *h);
205 #ifdef CONFIG_PROC_FS
206 static void cciss_procinit(ctlr_info_t *h);
208 static void cciss_procinit(ctlr_info_t *h)
211 #endif /* CONFIG_PROC_FS */
214 static int cciss_compat_ioctl(struct block_device *, fmode_t,
215 unsigned, unsigned long);
218 static const struct block_device_operations cciss_fops = {
219 .owner = THIS_MODULE,
220 .open = cciss_unlocked_open,
221 .release = cciss_release,
223 .getgeo = cciss_getgeo,
225 .compat_ioctl = cciss_compat_ioctl,
227 .revalidate_disk = cciss_revalidate,
230 /* set_performant_mode: Modify the tag for cciss performant
231 * set bit 0 for pull model, bits 3-1 for block fetch
234 static void set_performant_mode(ctlr_info_t *h, CommandList_struct *c)
236 if (likely(h->transMethod & CFGTBL_Trans_Performant))
237 c->busaddr |= 1 | (h->blockFetchTable[c->Header.SGList] << 1);
241 * Enqueuing and dequeuing functions for cmdlists.
243 static inline void addQ(struct list_head *list, CommandList_struct *c)
245 list_add_tail(&c->list, list);
248 static inline void removeQ(CommandList_struct *c)
251 * After kexec/dump some commands might still
252 * be in flight, which the firmware will try
253 * to complete. Resetting the firmware doesn't work
254 * with old fw revisions, so we have to mark
255 * them off as 'stale' to prevent the driver from
258 if (WARN_ON(list_empty(&c->list))) {
259 c->cmd_type = CMD_MSG_STALE;
263 list_del_init(&c->list);
266 static void enqueue_cmd_and_start_io(ctlr_info_t *h,
267 CommandList_struct *c)
270 set_performant_mode(h, c);
271 spin_lock_irqsave(&h->lock, flags);
274 if (h->Qdepth > h->maxQsinceinit)
275 h->maxQsinceinit = h->Qdepth;
277 spin_unlock_irqrestore(&h->lock, flags);
280 static void cciss_free_sg_chain_blocks(SGDescriptor_struct **cmd_sg_list,
287 for (i = 0; i < nr_cmds; i++) {
288 kfree(cmd_sg_list[i]);
289 cmd_sg_list[i] = NULL;
294 static SGDescriptor_struct **cciss_allocate_sg_chain_blocks(
295 ctlr_info_t *h, int chainsize, int nr_cmds)
298 SGDescriptor_struct **cmd_sg_list;
303 cmd_sg_list = kmalloc(sizeof(*cmd_sg_list) * nr_cmds, GFP_KERNEL);
307 /* Build up chain blocks for each command */
308 for (j = 0; j < nr_cmds; j++) {
309 /* Need a block of chainsized s/g elements. */
310 cmd_sg_list[j] = kmalloc((chainsize *
311 sizeof(*cmd_sg_list[j])), GFP_KERNEL);
312 if (!cmd_sg_list[j]) {
313 dev_err(&h->pdev->dev, "Cannot get memory "
314 "for s/g chains.\n");
320 cciss_free_sg_chain_blocks(cmd_sg_list, nr_cmds);
324 static void cciss_unmap_sg_chain_block(ctlr_info_t *h, CommandList_struct *c)
326 SGDescriptor_struct *chain_sg;
329 if (c->Header.SGTotal <= h->max_cmd_sgentries)
332 chain_sg = &c->SG[h->max_cmd_sgentries - 1];
333 temp64.val32.lower = chain_sg->Addr.lower;
334 temp64.val32.upper = chain_sg->Addr.upper;
335 pci_unmap_single(h->pdev, temp64.val, chain_sg->Len, PCI_DMA_TODEVICE);
338 static void cciss_map_sg_chain_block(ctlr_info_t *h, CommandList_struct *c,
339 SGDescriptor_struct *chain_block, int len)
341 SGDescriptor_struct *chain_sg;
344 chain_sg = &c->SG[h->max_cmd_sgentries - 1];
345 chain_sg->Ext = CCISS_SG_CHAIN;
347 temp64.val = pci_map_single(h->pdev, chain_block, len,
349 chain_sg->Addr.lower = temp64.val32.lower;
350 chain_sg->Addr.upper = temp64.val32.upper;
353 #include "cciss_scsi.c" /* For SCSI tape support */
355 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
358 #define RAID_UNKNOWN (ARRAY_SIZE(raid_label)-1)
360 #ifdef CONFIG_PROC_FS
363 * Report information about this controller.
365 #define ENG_GIG 1000000000
366 #define ENG_GIG_FACTOR (ENG_GIG/512)
367 #define ENGAGE_SCSI "engage scsi"
369 static void cciss_seq_show_header(struct seq_file *seq)
371 ctlr_info_t *h = seq->private;
373 seq_printf(seq, "%s: HP %s Controller\n"
374 "Board ID: 0x%08lx\n"
375 "Firmware Version: %c%c%c%c\n"
377 "Logical drives: %d\n"
378 "Current Q depth: %d\n"
379 "Current # commands on controller: %d\n"
380 "Max Q depth since init: %d\n"
381 "Max # commands on controller since init: %d\n"
382 "Max SG entries since init: %d\n",
385 (unsigned long)h->board_id,
386 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
387 h->firm_ver[3], (unsigned int)h->intr[PERF_MODE_INT],
389 h->Qdepth, h->commands_outstanding,
390 h->maxQsinceinit, h->max_outstanding, h->maxSG);
392 #ifdef CONFIG_CISS_SCSI_TAPE
393 cciss_seq_tape_report(seq, h);
394 #endif /* CONFIG_CISS_SCSI_TAPE */
397 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
399 ctlr_info_t *h = seq->private;
402 /* prevent displaying bogus info during configuration
403 * or deconfiguration of a logical volume
405 spin_lock_irqsave(&h->lock, flags);
406 if (h->busy_configuring) {
407 spin_unlock_irqrestore(&h->lock, flags);
408 return ERR_PTR(-EBUSY);
410 h->busy_configuring = 1;
411 spin_unlock_irqrestore(&h->lock, flags);
414 cciss_seq_show_header(seq);
419 static int cciss_seq_show(struct seq_file *seq, void *v)
421 sector_t vol_sz, vol_sz_frac;
422 ctlr_info_t *h = seq->private;
423 unsigned ctlr = h->ctlr;
425 drive_info_struct *drv = h->drv[*pos];
427 if (*pos > h->highest_lun)
430 if (drv == NULL) /* it's possible for h->drv[] to have holes. */
436 vol_sz = drv->nr_blocks;
437 vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
439 sector_div(vol_sz_frac, ENG_GIG_FACTOR);
441 if (drv->raid_level < 0 || drv->raid_level > RAID_UNKNOWN)
442 drv->raid_level = RAID_UNKNOWN;
443 seq_printf(seq, "cciss/c%dd%d:"
444 "\t%4u.%02uGB\tRAID %s\n",
445 ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
446 raid_label[drv->raid_level]);
450 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
452 ctlr_info_t *h = seq->private;
454 if (*pos > h->highest_lun)
461 static void cciss_seq_stop(struct seq_file *seq, void *v)
463 ctlr_info_t *h = seq->private;
465 /* Only reset h->busy_configuring if we succeeded in setting
466 * it during cciss_seq_start. */
467 if (v == ERR_PTR(-EBUSY))
470 h->busy_configuring = 0;
473 static const struct seq_operations cciss_seq_ops = {
474 .start = cciss_seq_start,
475 .show = cciss_seq_show,
476 .next = cciss_seq_next,
477 .stop = cciss_seq_stop,
480 static int cciss_seq_open(struct inode *inode, struct file *file)
482 int ret = seq_open(file, &cciss_seq_ops);
483 struct seq_file *seq = file->private_data;
486 seq->private = PDE(inode)->data;
492 cciss_proc_write(struct file *file, const char __user *buf,
493 size_t length, loff_t *ppos)
498 #ifndef CONFIG_CISS_SCSI_TAPE
502 if (!buf || length > PAGE_SIZE - 1)
505 buffer = (char *)__get_free_page(GFP_KERNEL);
510 if (copy_from_user(buffer, buf, length))
512 buffer[length] = '\0';
514 #ifdef CONFIG_CISS_SCSI_TAPE
515 if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
516 struct seq_file *seq = file->private_data;
517 ctlr_info_t *h = seq->private;
519 err = cciss_engage_scsi(h);
523 #endif /* CONFIG_CISS_SCSI_TAPE */
525 /* might be nice to have "disengage" too, but it's not
526 safely possible. (only 1 module use count, lock issues.) */
529 free_page((unsigned long)buffer);
533 static const struct file_operations cciss_proc_fops = {
534 .owner = THIS_MODULE,
535 .open = cciss_seq_open,
538 .release = seq_release,
539 .write = cciss_proc_write,
542 static void __devinit cciss_procinit(ctlr_info_t *h)
544 struct proc_dir_entry *pde;
546 if (proc_cciss == NULL)
547 proc_cciss = proc_mkdir("driver/cciss", NULL);
550 pde = proc_create_data(h->devname, S_IWUSR | S_IRUSR | S_IRGRP |
552 &cciss_proc_fops, h);
554 #endif /* CONFIG_PROC_FS */
556 #define MAX_PRODUCT_NAME_LEN 19
558 #define to_hba(n) container_of(n, struct ctlr_info, dev)
559 #define to_drv(n) container_of(n, drive_info_struct, dev)
561 /* List of controllers which cannot be reset on kexec with reset_devices */
562 static u32 unresettable_controller[] = {
563 0x324a103C, /* Smart Array P712m */
564 0x324b103C, /* SmartArray P711m */
565 0x3223103C, /* Smart Array P800 */
566 0x3234103C, /* Smart Array P400 */
567 0x3235103C, /* Smart Array P400i */
568 0x3211103C, /* Smart Array E200i */
569 0x3212103C, /* Smart Array E200 */
570 0x3213103C, /* Smart Array E200i */
571 0x3214103C, /* Smart Array E200i */
572 0x3215103C, /* Smart Array E200i */
573 0x3237103C, /* Smart Array E500 */
574 0x323D103C, /* Smart Array P700m */
575 0x409C0E11, /* Smart Array 6400 */
576 0x409D0E11, /* Smart Array 6400 EM */
579 static int ctlr_is_resettable(struct ctlr_info *h)
583 for (i = 0; i < ARRAY_SIZE(unresettable_controller); i++)
584 if (unresettable_controller[i] == h->board_id)
589 static ssize_t host_show_resettable(struct device *dev,
590 struct device_attribute *attr,
593 struct ctlr_info *h = to_hba(dev);
595 return snprintf(buf, 20, "%d\n", ctlr_is_resettable(h));
597 static DEVICE_ATTR(resettable, S_IRUGO, host_show_resettable, NULL);
599 static ssize_t host_store_rescan(struct device *dev,
600 struct device_attribute *attr,
601 const char *buf, size_t count)
603 struct ctlr_info *h = to_hba(dev);
606 wake_up_process(cciss_scan_thread);
607 wait_for_completion_interruptible(&h->scan_wait);
611 static DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
613 static ssize_t dev_show_unique_id(struct device *dev,
614 struct device_attribute *attr,
617 drive_info_struct *drv = to_drv(dev);
618 struct ctlr_info *h = to_hba(drv->dev.parent);
623 spin_lock_irqsave(&h->lock, flags);
624 if (h->busy_configuring)
627 memcpy(sn, drv->serial_no, sizeof(sn));
628 spin_unlock_irqrestore(&h->lock, flags);
633 return snprintf(buf, 16 * 2 + 2,
634 "%02X%02X%02X%02X%02X%02X%02X%02X"
635 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
636 sn[0], sn[1], sn[2], sn[3],
637 sn[4], sn[5], sn[6], sn[7],
638 sn[8], sn[9], sn[10], sn[11],
639 sn[12], sn[13], sn[14], sn[15]);
641 static DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
643 static ssize_t dev_show_vendor(struct device *dev,
644 struct device_attribute *attr,
647 drive_info_struct *drv = to_drv(dev);
648 struct ctlr_info *h = to_hba(drv->dev.parent);
649 char vendor[VENDOR_LEN + 1];
653 spin_lock_irqsave(&h->lock, flags);
654 if (h->busy_configuring)
657 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
658 spin_unlock_irqrestore(&h->lock, flags);
663 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
665 static DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
667 static ssize_t dev_show_model(struct device *dev,
668 struct device_attribute *attr,
671 drive_info_struct *drv = to_drv(dev);
672 struct ctlr_info *h = to_hba(drv->dev.parent);
673 char model[MODEL_LEN + 1];
677 spin_lock_irqsave(&h->lock, flags);
678 if (h->busy_configuring)
681 memcpy(model, drv->model, MODEL_LEN + 1);
682 spin_unlock_irqrestore(&h->lock, flags);
687 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
689 static DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
691 static ssize_t dev_show_rev(struct device *dev,
692 struct device_attribute *attr,
695 drive_info_struct *drv = to_drv(dev);
696 struct ctlr_info *h = to_hba(drv->dev.parent);
697 char rev[REV_LEN + 1];
701 spin_lock_irqsave(&h->lock, flags);
702 if (h->busy_configuring)
705 memcpy(rev, drv->rev, REV_LEN + 1);
706 spin_unlock_irqrestore(&h->lock, flags);
711 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
713 static DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
715 static ssize_t cciss_show_lunid(struct device *dev,
716 struct device_attribute *attr, char *buf)
718 drive_info_struct *drv = to_drv(dev);
719 struct ctlr_info *h = to_hba(drv->dev.parent);
721 unsigned char lunid[8];
723 spin_lock_irqsave(&h->lock, flags);
724 if (h->busy_configuring) {
725 spin_unlock_irqrestore(&h->lock, flags);
729 spin_unlock_irqrestore(&h->lock, flags);
732 memcpy(lunid, drv->LunID, sizeof(lunid));
733 spin_unlock_irqrestore(&h->lock, flags);
734 return snprintf(buf, 20, "0x%02x%02x%02x%02x%02x%02x%02x%02x\n",
735 lunid[0], lunid[1], lunid[2], lunid[3],
736 lunid[4], lunid[5], lunid[6], lunid[7]);
738 static DEVICE_ATTR(lunid, S_IRUGO, cciss_show_lunid, NULL);
740 static ssize_t cciss_show_raid_level(struct device *dev,
741 struct device_attribute *attr, char *buf)
743 drive_info_struct *drv = to_drv(dev);
744 struct ctlr_info *h = to_hba(drv->dev.parent);
748 spin_lock_irqsave(&h->lock, flags);
749 if (h->busy_configuring) {
750 spin_unlock_irqrestore(&h->lock, flags);
753 raid = drv->raid_level;
754 spin_unlock_irqrestore(&h->lock, flags);
755 if (raid < 0 || raid > RAID_UNKNOWN)
758 return snprintf(buf, strlen(raid_label[raid]) + 7, "RAID %s\n",
761 static DEVICE_ATTR(raid_level, S_IRUGO, cciss_show_raid_level, NULL);
763 static ssize_t cciss_show_usage_count(struct device *dev,
764 struct device_attribute *attr, char *buf)
766 drive_info_struct *drv = to_drv(dev);
767 struct ctlr_info *h = to_hba(drv->dev.parent);
771 spin_lock_irqsave(&h->lock, flags);
772 if (h->busy_configuring) {
773 spin_unlock_irqrestore(&h->lock, flags);
776 count = drv->usage_count;
777 spin_unlock_irqrestore(&h->lock, flags);
778 return snprintf(buf, 20, "%d\n", count);
780 static DEVICE_ATTR(usage_count, S_IRUGO, cciss_show_usage_count, NULL);
782 static struct attribute *cciss_host_attrs[] = {
783 &dev_attr_rescan.attr,
784 &dev_attr_resettable.attr,
788 static struct attribute_group cciss_host_attr_group = {
789 .attrs = cciss_host_attrs,
792 static const struct attribute_group *cciss_host_attr_groups[] = {
793 &cciss_host_attr_group,
797 static struct device_type cciss_host_type = {
798 .name = "cciss_host",
799 .groups = cciss_host_attr_groups,
800 .release = cciss_hba_release,
803 static struct attribute *cciss_dev_attrs[] = {
804 &dev_attr_unique_id.attr,
805 &dev_attr_model.attr,
806 &dev_attr_vendor.attr,
808 &dev_attr_lunid.attr,
809 &dev_attr_raid_level.attr,
810 &dev_attr_usage_count.attr,
814 static struct attribute_group cciss_dev_attr_group = {
815 .attrs = cciss_dev_attrs,
818 static const struct attribute_group *cciss_dev_attr_groups[] = {
819 &cciss_dev_attr_group,
823 static struct device_type cciss_dev_type = {
824 .name = "cciss_device",
825 .groups = cciss_dev_attr_groups,
826 .release = cciss_device_release,
829 static struct bus_type cciss_bus_type = {
834 * cciss_hba_release is called when the reference count
835 * of h->dev goes to zero.
837 static void cciss_hba_release(struct device *dev)
840 * nothing to do, but need this to avoid a warning
841 * about not having a release handler from lib/kref.c.
846 * Initialize sysfs entry for each controller. This sets up and registers
847 * the 'cciss#' directory for each individual controller under
848 * /sys/bus/pci/devices/<dev>/.
850 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
852 device_initialize(&h->dev);
853 h->dev.type = &cciss_host_type;
854 h->dev.bus = &cciss_bus_type;
855 dev_set_name(&h->dev, "%s", h->devname);
856 h->dev.parent = &h->pdev->dev;
858 return device_add(&h->dev);
862 * Remove sysfs entries for an hba.
864 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
867 put_device(&h->dev); /* final put. */
870 /* cciss_device_release is called when the reference count
871 * of h->drv[x]dev goes to zero.
873 static void cciss_device_release(struct device *dev)
875 drive_info_struct *drv = to_drv(dev);
880 * Initialize sysfs for each logical drive. This sets up and registers
881 * the 'c#d#' directory for each individual logical drive under
882 * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
883 * /sys/block/cciss!c#d# to this entry.
885 static long cciss_create_ld_sysfs_entry(struct ctlr_info *h,
890 if (h->drv[drv_index]->device_initialized)
893 dev = &h->drv[drv_index]->dev;
894 device_initialize(dev);
895 dev->type = &cciss_dev_type;
896 dev->bus = &cciss_bus_type;
897 dev_set_name(dev, "c%dd%d", h->ctlr, drv_index);
898 dev->parent = &h->dev;
899 h->drv[drv_index]->device_initialized = 1;
900 return device_add(dev);
904 * Remove sysfs entries for a logical drive.
906 static void cciss_destroy_ld_sysfs_entry(struct ctlr_info *h, int drv_index,
909 struct device *dev = &h->drv[drv_index]->dev;
911 /* special case for c*d0, we only destroy it on controller exit */
912 if (drv_index == 0 && !ctlr_exiting)
916 put_device(dev); /* the "final" put. */
917 h->drv[drv_index] = NULL;
921 * For operations that cannot sleep, a command block is allocated at init,
922 * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
923 * which ones are free or in use.
925 static CommandList_struct *cmd_alloc(ctlr_info_t *h)
927 CommandList_struct *c;
930 dma_addr_t cmd_dma_handle, err_dma_handle;
933 i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
936 } while (test_and_set_bit(i & (BITS_PER_LONG - 1),
937 h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
939 memset(c, 0, sizeof(CommandList_struct));
940 cmd_dma_handle = h->cmd_pool_dhandle + i * sizeof(CommandList_struct);
941 c->err_info = h->errinfo_pool + i;
942 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
943 err_dma_handle = h->errinfo_pool_dhandle
944 + i * sizeof(ErrorInfo_struct);
949 INIT_LIST_HEAD(&c->list);
950 c->busaddr = (__u32) cmd_dma_handle;
951 temp64.val = (__u64) err_dma_handle;
952 c->ErrDesc.Addr.lower = temp64.val32.lower;
953 c->ErrDesc.Addr.upper = temp64.val32.upper;
954 c->ErrDesc.Len = sizeof(ErrorInfo_struct);
960 /* allocate a command using pci_alloc_consistent, used for ioctls,
961 * etc., not for the main i/o path.
963 static CommandList_struct *cmd_special_alloc(ctlr_info_t *h)
965 CommandList_struct *c;
967 dma_addr_t cmd_dma_handle, err_dma_handle;
969 c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
970 sizeof(CommandList_struct), &cmd_dma_handle);
973 memset(c, 0, sizeof(CommandList_struct));
977 c->err_info = (ErrorInfo_struct *)
978 pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
981 if (c->err_info == NULL) {
982 pci_free_consistent(h->pdev,
983 sizeof(CommandList_struct), c, cmd_dma_handle);
986 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
988 INIT_LIST_HEAD(&c->list);
989 c->busaddr = (__u32) cmd_dma_handle;
990 temp64.val = (__u64) err_dma_handle;
991 c->ErrDesc.Addr.lower = temp64.val32.lower;
992 c->ErrDesc.Addr.upper = temp64.val32.upper;
993 c->ErrDesc.Len = sizeof(ErrorInfo_struct);
999 static void cmd_free(ctlr_info_t *h, CommandList_struct *c)
1003 i = c - h->cmd_pool;
1004 clear_bit(i & (BITS_PER_LONG - 1),
1005 h->cmd_pool_bits + (i / BITS_PER_LONG));
1009 static void cmd_special_free(ctlr_info_t *h, CommandList_struct *c)
1013 temp64.val32.lower = c->ErrDesc.Addr.lower;
1014 temp64.val32.upper = c->ErrDesc.Addr.upper;
1015 pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
1016 c->err_info, (dma_addr_t) temp64.val);
1017 pci_free_consistent(h->pdev, sizeof(CommandList_struct), c,
1018 (dma_addr_t) cciss_tag_discard_error_bits(h, (u32) c->busaddr));
1021 static inline ctlr_info_t *get_host(struct gendisk *disk)
1023 return disk->queue->queuedata;
1026 static inline drive_info_struct *get_drv(struct gendisk *disk)
1028 return disk->private_data;
1032 * Open. Make sure the device is really there.
1034 static int cciss_open(struct block_device *bdev, fmode_t mode)
1036 ctlr_info_t *h = get_host(bdev->bd_disk);
1037 drive_info_struct *drv = get_drv(bdev->bd_disk);
1039 dev_dbg(&h->pdev->dev, "cciss_open %s\n", bdev->bd_disk->disk_name);
1040 if (drv->busy_configuring)
1043 * Root is allowed to open raw volume zero even if it's not configured
1044 * so array config can still work. Root is also allowed to open any
1045 * volume that has a LUN ID, so it can issue IOCTL to reread the
1046 * disk information. I don't think I really like this
1047 * but I'm already using way to many device nodes to claim another one
1048 * for "raw controller".
1050 if (drv->heads == 0) {
1051 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
1052 /* if not node 0 make sure it is a partition = 0 */
1053 if (MINOR(bdev->bd_dev) & 0x0f) {
1055 /* if it is, make sure we have a LUN ID */
1056 } else if (memcmp(drv->LunID, CTLR_LUNID,
1057 sizeof(drv->LunID))) {
1061 if (!capable(CAP_SYS_ADMIN))
1069 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode)
1073 mutex_lock(&cciss_mutex);
1074 ret = cciss_open(bdev, mode);
1075 mutex_unlock(&cciss_mutex);
1081 * Close. Sync first.
1083 static int cciss_release(struct gendisk *disk, fmode_t mode)
1086 drive_info_struct *drv;
1088 mutex_lock(&cciss_mutex);
1090 drv = get_drv(disk);
1091 dev_dbg(&h->pdev->dev, "cciss_release %s\n", disk->disk_name);
1094 mutex_unlock(&cciss_mutex);
1098 static int do_ioctl(struct block_device *bdev, fmode_t mode,
1099 unsigned cmd, unsigned long arg)
1102 mutex_lock(&cciss_mutex);
1103 ret = cciss_ioctl(bdev, mode, cmd, arg);
1104 mutex_unlock(&cciss_mutex);
1108 #ifdef CONFIG_COMPAT
1110 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1111 unsigned cmd, unsigned long arg);
1112 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1113 unsigned cmd, unsigned long arg);
1115 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
1116 unsigned cmd, unsigned long arg)
1119 case CCISS_GETPCIINFO:
1120 case CCISS_GETINTINFO:
1121 case CCISS_SETINTINFO:
1122 case CCISS_GETNODENAME:
1123 case CCISS_SETNODENAME:
1124 case CCISS_GETHEARTBEAT:
1125 case CCISS_GETBUSTYPES:
1126 case CCISS_GETFIRMVER:
1127 case CCISS_GETDRIVVER:
1128 case CCISS_REVALIDVOLS:
1129 case CCISS_DEREGDISK:
1130 case CCISS_REGNEWDISK:
1132 case CCISS_RESCANDISK:
1133 case CCISS_GETLUNINFO:
1134 return do_ioctl(bdev, mode, cmd, arg);
1136 case CCISS_PASSTHRU32:
1137 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
1138 case CCISS_BIG_PASSTHRU32:
1139 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
1142 return -ENOIOCTLCMD;
1146 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1147 unsigned cmd, unsigned long arg)
1149 IOCTL32_Command_struct __user *arg32 =
1150 (IOCTL32_Command_struct __user *) arg;
1151 IOCTL_Command_struct arg64;
1152 IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
1158 copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1159 sizeof(arg64.LUN_info));
1161 copy_from_user(&arg64.Request, &arg32->Request,
1162 sizeof(arg64.Request));
1164 copy_from_user(&arg64.error_info, &arg32->error_info,
1165 sizeof(arg64.error_info));
1166 err |= get_user(arg64.buf_size, &arg32->buf_size);
1167 err |= get_user(cp, &arg32->buf);
1168 arg64.buf = compat_ptr(cp);
1169 err |= copy_to_user(p, &arg64, sizeof(arg64));
1174 err = do_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
1178 copy_in_user(&arg32->error_info, &p->error_info,
1179 sizeof(arg32->error_info));
1185 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1186 unsigned cmd, unsigned long arg)
1188 BIG_IOCTL32_Command_struct __user *arg32 =
1189 (BIG_IOCTL32_Command_struct __user *) arg;
1190 BIG_IOCTL_Command_struct arg64;
1191 BIG_IOCTL_Command_struct __user *p =
1192 compat_alloc_user_space(sizeof(arg64));
1196 memset(&arg64, 0, sizeof(arg64));
1199 copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1200 sizeof(arg64.LUN_info));
1202 copy_from_user(&arg64.Request, &arg32->Request,
1203 sizeof(arg64.Request));
1205 copy_from_user(&arg64.error_info, &arg32->error_info,
1206 sizeof(arg64.error_info));
1207 err |= get_user(arg64.buf_size, &arg32->buf_size);
1208 err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1209 err |= get_user(cp, &arg32->buf);
1210 arg64.buf = compat_ptr(cp);
1211 err |= copy_to_user(p, &arg64, sizeof(arg64));
1216 err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1220 copy_in_user(&arg32->error_info, &p->error_info,
1221 sizeof(arg32->error_info));
1228 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1230 drive_info_struct *drv = get_drv(bdev->bd_disk);
1232 if (!drv->cylinders)
1235 geo->heads = drv->heads;
1236 geo->sectors = drv->sectors;
1237 geo->cylinders = drv->cylinders;
1241 static void check_ioctl_unit_attention(ctlr_info_t *h, CommandList_struct *c)
1243 if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1244 c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1245 (void)check_for_unit_attention(h, c);
1248 static int cciss_getpciinfo(ctlr_info_t *h, void __user *argp)
1250 cciss_pci_info_struct pciinfo;
1254 pciinfo.domain = pci_domain_nr(h->pdev->bus);
1255 pciinfo.bus = h->pdev->bus->number;
1256 pciinfo.dev_fn = h->pdev->devfn;
1257 pciinfo.board_id = h->board_id;
1258 if (copy_to_user(argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1263 static int cciss_getintinfo(ctlr_info_t *h, void __user *argp)
1265 cciss_coalint_struct intinfo;
1269 intinfo.delay = readl(&h->cfgtable->HostWrite.CoalIntDelay);
1270 intinfo.count = readl(&h->cfgtable->HostWrite.CoalIntCount);
1272 (argp, &intinfo, sizeof(cciss_coalint_struct)))
1277 static int cciss_setintinfo(ctlr_info_t *h, void __user *argp)
1279 cciss_coalint_struct intinfo;
1280 unsigned long flags;
1285 if (!capable(CAP_SYS_ADMIN))
1287 if (copy_from_user(&intinfo, argp, sizeof(intinfo)))
1289 if ((intinfo.delay == 0) && (intinfo.count == 0))
1291 spin_lock_irqsave(&h->lock, flags);
1292 /* Update the field, and then ring the doorbell */
1293 writel(intinfo.delay, &(h->cfgtable->HostWrite.CoalIntDelay));
1294 writel(intinfo.count, &(h->cfgtable->HostWrite.CoalIntCount));
1295 writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1297 for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1298 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
1300 udelay(1000); /* delay and try again */
1302 spin_unlock_irqrestore(&h->lock, flags);
1303 if (i >= MAX_IOCTL_CONFIG_WAIT)
1308 static int cciss_getnodename(ctlr_info_t *h, void __user *argp)
1310 NodeName_type NodeName;
1315 for (i = 0; i < 16; i++)
1316 NodeName[i] = readb(&h->cfgtable->ServerName[i]);
1317 if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1322 static int cciss_setnodename(ctlr_info_t *h, void __user *argp)
1324 NodeName_type NodeName;
1325 unsigned long flags;
1330 if (!capable(CAP_SYS_ADMIN))
1332 if (copy_from_user(NodeName, argp, sizeof(NodeName_type)))
1334 spin_lock_irqsave(&h->lock, flags);
1335 /* Update the field, and then ring the doorbell */
1336 for (i = 0; i < 16; i++)
1337 writeb(NodeName[i], &h->cfgtable->ServerName[i]);
1338 writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1339 for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1340 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
1342 udelay(1000); /* delay and try again */
1344 spin_unlock_irqrestore(&h->lock, flags);
1345 if (i >= MAX_IOCTL_CONFIG_WAIT)
1350 static int cciss_getheartbeat(ctlr_info_t *h, void __user *argp)
1352 Heartbeat_type heartbeat;
1356 heartbeat = readl(&h->cfgtable->HeartBeat);
1357 if (copy_to_user(argp, &heartbeat, sizeof(Heartbeat_type)))
1362 static int cciss_getbustypes(ctlr_info_t *h, void __user *argp)
1364 BusTypes_type BusTypes;
1368 BusTypes = readl(&h->cfgtable->BusTypes);
1369 if (copy_to_user(argp, &BusTypes, sizeof(BusTypes_type)))
1374 static int cciss_getfirmver(ctlr_info_t *h, void __user *argp)
1376 FirmwareVer_type firmware;
1380 memcpy(firmware, h->firm_ver, 4);
1383 (argp, firmware, sizeof(FirmwareVer_type)))
1388 static int cciss_getdrivver(ctlr_info_t *h, void __user *argp)
1390 DriverVer_type DriverVer = DRIVER_VERSION;
1394 if (copy_to_user(argp, &DriverVer, sizeof(DriverVer_type)))
1399 static int cciss_getluninfo(ctlr_info_t *h,
1400 struct gendisk *disk, void __user *argp)
1402 LogvolInfo_struct luninfo;
1403 drive_info_struct *drv = get_drv(disk);
1407 memcpy(&luninfo.LunID, drv->LunID, sizeof(luninfo.LunID));
1408 luninfo.num_opens = drv->usage_count;
1409 luninfo.num_parts = 0;
1410 if (copy_to_user(argp, &luninfo, sizeof(LogvolInfo_struct)))
1415 static int cciss_passthru(ctlr_info_t *h, void __user *argp)
1417 IOCTL_Command_struct iocommand;
1418 CommandList_struct *c;
1421 DECLARE_COMPLETION_ONSTACK(wait);
1426 if (!capable(CAP_SYS_RAWIO))
1430 (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1432 if ((iocommand.buf_size < 1) &&
1433 (iocommand.Request.Type.Direction != XFER_NONE)) {
1436 if (iocommand.buf_size > 0) {
1437 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1441 if (iocommand.Request.Type.Direction == XFER_WRITE) {
1442 /* Copy the data into the buffer we created */
1443 if (copy_from_user(buff, iocommand.buf, iocommand.buf_size)) {
1448 memset(buff, 0, iocommand.buf_size);
1450 c = cmd_special_alloc(h);
1455 /* Fill in the command type */
1456 c->cmd_type = CMD_IOCTL_PEND;
1457 /* Fill in Command Header */
1458 c->Header.ReplyQueue = 0; /* unused in simple mode */
1459 if (iocommand.buf_size > 0) { /* buffer to fill */
1460 c->Header.SGList = 1;
1461 c->Header.SGTotal = 1;
1462 } else { /* no buffers to fill */
1463 c->Header.SGList = 0;
1464 c->Header.SGTotal = 0;
1466 c->Header.LUN = iocommand.LUN_info;
1467 /* use the kernel address the cmd block for tag */
1468 c->Header.Tag.lower = c->busaddr;
1470 /* Fill in Request block */
1471 c->Request = iocommand.Request;
1473 /* Fill in the scatter gather information */
1474 if (iocommand.buf_size > 0) {
1475 temp64.val = pci_map_single(h->pdev, buff,
1476 iocommand.buf_size, PCI_DMA_BIDIRECTIONAL);
1477 c->SG[0].Addr.lower = temp64.val32.lower;
1478 c->SG[0].Addr.upper = temp64.val32.upper;
1479 c->SG[0].Len = iocommand.buf_size;
1480 c->SG[0].Ext = 0; /* we are not chaining */
1484 enqueue_cmd_and_start_io(h, c);
1485 wait_for_completion(&wait);
1487 /* unlock the buffers from DMA */
1488 temp64.val32.lower = c->SG[0].Addr.lower;
1489 temp64.val32.upper = c->SG[0].Addr.upper;
1490 pci_unmap_single(h->pdev, (dma_addr_t) temp64.val, iocommand.buf_size,
1491 PCI_DMA_BIDIRECTIONAL);
1492 check_ioctl_unit_attention(h, c);
1494 /* Copy the error information out */
1495 iocommand.error_info = *(c->err_info);
1496 if (copy_to_user(argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1498 cmd_special_free(h, c);
1502 if (iocommand.Request.Type.Direction == XFER_READ) {
1503 /* Copy the data out of the buffer we created */
1504 if (copy_to_user(iocommand.buf, buff, iocommand.buf_size)) {
1506 cmd_special_free(h, c);
1511 cmd_special_free(h, c);
1515 static int cciss_bigpassthru(ctlr_info_t *h, void __user *argp)
1517 BIG_IOCTL_Command_struct *ioc;
1518 CommandList_struct *c;
1519 unsigned char **buff = NULL;
1520 int *buff_size = NULL;
1525 DECLARE_COMPLETION_ONSTACK(wait);
1528 BYTE __user *data_ptr;
1532 if (!capable(CAP_SYS_RAWIO))
1534 ioc = kmalloc(sizeof(*ioc), GFP_KERNEL);
1539 if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1543 if ((ioc->buf_size < 1) &&
1544 (ioc->Request.Type.Direction != XFER_NONE)) {
1548 /* Check kmalloc limits using all SGs */
1549 if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1553 if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1557 buff = kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1562 buff_size = kmalloc(MAXSGENTRIES * sizeof(int), GFP_KERNEL);
1567 left = ioc->buf_size;
1568 data_ptr = ioc->buf;
1570 sz = (left > ioc->malloc_size) ? ioc->malloc_size : left;
1571 buff_size[sg_used] = sz;
1572 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1573 if (buff[sg_used] == NULL) {
1577 if (ioc->Request.Type.Direction == XFER_WRITE) {
1578 if (copy_from_user(buff[sg_used], data_ptr, sz)) {
1583 memset(buff[sg_used], 0, sz);
1589 c = cmd_special_alloc(h);
1594 c->cmd_type = CMD_IOCTL_PEND;
1595 c->Header.ReplyQueue = 0;
1596 c->Header.SGList = sg_used;
1597 c->Header.SGTotal = sg_used;
1598 c->Header.LUN = ioc->LUN_info;
1599 c->Header.Tag.lower = c->busaddr;
1601 c->Request = ioc->Request;
1602 for (i = 0; i < sg_used; i++) {
1603 temp64.val = pci_map_single(h->pdev, buff[i], buff_size[i],
1604 PCI_DMA_BIDIRECTIONAL);
1605 c->SG[i].Addr.lower = temp64.val32.lower;
1606 c->SG[i].Addr.upper = temp64.val32.upper;
1607 c->SG[i].Len = buff_size[i];
1608 c->SG[i].Ext = 0; /* we are not chaining */
1611 enqueue_cmd_and_start_io(h, c);
1612 wait_for_completion(&wait);
1613 /* unlock the buffers from DMA */
1614 for (i = 0; i < sg_used; i++) {
1615 temp64.val32.lower = c->SG[i].Addr.lower;
1616 temp64.val32.upper = c->SG[i].Addr.upper;
1617 pci_unmap_single(h->pdev,
1618 (dma_addr_t) temp64.val, buff_size[i],
1619 PCI_DMA_BIDIRECTIONAL);
1621 check_ioctl_unit_attention(h, c);
1622 /* Copy the error information out */
1623 ioc->error_info = *(c->err_info);
1624 if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1625 cmd_special_free(h, c);
1629 if (ioc->Request.Type.Direction == XFER_READ) {
1630 /* Copy the data out of the buffer we created */
1631 BYTE __user *ptr = ioc->buf;
1632 for (i = 0; i < sg_used; i++) {
1633 if (copy_to_user(ptr, buff[i], buff_size[i])) {
1634 cmd_special_free(h, c);
1638 ptr += buff_size[i];
1641 cmd_special_free(h, c);
1645 for (i = 0; i < sg_used; i++)
1654 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1655 unsigned int cmd, unsigned long arg)
1657 struct gendisk *disk = bdev->bd_disk;
1658 ctlr_info_t *h = get_host(disk);
1659 void __user *argp = (void __user *)arg;
1661 dev_dbg(&h->pdev->dev, "cciss_ioctl: Called with cmd=%x %lx\n",
1664 case CCISS_GETPCIINFO:
1665 return cciss_getpciinfo(h, argp);
1666 case CCISS_GETINTINFO:
1667 return cciss_getintinfo(h, argp);
1668 case CCISS_SETINTINFO:
1669 return cciss_setintinfo(h, argp);
1670 case CCISS_GETNODENAME:
1671 return cciss_getnodename(h, argp);
1672 case CCISS_SETNODENAME:
1673 return cciss_setnodename(h, argp);
1674 case CCISS_GETHEARTBEAT:
1675 return cciss_getheartbeat(h, argp);
1676 case CCISS_GETBUSTYPES:
1677 return cciss_getbustypes(h, argp);
1678 case CCISS_GETFIRMVER:
1679 return cciss_getfirmver(h, argp);
1680 case CCISS_GETDRIVVER:
1681 return cciss_getdrivver(h, argp);
1682 case CCISS_DEREGDISK:
1684 case CCISS_REVALIDVOLS:
1685 return rebuild_lun_table(h, 0, 1);
1686 case CCISS_GETLUNINFO:
1687 return cciss_getluninfo(h, disk, argp);
1688 case CCISS_PASSTHRU:
1689 return cciss_passthru(h, argp);
1690 case CCISS_BIG_PASSTHRU:
1691 return cciss_bigpassthru(h, argp);
1693 /* scsi_cmd_ioctl handles these, below, though some are not */
1694 /* very meaningful for cciss. SG_IO is the main one people want. */
1696 case SG_GET_VERSION_NUM:
1697 case SG_SET_TIMEOUT:
1698 case SG_GET_TIMEOUT:
1699 case SG_GET_RESERVED_SIZE:
1700 case SG_SET_RESERVED_SIZE:
1701 case SG_EMULATED_HOST:
1703 case SCSI_IOCTL_SEND_COMMAND:
1704 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
1706 /* scsi_cmd_ioctl would normally handle these, below, but */
1707 /* they aren't a good fit for cciss, as CD-ROMs are */
1708 /* not supported, and we don't have any bus/target/lun */
1709 /* which we present to the kernel. */
1711 case CDROM_SEND_PACKET:
1712 case CDROMCLOSETRAY:
1714 case SCSI_IOCTL_GET_IDLUN:
1715 case SCSI_IOCTL_GET_BUS_NUMBER:
1721 static void cciss_check_queues(ctlr_info_t *h)
1723 int start_queue = h->next_to_run;
1726 /* check to see if we have maxed out the number of commands that can
1727 * be placed on the queue. If so then exit. We do this check here
1728 * in case the interrupt we serviced was from an ioctl and did not
1729 * free any new commands.
1731 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1734 /* We have room on the queue for more commands. Now we need to queue
1735 * them up. We will also keep track of the next queue to run so
1736 * that every queue gets a chance to be started first.
1738 for (i = 0; i < h->highest_lun + 1; i++) {
1739 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1740 /* make sure the disk has been added and the drive is real
1741 * because this can be called from the middle of init_one.
1743 if (!h->drv[curr_queue])
1745 if (!(h->drv[curr_queue]->queue) ||
1746 !(h->drv[curr_queue]->heads))
1748 blk_start_queue(h->gendisk[curr_queue]->queue);
1750 /* check to see if we have maxed out the number of commands
1751 * that can be placed on the queue.
1753 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1754 if (curr_queue == start_queue) {
1756 (start_queue + 1) % (h->highest_lun + 1);
1759 h->next_to_run = curr_queue;
1766 static void cciss_softirq_done(struct request *rq)
1768 CommandList_struct *c = rq->completion_data;
1769 ctlr_info_t *h = hba[c->ctlr];
1770 SGDescriptor_struct *curr_sg = c->SG;
1772 unsigned long flags;
1776 if (c->Request.Type.Direction == XFER_READ)
1777 ddir = PCI_DMA_FROMDEVICE;
1779 ddir = PCI_DMA_TODEVICE;
1781 /* command did not need to be retried */
1782 /* unmap the DMA mapping for all the scatter gather elements */
1783 for (i = 0; i < c->Header.SGList; i++) {
1784 if (curr_sg[sg_index].Ext == CCISS_SG_CHAIN) {
1785 cciss_unmap_sg_chain_block(h, c);
1786 /* Point to the next block */
1787 curr_sg = h->cmd_sg_list[c->cmdindex];
1790 temp64.val32.lower = curr_sg[sg_index].Addr.lower;
1791 temp64.val32.upper = curr_sg[sg_index].Addr.upper;
1792 pci_unmap_page(h->pdev, temp64.val, curr_sg[sg_index].Len,
1797 dev_dbg(&h->pdev->dev, "Done with %p\n", rq);
1799 /* set the residual count for pc requests */
1800 if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
1801 rq->resid_len = c->err_info->ResidualCnt;
1803 blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1805 spin_lock_irqsave(&h->lock, flags);
1807 cciss_check_queues(h);
1808 spin_unlock_irqrestore(&h->lock, flags);
1811 static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1812 unsigned char scsi3addr[], uint32_t log_unit)
1814 memcpy(scsi3addr, h->drv[log_unit]->LunID,
1815 sizeof(h->drv[log_unit]->LunID));
1818 /* This function gets the SCSI vendor, model, and revision of a logical drive
1819 * via the inquiry page 0. Model, vendor, and rev are set to empty strings if
1820 * they cannot be read.
1822 static void cciss_get_device_descr(ctlr_info_t *h, int logvol,
1823 char *vendor, char *model, char *rev)
1826 InquiryData_struct *inq_buf;
1827 unsigned char scsi3addr[8];
1833 inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1837 log_unit_to_scsi3addr(h, scsi3addr, logvol);
1838 rc = sendcmd_withirq(h, CISS_INQUIRY, inq_buf, sizeof(*inq_buf), 0,
1839 scsi3addr, TYPE_CMD);
1841 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1842 vendor[VENDOR_LEN] = '\0';
1843 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1844 model[MODEL_LEN] = '\0';
1845 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1846 rev[REV_LEN] = '\0';
1853 /* This function gets the serial number of a logical drive via
1854 * inquiry page 0x83. Serial no. is 16 bytes. If the serial
1855 * number cannot be had, for whatever reason, 16 bytes of 0xff
1856 * are returned instead.
1858 static void cciss_get_serial_no(ctlr_info_t *h, int logvol,
1859 unsigned char *serial_no, int buflen)
1861 #define PAGE_83_INQ_BYTES 64
1864 unsigned char scsi3addr[8];
1868 memset(serial_no, 0xff, buflen);
1869 buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1872 memset(serial_no, 0, buflen);
1873 log_unit_to_scsi3addr(h, scsi3addr, logvol);
1874 rc = sendcmd_withirq(h, CISS_INQUIRY, buf,
1875 PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1877 memcpy(serial_no, &buf[8], buflen);
1883 * cciss_add_disk sets up the block device queue for a logical drive
1885 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1888 disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1890 goto init_queue_failure;
1891 sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1892 disk->major = h->major;
1893 disk->first_minor = drv_index << NWD_SHIFT;
1894 disk->fops = &cciss_fops;
1895 if (cciss_create_ld_sysfs_entry(h, drv_index))
1897 disk->private_data = h->drv[drv_index];
1898 disk->driverfs_dev = &h->drv[drv_index]->dev;
1900 /* Set up queue information */
1901 blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1903 /* This is a hardware imposed limit. */
1904 blk_queue_max_segments(disk->queue, h->maxsgentries);
1906 blk_queue_max_hw_sectors(disk->queue, h->cciss_max_sectors);
1908 blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1910 disk->queue->queuedata = h;
1912 blk_queue_logical_block_size(disk->queue,
1913 h->drv[drv_index]->block_size);
1915 /* Make sure all queue data is written out before */
1916 /* setting h->drv[drv_index]->queue, as setting this */
1917 /* allows the interrupt handler to start the queue */
1919 h->drv[drv_index]->queue = disk->queue;
1924 blk_cleanup_queue(disk->queue);
1930 /* This function will check the usage_count of the drive to be updated/added.
1931 * If the usage_count is zero and it is a heretofore unknown drive, or,
1932 * the drive's capacity, geometry, or serial number has changed,
1933 * then the drive information will be updated and the disk will be
1934 * re-registered with the kernel. If these conditions don't hold,
1935 * then it will be left alone for the next reboot. The exception to this
1936 * is disk 0 which will always be left registered with the kernel since it
1937 * is also the controller node. Any changes to disk 0 will show up on
1940 static void cciss_update_drive_info(ctlr_info_t *h, int drv_index,
1941 int first_time, int via_ioctl)
1943 struct gendisk *disk;
1944 InquiryData_struct *inq_buff = NULL;
1945 unsigned int block_size;
1946 sector_t total_size;
1947 unsigned long flags = 0;
1949 drive_info_struct *drvinfo;
1951 /* Get information about the disk and modify the driver structure */
1952 inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1953 drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
1954 if (inq_buff == NULL || drvinfo == NULL)
1957 /* testing to see if 16-byte CDBs are already being used */
1958 if (h->cciss_read == CCISS_READ_16) {
1959 cciss_read_capacity_16(h, drv_index,
1960 &total_size, &block_size);
1963 cciss_read_capacity(h, drv_index, &total_size, &block_size);
1964 /* if read_capacity returns all F's this volume is >2TB */
1965 /* in size so we switch to 16-byte CDB's for all */
1966 /* read/write ops */
1967 if (total_size == 0xFFFFFFFFULL) {
1968 cciss_read_capacity_16(h, drv_index,
1969 &total_size, &block_size);
1970 h->cciss_read = CCISS_READ_16;
1971 h->cciss_write = CCISS_WRITE_16;
1973 h->cciss_read = CCISS_READ_10;
1974 h->cciss_write = CCISS_WRITE_10;
1978 cciss_geometry_inquiry(h, drv_index, total_size, block_size,
1980 drvinfo->block_size = block_size;
1981 drvinfo->nr_blocks = total_size + 1;
1983 cciss_get_device_descr(h, drv_index, drvinfo->vendor,
1984 drvinfo->model, drvinfo->rev);
1985 cciss_get_serial_no(h, drv_index, drvinfo->serial_no,
1986 sizeof(drvinfo->serial_no));
1987 /* Save the lunid in case we deregister the disk, below. */
1988 memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
1989 sizeof(drvinfo->LunID));
1991 /* Is it the same disk we already know, and nothing's changed? */
1992 if (h->drv[drv_index]->raid_level != -1 &&
1993 ((memcmp(drvinfo->serial_no,
1994 h->drv[drv_index]->serial_no, 16) == 0) &&
1995 drvinfo->block_size == h->drv[drv_index]->block_size &&
1996 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
1997 drvinfo->heads == h->drv[drv_index]->heads &&
1998 drvinfo->sectors == h->drv[drv_index]->sectors &&
1999 drvinfo->cylinders == h->drv[drv_index]->cylinders))
2000 /* The disk is unchanged, nothing to update */
2003 /* If we get here it's not the same disk, or something's changed,
2004 * so we need to * deregister it, and re-register it, if it's not
2006 * If the disk already exists then deregister it before proceeding
2007 * (unless it's the first disk (for the controller node).
2009 if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
2010 dev_warn(&h->pdev->dev, "disk %d has changed.\n", drv_index);
2011 spin_lock_irqsave(&h->lock, flags);
2012 h->drv[drv_index]->busy_configuring = 1;
2013 spin_unlock_irqrestore(&h->lock, flags);
2015 /* deregister_disk sets h->drv[drv_index]->queue = NULL
2016 * which keeps the interrupt handler from starting
2019 ret = deregister_disk(h, drv_index, 0, via_ioctl);
2022 /* If the disk is in use return */
2026 /* Save the new information from cciss_geometry_inquiry
2027 * and serial number inquiry. If the disk was deregistered
2028 * above, then h->drv[drv_index] will be NULL.
2030 if (h->drv[drv_index] == NULL) {
2031 drvinfo->device_initialized = 0;
2032 h->drv[drv_index] = drvinfo;
2033 drvinfo = NULL; /* so it won't be freed below. */
2035 /* special case for cxd0 */
2036 h->drv[drv_index]->block_size = drvinfo->block_size;
2037 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
2038 h->drv[drv_index]->heads = drvinfo->heads;
2039 h->drv[drv_index]->sectors = drvinfo->sectors;
2040 h->drv[drv_index]->cylinders = drvinfo->cylinders;
2041 h->drv[drv_index]->raid_level = drvinfo->raid_level;
2042 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
2043 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
2045 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
2046 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
2050 disk = h->gendisk[drv_index];
2051 set_capacity(disk, h->drv[drv_index]->nr_blocks);
2053 /* If it's not disk 0 (drv_index != 0)
2054 * or if it was disk 0, but there was previously
2055 * no actual corresponding configured logical drive
2056 * (raid_leve == -1) then we want to update the
2057 * logical drive's information.
2059 if (drv_index || first_time) {
2060 if (cciss_add_disk(h, disk, drv_index) != 0) {
2061 cciss_free_gendisk(h, drv_index);
2062 cciss_free_drive_info(h, drv_index);
2063 dev_warn(&h->pdev->dev, "could not update disk %d\n",
2074 dev_err(&h->pdev->dev, "out of memory\n");
2078 /* This function will find the first index of the controllers drive array
2079 * that has a null drv pointer and allocate the drive info struct and
2080 * will return that index This is where new drives will be added.
2081 * If the index to be returned is greater than the highest_lun index for
2082 * the controller then highest_lun is set * to this new index.
2083 * If there are no available indexes or if tha allocation fails, then -1
2084 * is returned. * "controller_node" is used to know if this is a real
2085 * logical drive, or just the controller node, which determines if this
2086 * counts towards highest_lun.
2088 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
2091 drive_info_struct *drv;
2093 /* Search for an empty slot for our drive info */
2094 for (i = 0; i < CISS_MAX_LUN; i++) {
2096 /* if not cxd0 case, and it's occupied, skip it. */
2097 if (h->drv[i] && i != 0)
2100 * If it's cxd0 case, and drv is alloc'ed already, and a
2101 * disk is configured there, skip it.
2103 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2107 * We've found an empty slot. Update highest_lun
2108 * provided this isn't just the fake cxd0 controller node.
2110 if (i > h->highest_lun && !controller_node)
2113 /* If adding a real disk at cxd0, and it's already alloc'ed */
2114 if (i == 0 && h->drv[i] != NULL)
2118 * Found an empty slot, not already alloc'ed. Allocate it.
2119 * Mark it with raid_level == -1, so we know it's new later on.
2121 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2124 drv->raid_level = -1; /* so we know it's new */
2131 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2133 kfree(h->drv[drv_index]);
2134 h->drv[drv_index] = NULL;
2137 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2139 put_disk(h->gendisk[drv_index]);
2140 h->gendisk[drv_index] = NULL;
2143 /* cciss_add_gendisk finds a free hba[]->drv structure
2144 * and allocates a gendisk if needed, and sets the lunid
2145 * in the drvinfo structure. It returns the index into
2146 * the ->drv[] array, or -1 if none are free.
2147 * is_controller_node indicates whether highest_lun should
2148 * count this disk, or if it's only being added to provide
2149 * a means to talk to the controller in case no logical
2150 * drives have yet been configured.
2152 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2153 int controller_node)
2157 drv_index = cciss_alloc_drive_info(h, controller_node);
2158 if (drv_index == -1)
2161 /*Check if the gendisk needs to be allocated */
2162 if (!h->gendisk[drv_index]) {
2163 h->gendisk[drv_index] =
2164 alloc_disk(1 << NWD_SHIFT);
2165 if (!h->gendisk[drv_index]) {
2166 dev_err(&h->pdev->dev,
2167 "could not allocate a new disk %d\n",
2169 goto err_free_drive_info;
2172 memcpy(h->drv[drv_index]->LunID, lunid,
2173 sizeof(h->drv[drv_index]->LunID));
2174 if (cciss_create_ld_sysfs_entry(h, drv_index))
2176 /* Don't need to mark this busy because nobody */
2177 /* else knows about this disk yet to contend */
2178 /* for access to it. */
2179 h->drv[drv_index]->busy_configuring = 0;
2184 cciss_free_gendisk(h, drv_index);
2185 err_free_drive_info:
2186 cciss_free_drive_info(h, drv_index);
2190 /* This is for the special case of a controller which
2191 * has no logical drives. In this case, we still need
2192 * to register a disk so the controller can be accessed
2193 * by the Array Config Utility.
2195 static void cciss_add_controller_node(ctlr_info_t *h)
2197 struct gendisk *disk;
2200 if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2203 drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2204 if (drv_index == -1)
2206 h->drv[drv_index]->block_size = 512;
2207 h->drv[drv_index]->nr_blocks = 0;
2208 h->drv[drv_index]->heads = 0;
2209 h->drv[drv_index]->sectors = 0;
2210 h->drv[drv_index]->cylinders = 0;
2211 h->drv[drv_index]->raid_level = -1;
2212 memset(h->drv[drv_index]->serial_no, 0, 16);
2213 disk = h->gendisk[drv_index];
2214 if (cciss_add_disk(h, disk, drv_index) == 0)
2216 cciss_free_gendisk(h, drv_index);
2217 cciss_free_drive_info(h, drv_index);
2219 dev_warn(&h->pdev->dev, "could not add disk 0.\n");
2223 /* This function will add and remove logical drives from the Logical
2224 * drive array of the controller and maintain persistency of ordering
2225 * so that mount points are preserved until the next reboot. This allows
2226 * for the removal of logical drives in the middle of the drive array
2227 * without a re-ordering of those drives.
2229 * h = The controller to perform the operations on
2231 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2235 ReportLunData_struct *ld_buff = NULL;
2241 unsigned char lunid[8] = CTLR_LUNID;
2242 unsigned long flags;
2244 if (!capable(CAP_SYS_RAWIO))
2247 /* Set busy_configuring flag for this operation */
2248 spin_lock_irqsave(&h->lock, flags);
2249 if (h->busy_configuring) {
2250 spin_unlock_irqrestore(&h->lock, flags);
2253 h->busy_configuring = 1;
2254 spin_unlock_irqrestore(&h->lock, flags);
2256 ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2257 if (ld_buff == NULL)
2260 return_code = sendcmd_withirq(h, CISS_REPORT_LOG, ld_buff,
2261 sizeof(ReportLunData_struct),
2262 0, CTLR_LUNID, TYPE_CMD);
2264 if (return_code == IO_OK)
2265 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2266 else { /* reading number of logical volumes failed */
2267 dev_warn(&h->pdev->dev,
2268 "report logical volume command failed\n");
2273 num_luns = listlength / 8; /* 8 bytes per entry */
2274 if (num_luns > CISS_MAX_LUN) {
2275 num_luns = CISS_MAX_LUN;
2276 dev_warn(&h->pdev->dev, "more luns configured"
2277 " on controller than can be handled by"
2282 cciss_add_controller_node(h);
2284 /* Compare controller drive array to driver's drive array
2285 * to see if any drives are missing on the controller due
2286 * to action of Array Config Utility (user deletes drive)
2287 * and deregister logical drives which have disappeared.
2289 for (i = 0; i <= h->highest_lun; i++) {
2293 /* skip holes in the array from already deleted drives */
2294 if (h->drv[i] == NULL)
2297 for (j = 0; j < num_luns; j++) {
2298 memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2299 if (memcmp(h->drv[i]->LunID, lunid,
2300 sizeof(lunid)) == 0) {
2306 /* Deregister it from the OS, it's gone. */
2307 spin_lock_irqsave(&h->lock, flags);
2308 h->drv[i]->busy_configuring = 1;
2309 spin_unlock_irqrestore(&h->lock, flags);
2310 return_code = deregister_disk(h, i, 1, via_ioctl);
2311 if (h->drv[i] != NULL)
2312 h->drv[i]->busy_configuring = 0;
2316 /* Compare controller drive array to driver's drive array.
2317 * Check for updates in the drive information and any new drives
2318 * on the controller due to ACU adding logical drives, or changing
2319 * a logical drive's size, etc. Reregister any new/changed drives
2321 for (i = 0; i < num_luns; i++) {
2326 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2327 /* Find if the LUN is already in the drive array
2328 * of the driver. If so then update its info
2329 * if not in use. If it does not exist then find
2330 * the first free index and add it.
2332 for (j = 0; j <= h->highest_lun; j++) {
2333 if (h->drv[j] != NULL &&
2334 memcmp(h->drv[j]->LunID, lunid,
2335 sizeof(h->drv[j]->LunID)) == 0) {
2342 /* check if the drive was found already in the array */
2344 drv_index = cciss_add_gendisk(h, lunid, 0);
2345 if (drv_index == -1)
2348 cciss_update_drive_info(h, drv_index, first_time, via_ioctl);
2353 h->busy_configuring = 0;
2354 /* We return -1 here to tell the ACU that we have registered/updated
2355 * all of the drives that we can and to keep it from calling us
2360 dev_err(&h->pdev->dev, "out of memory\n");
2361 h->busy_configuring = 0;
2365 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2367 /* zero out the disk size info */
2368 drive_info->nr_blocks = 0;
2369 drive_info->block_size = 0;
2370 drive_info->heads = 0;
2371 drive_info->sectors = 0;
2372 drive_info->cylinders = 0;
2373 drive_info->raid_level = -1;
2374 memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2375 memset(drive_info->model, 0, sizeof(drive_info->model));
2376 memset(drive_info->rev, 0, sizeof(drive_info->rev));
2377 memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2379 * don't clear the LUNID though, we need to remember which
2384 /* This function will deregister the disk and it's queue from the
2385 * kernel. It must be called with the controller lock held and the
2386 * drv structures busy_configuring flag set. It's parameters are:
2388 * disk = This is the disk to be deregistered
2389 * drv = This is the drive_info_struct associated with the disk to be
2390 * deregistered. It contains information about the disk used
2392 * clear_all = This flag determines whether or not the disk information
2393 * is going to be completely cleared out and the highest_lun
2394 * reset. Sometimes we want to clear out information about
2395 * the disk in preparation for re-adding it. In this case
2396 * the highest_lun should be left unchanged and the LunID
2397 * should not be cleared.
2399 * This indicates whether we've reached this path via ioctl.
2400 * This affects the maximum usage count allowed for c0d0 to be messed with.
2401 * If this path is reached via ioctl(), then the max_usage_count will
2402 * be 1, as the process calling ioctl() has got to have the device open.
2403 * If we get here via sysfs, then the max usage count will be zero.
2405 static int deregister_disk(ctlr_info_t *h, int drv_index,
2406 int clear_all, int via_ioctl)
2409 struct gendisk *disk;
2410 drive_info_struct *drv;
2411 int recalculate_highest_lun;
2413 if (!capable(CAP_SYS_RAWIO))
2416 drv = h->drv[drv_index];
2417 disk = h->gendisk[drv_index];
2419 /* make sure logical volume is NOT is use */
2420 if (clear_all || (h->gendisk[0] == disk)) {
2421 if (drv->usage_count > via_ioctl)
2423 } else if (drv->usage_count > 0)
2426 recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2428 /* invalidate the devices and deregister the disk. If it is disk
2429 * zero do not deregister it but just zero out it's values. This
2430 * allows us to delete disk zero but keep the controller registered.
2432 if (h->gendisk[0] != disk) {
2433 struct request_queue *q = disk->queue;
2434 if (disk->flags & GENHD_FL_UP) {
2435 cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2439 blk_cleanup_queue(q);
2440 /* If clear_all is set then we are deleting the logical
2441 * drive, not just refreshing its info. For drives
2442 * other than disk 0 we will call put_disk. We do not
2443 * do this for disk 0 as we need it to be able to
2444 * configure the controller.
2447 /* This isn't pretty, but we need to find the
2448 * disk in our array and NULL our the pointer.
2449 * This is so that we will call alloc_disk if
2450 * this index is used again later.
2452 for (i=0; i < CISS_MAX_LUN; i++){
2453 if (h->gendisk[i] == disk) {
2454 h->gendisk[i] = NULL;
2461 set_capacity(disk, 0);
2462 cciss_clear_drive_info(drv);
2467 /* if it was the last disk, find the new hightest lun */
2468 if (clear_all && recalculate_highest_lun) {
2469 int newhighest = -1;
2470 for (i = 0; i <= h->highest_lun; i++) {
2471 /* if the disk has size > 0, it is available */
2472 if (h->drv[i] && h->drv[i]->heads)
2475 h->highest_lun = newhighest;
2480 static int __devinit cciss_send_reset(ctlr_info_t *h, unsigned char *scsi3addr,
2483 CommandList_struct *c;
2489 return_status = fill_cmd(h, c, CCISS_RESET_MSG, NULL, 0, 0,
2490 CTLR_LUNID, TYPE_MSG);
2491 c->Request.CDB[1] = reset_type; /* fill_cmd defaults to target reset */
2492 if (return_status != IO_OK) {
2493 cmd_special_free(h, c);
2494 return return_status;
2497 enqueue_cmd_and_start_io(h, c);
2498 /* Don't wait for completion, the reset won't complete. Don't free
2499 * the command either. This is the last command we will send before
2500 * re-initializing everything, so it doesn't matter and won't leak.
2505 static int fill_cmd(ctlr_info_t *h, CommandList_struct *c, __u8 cmd, void *buff,
2506 size_t size, __u8 page_code, unsigned char *scsi3addr,
2509 u64bit buff_dma_handle;
2512 c->cmd_type = CMD_IOCTL_PEND;
2513 c->Header.ReplyQueue = 0;
2515 c->Header.SGList = 1;
2516 c->Header.SGTotal = 1;
2518 c->Header.SGList = 0;
2519 c->Header.SGTotal = 0;
2521 c->Header.Tag.lower = c->busaddr;
2522 memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2524 c->Request.Type.Type = cmd_type;
2525 if (cmd_type == TYPE_CMD) {
2528 /* are we trying to read a vital product page */
2529 if (page_code != 0) {
2530 c->Request.CDB[1] = 0x01;
2531 c->Request.CDB[2] = page_code;
2533 c->Request.CDBLen = 6;
2534 c->Request.Type.Attribute = ATTR_SIMPLE;
2535 c->Request.Type.Direction = XFER_READ;
2536 c->Request.Timeout = 0;
2537 c->Request.CDB[0] = CISS_INQUIRY;
2538 c->Request.CDB[4] = size & 0xFF;
2540 case CISS_REPORT_LOG:
2541 case CISS_REPORT_PHYS:
2542 /* Talking to controller so It's a physical command
2543 mode = 00 target = 0. Nothing to write.
2545 c->Request.CDBLen = 12;
2546 c->Request.Type.Attribute = ATTR_SIMPLE;
2547 c->Request.Type.Direction = XFER_READ;
2548 c->Request.Timeout = 0;
2549 c->Request.CDB[0] = cmd;
2550 c->Request.CDB[6] = (size >> 24) & 0xFF; /* MSB */
2551 c->Request.CDB[7] = (size >> 16) & 0xFF;
2552 c->Request.CDB[8] = (size >> 8) & 0xFF;
2553 c->Request.CDB[9] = size & 0xFF;
2556 case CCISS_READ_CAPACITY:
2557 c->Request.CDBLen = 10;
2558 c->Request.Type.Attribute = ATTR_SIMPLE;
2559 c->Request.Type.Direction = XFER_READ;
2560 c->Request.Timeout = 0;
2561 c->Request.CDB[0] = cmd;
2563 case CCISS_READ_CAPACITY_16:
2564 c->Request.CDBLen = 16;
2565 c->Request.Type.Attribute = ATTR_SIMPLE;
2566 c->Request.Type.Direction = XFER_READ;
2567 c->Request.Timeout = 0;
2568 c->Request.CDB[0] = cmd;
2569 c->Request.CDB[1] = 0x10;
2570 c->Request.CDB[10] = (size >> 24) & 0xFF;
2571 c->Request.CDB[11] = (size >> 16) & 0xFF;
2572 c->Request.CDB[12] = (size >> 8) & 0xFF;
2573 c->Request.CDB[13] = size & 0xFF;
2574 c->Request.Timeout = 0;
2575 c->Request.CDB[0] = cmd;
2577 case CCISS_CACHE_FLUSH:
2578 c->Request.CDBLen = 12;
2579 c->Request.Type.Attribute = ATTR_SIMPLE;
2580 c->Request.Type.Direction = XFER_WRITE;
2581 c->Request.Timeout = 0;
2582 c->Request.CDB[0] = BMIC_WRITE;
2583 c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2585 case TEST_UNIT_READY:
2586 c->Request.CDBLen = 6;
2587 c->Request.Type.Attribute = ATTR_SIMPLE;
2588 c->Request.Type.Direction = XFER_NONE;
2589 c->Request.Timeout = 0;
2592 dev_warn(&h->pdev->dev, "Unknown Command 0x%c\n", cmd);
2595 } else if (cmd_type == TYPE_MSG) {
2597 case CCISS_ABORT_MSG:
2598 c->Request.CDBLen = 12;
2599 c->Request.Type.Attribute = ATTR_SIMPLE;
2600 c->Request.Type.Direction = XFER_WRITE;
2601 c->Request.Timeout = 0;
2602 c->Request.CDB[0] = cmd; /* abort */
2603 c->Request.CDB[1] = 0; /* abort a command */
2604 /* buff contains the tag of the command to abort */
2605 memcpy(&c->Request.CDB[4], buff, 8);
2607 case CCISS_RESET_MSG:
2608 c->Request.CDBLen = 16;
2609 c->Request.Type.Attribute = ATTR_SIMPLE;
2610 c->Request.Type.Direction = XFER_NONE;
2611 c->Request.Timeout = 0;
2612 memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2613 c->Request.CDB[0] = cmd; /* reset */
2614 c->Request.CDB[1] = CCISS_RESET_TYPE_TARGET;
2616 case CCISS_NOOP_MSG:
2617 c->Request.CDBLen = 1;
2618 c->Request.Type.Attribute = ATTR_SIMPLE;
2619 c->Request.Type.Direction = XFER_WRITE;
2620 c->Request.Timeout = 0;
2621 c->Request.CDB[0] = cmd;
2624 dev_warn(&h->pdev->dev,
2625 "unknown message type %d\n", cmd);
2629 dev_warn(&h->pdev->dev, "unknown command type %d\n", cmd_type);
2632 /* Fill in the scatter gather information */
2634 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2636 PCI_DMA_BIDIRECTIONAL);
2637 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2638 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2639 c->SG[0].Len = size;
2640 c->SG[0].Ext = 0; /* we are not chaining */
2645 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2647 switch (c->err_info->ScsiStatus) {
2650 case SAM_STAT_CHECK_CONDITION:
2651 switch (0xf & c->err_info->SenseInfo[2]) {
2652 case 0: return IO_OK; /* no sense */
2653 case 1: return IO_OK; /* recovered error */
2655 if (check_for_unit_attention(h, c))
2656 return IO_NEEDS_RETRY;
2657 dev_warn(&h->pdev->dev, "cmd 0x%02x "
2658 "check condition, sense key = 0x%02x\n",
2659 c->Request.CDB[0], c->err_info->SenseInfo[2]);
2663 dev_warn(&h->pdev->dev, "cmd 0x%02x"
2664 "scsi status = 0x%02x\n",
2665 c->Request.CDB[0], c->err_info->ScsiStatus);
2671 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2673 int return_status = IO_OK;
2675 if (c->err_info->CommandStatus == CMD_SUCCESS)
2678 switch (c->err_info->CommandStatus) {
2679 case CMD_TARGET_STATUS:
2680 return_status = check_target_status(h, c);
2682 case CMD_DATA_UNDERRUN:
2683 case CMD_DATA_OVERRUN:
2684 /* expected for inquiry and report lun commands */
2687 dev_warn(&h->pdev->dev, "cmd 0x%02x is "
2688 "reported invalid\n", c->Request.CDB[0]);
2689 return_status = IO_ERROR;
2691 case CMD_PROTOCOL_ERR:
2692 dev_warn(&h->pdev->dev, "cmd 0x%02x has "
2693 "protocol error\n", c->Request.CDB[0]);
2694 return_status = IO_ERROR;
2696 case CMD_HARDWARE_ERR:
2697 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2698 " hardware error\n", c->Request.CDB[0]);
2699 return_status = IO_ERROR;
2701 case CMD_CONNECTION_LOST:
2702 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2703 "connection lost\n", c->Request.CDB[0]);
2704 return_status = IO_ERROR;
2707 dev_warn(&h->pdev->dev, "cmd 0x%02x was "
2708 "aborted\n", c->Request.CDB[0]);
2709 return_status = IO_ERROR;
2711 case CMD_ABORT_FAILED:
2712 dev_warn(&h->pdev->dev, "cmd 0x%02x reports "
2713 "abort failed\n", c->Request.CDB[0]);
2714 return_status = IO_ERROR;
2716 case CMD_UNSOLICITED_ABORT:
2717 dev_warn(&h->pdev->dev, "unsolicited abort 0x%02x\n",
2719 return_status = IO_NEEDS_RETRY;
2721 case CMD_UNABORTABLE:
2722 dev_warn(&h->pdev->dev, "cmd unabortable\n");
2723 return_status = IO_ERROR;
2726 dev_warn(&h->pdev->dev, "cmd 0x%02x returned "
2727 "unknown status %x\n", c->Request.CDB[0],
2728 c->err_info->CommandStatus);
2729 return_status = IO_ERROR;
2731 return return_status;
2734 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2737 DECLARE_COMPLETION_ONSTACK(wait);
2738 u64bit buff_dma_handle;
2739 int return_status = IO_OK;
2743 enqueue_cmd_and_start_io(h, c);
2745 wait_for_completion(&wait);
2747 if (c->err_info->CommandStatus == 0 || !attempt_retry)
2750 return_status = process_sendcmd_error(h, c);
2752 if (return_status == IO_NEEDS_RETRY &&
2753 c->retry_count < MAX_CMD_RETRIES) {
2754 dev_warn(&h->pdev->dev, "retrying 0x%02x\n",
2757 /* erase the old error information */
2758 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2759 return_status = IO_OK;
2760 INIT_COMPLETION(wait);
2765 /* unlock the buffers from DMA */
2766 buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2767 buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2768 pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2769 c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2770 return return_status;
2773 static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
2774 __u8 page_code, unsigned char scsi3addr[],
2777 CommandList_struct *c;
2780 c = cmd_special_alloc(h);
2783 return_status = fill_cmd(h, c, cmd, buff, size, page_code,
2784 scsi3addr, cmd_type);
2785 if (return_status == IO_OK)
2786 return_status = sendcmd_withirq_core(h, c, 1);
2788 cmd_special_free(h, c);
2789 return return_status;
2792 static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
2793 sector_t total_size,
2794 unsigned int block_size,
2795 InquiryData_struct *inq_buff,
2796 drive_info_struct *drv)
2800 unsigned char scsi3addr[8];
2802 memset(inq_buff, 0, sizeof(InquiryData_struct));
2803 log_unit_to_scsi3addr(h, scsi3addr, logvol);
2804 return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
2805 sizeof(*inq_buff), 0xC1, scsi3addr, TYPE_CMD);
2806 if (return_code == IO_OK) {
2807 if (inq_buff->data_byte[8] == 0xFF) {
2808 dev_warn(&h->pdev->dev,
2809 "reading geometry failed, volume "
2810 "does not support reading geometry\n");
2812 drv->sectors = 32; /* Sectors per track */
2813 drv->cylinders = total_size + 1;
2814 drv->raid_level = RAID_UNKNOWN;
2816 drv->heads = inq_buff->data_byte[6];
2817 drv->sectors = inq_buff->data_byte[7];
2818 drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2819 drv->cylinders += inq_buff->data_byte[5];
2820 drv->raid_level = inq_buff->data_byte[8];
2822 drv->block_size = block_size;
2823 drv->nr_blocks = total_size + 1;
2824 t = drv->heads * drv->sectors;
2826 sector_t real_size = total_size + 1;
2827 unsigned long rem = sector_div(real_size, t);
2830 drv->cylinders = real_size;
2832 } else { /* Get geometry failed */
2833 dev_warn(&h->pdev->dev, "reading geometry failed\n");
2838 cciss_read_capacity(ctlr_info_t *h, int logvol, sector_t *total_size,
2839 unsigned int *block_size)
2841 ReadCapdata_struct *buf;
2843 unsigned char scsi3addr[8];
2845 buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2847 dev_warn(&h->pdev->dev, "out of memory\n");
2851 log_unit_to_scsi3addr(h, scsi3addr, logvol);
2852 return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY, buf,
2853 sizeof(ReadCapdata_struct), 0, scsi3addr, TYPE_CMD);
2854 if (return_code == IO_OK) {
2855 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2856 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2857 } else { /* read capacity command failed */
2858 dev_warn(&h->pdev->dev, "read capacity failed\n");
2860 *block_size = BLOCK_SIZE;
2865 static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
2866 sector_t *total_size, unsigned int *block_size)
2868 ReadCapdata_struct_16 *buf;
2870 unsigned char scsi3addr[8];
2872 buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2874 dev_warn(&h->pdev->dev, "out of memory\n");
2878 log_unit_to_scsi3addr(h, scsi3addr, logvol);
2879 return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY_16,
2880 buf, sizeof(ReadCapdata_struct_16),
2881 0, scsi3addr, TYPE_CMD);
2882 if (return_code == IO_OK) {
2883 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2884 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2885 } else { /* read capacity command failed */
2886 dev_warn(&h->pdev->dev, "read capacity failed\n");
2888 *block_size = BLOCK_SIZE;
2890 dev_info(&h->pdev->dev, " blocks= %llu block_size= %d\n",
2891 (unsigned long long)*total_size+1, *block_size);
2895 static int cciss_revalidate(struct gendisk *disk)
2897 ctlr_info_t *h = get_host(disk);
2898 drive_info_struct *drv = get_drv(disk);
2901 unsigned int block_size;
2902 sector_t total_size;
2903 InquiryData_struct *inq_buff = NULL;
2905 for (logvol = 0; logvol <= h->highest_lun; logvol++) {
2906 if (!h->drv[logvol])
2908 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2909 sizeof(drv->LunID)) == 0) {
2918 inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2919 if (inq_buff == NULL) {
2920 dev_warn(&h->pdev->dev, "out of memory\n");
2923 if (h->cciss_read == CCISS_READ_10) {
2924 cciss_read_capacity(h, logvol,
2925 &total_size, &block_size);
2927 cciss_read_capacity_16(h, logvol,
2928 &total_size, &block_size);
2930 cciss_geometry_inquiry(h, logvol, total_size, block_size,
2933 blk_queue_logical_block_size(drv->queue, drv->block_size);
2934 set_capacity(disk, drv->nr_blocks);
2941 * Map (physical) PCI mem into (virtual) kernel space
2943 static void __iomem *remap_pci_mem(ulong base, ulong size)
2945 ulong page_base = ((ulong) base) & PAGE_MASK;
2946 ulong page_offs = ((ulong) base) - page_base;
2947 void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2949 return page_remapped ? (page_remapped + page_offs) : NULL;
2953 * Takes jobs of the Q and sends them to the hardware, then puts it on
2954 * the Q to wait for completion.
2956 static void start_io(ctlr_info_t *h)
2958 CommandList_struct *c;
2960 while (!list_empty(&h->reqQ)) {
2961 c = list_entry(h->reqQ.next, CommandList_struct, list);
2962 /* can't do anything if fifo is full */
2963 if ((h->access.fifo_full(h))) {
2964 dev_warn(&h->pdev->dev, "fifo full\n");
2968 /* Get the first entry from the Request Q */
2972 /* Tell the controller execute command */
2973 h->access.submit_command(h, c);
2975 /* Put job onto the completed Q */
2980 /* Assumes that h->lock is held. */
2981 /* Zeros out the error record and then resends the command back */
2982 /* to the controller */
2983 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2985 /* erase the old error information */
2986 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2988 /* add it to software queue and then send it to the controller */
2991 if (h->Qdepth > h->maxQsinceinit)
2992 h->maxQsinceinit = h->Qdepth;
2997 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
2998 unsigned int msg_byte, unsigned int host_byte,
2999 unsigned int driver_byte)
3001 /* inverse of macros in scsi.h */
3002 return (scsi_status_byte & 0xff) |
3003 ((msg_byte & 0xff) << 8) |
3004 ((host_byte & 0xff) << 16) |
3005 ((driver_byte & 0xff) << 24);
3008 static inline int evaluate_target_status(ctlr_info_t *h,
3009 CommandList_struct *cmd, int *retry_cmd)
3011 unsigned char sense_key;
3012 unsigned char status_byte, msg_byte, host_byte, driver_byte;
3016 /* If we get in here, it means we got "target status", that is, scsi status */
3017 status_byte = cmd->err_info->ScsiStatus;
3018 driver_byte = DRIVER_OK;
3019 msg_byte = cmd->err_info->CommandStatus; /* correct? seems too device specific */
3021 if (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC)
3022 host_byte = DID_PASSTHROUGH;
3026 error_value = make_status_bytes(status_byte, msg_byte,
3027 host_byte, driver_byte);
3029 if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
3030 if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC)
3031 dev_warn(&h->pdev->dev, "cmd %p "
3032 "has SCSI Status 0x%x\n",
3033 cmd, cmd->err_info->ScsiStatus);
3037 /* check the sense key */
3038 sense_key = 0xf & cmd->err_info->SenseInfo[2];
3039 /* no status or recovered error */
3040 if (((sense_key == 0x0) || (sense_key == 0x1)) &&
3041 (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC))
3044 if (check_for_unit_attention(h, cmd)) {
3045 *retry_cmd = !(cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC);
3049 /* Not SG_IO or similar? */
3050 if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC) {
3051 if (error_value != 0)
3052 dev_warn(&h->pdev->dev, "cmd %p has CHECK CONDITION"
3053 " sense key = 0x%x\n", cmd, sense_key);
3057 /* SG_IO or similar, copy sense data back */
3058 if (cmd->rq->sense) {
3059 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
3060 cmd->rq->sense_len = cmd->err_info->SenseLen;
3061 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
3062 cmd->rq->sense_len);
3064 cmd->rq->sense_len = 0;
3069 /* checks the status of the job and calls complete buffers to mark all
3070 * buffers for the completed job. Note that this function does not need
3071 * to hold the hba/queue lock.
3073 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
3077 struct request *rq = cmd->rq;
3082 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
3084 if (cmd->err_info->CommandStatus == 0) /* no error has occurred */
3085 goto after_error_processing;
3087 switch (cmd->err_info->CommandStatus) {
3088 case CMD_TARGET_STATUS:
3089 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
3091 case CMD_DATA_UNDERRUN:
3092 if (cmd->rq->cmd_type == REQ_TYPE_FS) {
3093 dev_warn(&h->pdev->dev, "cmd %p has"
3094 " completed with data underrun "
3096 cmd->rq->resid_len = cmd->err_info->ResidualCnt;
3099 case CMD_DATA_OVERRUN:
3100 if (cmd->rq->cmd_type == REQ_TYPE_FS)
3101 dev_warn(&h->pdev->dev, "cciss: cmd %p has"
3102 " completed with data overrun "
3106 dev_warn(&h->pdev->dev, "cciss: cmd %p is "
3107 "reported invalid\n", cmd);
3108 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3109 cmd->err_info->CommandStatus, DRIVER_OK,
3110 (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3111 DID_PASSTHROUGH : DID_ERROR);
3113 case CMD_PROTOCOL_ERR:
3114 dev_warn(&h->pdev->dev, "cciss: cmd %p has "
3115 "protocol error\n", cmd);
3116 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3117 cmd->err_info->CommandStatus, DRIVER_OK,
3118 (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3119 DID_PASSTHROUGH : DID_ERROR);
3121 case CMD_HARDWARE_ERR:
3122 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3123 " hardware error\n", cmd);
3124 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3125 cmd->err_info->CommandStatus, DRIVER_OK,
3126 (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3127 DID_PASSTHROUGH : DID_ERROR);
3129 case CMD_CONNECTION_LOST:
3130 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3131 "connection lost\n", cmd);
3132 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3133 cmd->err_info->CommandStatus, DRIVER_OK,
3134 (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3135 DID_PASSTHROUGH : DID_ERROR);
3138 dev_warn(&h->pdev->dev, "cciss: cmd %p was "
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_ABORT);
3145 case CMD_ABORT_FAILED:
3146 dev_warn(&h->pdev->dev, "cciss: cmd %p reports "
3147 "abort failed\n", cmd);
3148 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3149 cmd->err_info->CommandStatus, DRIVER_OK,
3150 (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3151 DID_PASSTHROUGH : DID_ERROR);
3153 case CMD_UNSOLICITED_ABORT:
3154 dev_warn(&h->pdev->dev, "cciss%d: unsolicited "
3155 "abort %p\n", h->ctlr, cmd);
3156 if (cmd->retry_count < MAX_CMD_RETRIES) {
3158 dev_warn(&h->pdev->dev, "retrying %p\n", cmd);
3161 dev_warn(&h->pdev->dev,
3162 "%p retried too many times\n", cmd);
3163 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3164 cmd->err_info->CommandStatus, DRIVER_OK,
3165 (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3166 DID_PASSTHROUGH : DID_ABORT);
3169 dev_warn(&h->pdev->dev, "cmd %p timedout\n", cmd);
3170 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3171 cmd->err_info->CommandStatus, DRIVER_OK,
3172 (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3173 DID_PASSTHROUGH : DID_ERROR);
3175 case CMD_UNABORTABLE:
3176 dev_warn(&h->pdev->dev, "cmd %p unabortable\n", cmd);
3177 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3178 cmd->err_info->CommandStatus, DRIVER_OK,
3179 cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC ?
3180 DID_PASSTHROUGH : DID_ERROR);
3183 dev_warn(&h->pdev->dev, "cmd %p returned "
3184 "unknown status %x\n", cmd,
3185 cmd->err_info->CommandStatus);
3186 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3187 cmd->err_info->CommandStatus, DRIVER_OK,
3188 (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3189 DID_PASSTHROUGH : DID_ERROR);
3192 after_error_processing:
3194 /* We need to return this command */
3196 resend_cciss_cmd(h, cmd);
3199 cmd->rq->completion_data = cmd;
3200 blk_complete_request(cmd->rq);
3203 static inline u32 cciss_tag_contains_index(u32 tag)
3205 #define DIRECT_LOOKUP_BIT 0x10
3206 return tag & DIRECT_LOOKUP_BIT;
3209 static inline u32 cciss_tag_to_index(u32 tag)
3211 #define DIRECT_LOOKUP_SHIFT 5
3212 return tag >> DIRECT_LOOKUP_SHIFT;
3215 static inline u32 cciss_tag_discard_error_bits(ctlr_info_t *h, u32 tag)
3217 #define CCISS_PERF_ERROR_BITS ((1 << DIRECT_LOOKUP_SHIFT) - 1)
3218 #define CCISS_SIMPLE_ERROR_BITS 0x03
3219 if (likely(h->transMethod & CFGTBL_Trans_Performant))
3220 return tag & ~CCISS_PERF_ERROR_BITS;
3221 return tag & ~CCISS_SIMPLE_ERROR_BITS;
3224 static inline void cciss_mark_tag_indexed(u32 *tag)
3226 *tag |= DIRECT_LOOKUP_BIT;
3229 static inline void cciss_set_tag_index(u32 *tag, u32 index)
3231 *tag |= (index << DIRECT_LOOKUP_SHIFT);
3235 * Get a request and submit it to the controller.
3237 static void do_cciss_request(struct request_queue *q)
3239 ctlr_info_t *h = q->queuedata;
3240 CommandList_struct *c;
3243 struct request *creq;
3245 struct scatterlist *tmp_sg;
3246 SGDescriptor_struct *curr_sg;
3247 drive_info_struct *drv;
3253 creq = blk_peek_request(q);
3257 BUG_ON(creq->nr_phys_segments > h->maxsgentries);
3263 blk_start_request(creq);
3265 tmp_sg = h->scatter_list[c->cmdindex];
3266 spin_unlock_irq(q->queue_lock);
3268 c->cmd_type = CMD_RWREQ;
3271 /* fill in the request */
3272 drv = creq->rq_disk->private_data;
3273 c->Header.ReplyQueue = 0; /* unused in simple mode */
3274 /* got command from pool, so use the command block index instead */
3275 /* for direct lookups. */
3276 /* The first 2 bits are reserved for controller error reporting. */
3277 cciss_set_tag_index(&c->Header.Tag.lower, c->cmdindex);
3278 cciss_mark_tag_indexed(&c->Header.Tag.lower);
3279 memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3280 c->Request.CDBLen = 10; /* 12 byte commands not in FW yet; */
3281 c->Request.Type.Type = TYPE_CMD; /* It is a command. */
3282 c->Request.Type.Attribute = ATTR_SIMPLE;
3283 c->Request.Type.Direction =
3284 (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3285 c->Request.Timeout = 0; /* Don't time out */
3287 (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3288 start_blk = blk_rq_pos(creq);
3289 dev_dbg(&h->pdev->dev, "sector =%d nr_sectors=%d\n",
3290 (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3291 sg_init_table(tmp_sg, h->maxsgentries);
3292 seg = blk_rq_map_sg(q, creq, tmp_sg);
3294 /* get the DMA records for the setup */
3295 if (c->Request.Type.Direction == XFER_READ)
3296 dir = PCI_DMA_FROMDEVICE;
3298 dir = PCI_DMA_TODEVICE;
3304 for (i = 0; i < seg; i++) {
3305 if (((sg_index+1) == (h->max_cmd_sgentries)) &&
3306 !chained && ((seg - i) > 1)) {
3307 /* Point to next chain block. */
3308 curr_sg = h->cmd_sg_list[c->cmdindex];
3312 curr_sg[sg_index].Len = tmp_sg[i].length;
3313 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3315 tmp_sg[i].length, dir);
3316 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3317 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3318 curr_sg[sg_index].Ext = 0; /* we are not chaining */
3322 cciss_map_sg_chain_block(h, c, h->cmd_sg_list[c->cmdindex],
3323 (seg - (h->max_cmd_sgentries - 1)) *
3324 sizeof(SGDescriptor_struct));
3326 /* track how many SG entries we are using */
3330 dev_dbg(&h->pdev->dev, "Submitting %u sectors in %d segments "
3332 blk_rq_sectors(creq), seg, chained);
3334 c->Header.SGTotal = seg + chained;
3335 if (seg <= h->max_cmd_sgentries)
3336 c->Header.SGList = c->Header.SGTotal;
3338 c->Header.SGList = h->max_cmd_sgentries;
3339 set_performant_mode(h, c);
3341 if (likely(creq->cmd_type == REQ_TYPE_FS)) {
3342 if(h->cciss_read == CCISS_READ_10) {
3343 c->Request.CDB[1] = 0;
3344 c->Request.CDB[2] = (start_blk >> 24) & 0xff; /* MSB */
3345 c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3346 c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3347 c->Request.CDB[5] = start_blk & 0xff;
3348 c->Request.CDB[6] = 0; /* (sect >> 24) & 0xff; MSB */
3349 c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3350 c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3351 c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3353 u32 upper32 = upper_32_bits(start_blk);
3355 c->Request.CDBLen = 16;
3356 c->Request.CDB[1]= 0;
3357 c->Request.CDB[2]= (upper32 >> 24) & 0xff; /* MSB */
3358 c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3359 c->Request.CDB[4]= (upper32 >> 8) & 0xff;
3360 c->Request.CDB[5]= upper32 & 0xff;
3361 c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3362 c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3363 c->Request.CDB[8]= (start_blk >> 8) & 0xff;
3364 c->Request.CDB[9]= start_blk & 0xff;
3365 c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3366 c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3367 c->Request.CDB[12]= (blk_rq_sectors(creq) >> 8) & 0xff;
3368 c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3369 c->Request.CDB[14] = c->Request.CDB[15] = 0;
3371 } else if (creq->cmd_type == REQ_TYPE_BLOCK_PC) {
3372 c->Request.CDBLen = creq->cmd_len;
3373 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3375 dev_warn(&h->pdev->dev, "bad request type %d\n",
3380 spin_lock_irq(q->queue_lock);
3384 if (h->Qdepth > h->maxQsinceinit)
3385 h->maxQsinceinit = h->Qdepth;
3391 /* We will already have the driver lock here so not need
3397 static inline unsigned long get_next_completion(ctlr_info_t *h)
3399 return h->access.command_completed(h);
3402 static inline int interrupt_pending(ctlr_info_t *h)
3404 return h->access.intr_pending(h);
3407 static inline long interrupt_not_for_us(ctlr_info_t *h)
3409 return ((h->access.intr_pending(h) == 0) ||
3410 (h->interrupts_enabled == 0));
3413 static inline int bad_tag(ctlr_info_t *h, u32 tag_index,
3416 if (unlikely(tag_index >= h->nr_cmds)) {
3417 dev_warn(&h->pdev->dev, "bad tag 0x%08x ignored.\n", raw_tag);
3423 static inline void finish_cmd(ctlr_info_t *h, CommandList_struct *c,
3427 if (likely(c->cmd_type == CMD_RWREQ))
3428 complete_command(h, c, 0);
3429 else if (c->cmd_type == CMD_IOCTL_PEND)
3430 complete(c->waiting);
3431 #ifdef CONFIG_CISS_SCSI_TAPE
3432 else if (c->cmd_type == CMD_SCSI)
3433 complete_scsi_command(c, 0, raw_tag);
3437 static inline u32 next_command(ctlr_info_t *h)
3441 if (unlikely(!(h->transMethod & CFGTBL_Trans_Performant)))
3442 return h->access.command_completed(h);
3444 if ((*(h->reply_pool_head) & 1) == (h->reply_pool_wraparound)) {
3445 a = *(h->reply_pool_head); /* Next cmd in ring buffer */
3446 (h->reply_pool_head)++;
3447 h->commands_outstanding--;
3451 /* Check for wraparound */
3452 if (h->reply_pool_head == (h->reply_pool + h->max_commands)) {
3453 h->reply_pool_head = h->reply_pool;
3454 h->reply_pool_wraparound ^= 1;
3459 /* process completion of an indexed ("direct lookup") command */
3460 static inline u32 process_indexed_cmd(ctlr_info_t *h, u32 raw_tag)
3463 CommandList_struct *c;
3465 tag_index = cciss_tag_to_index(raw_tag);
3466 if (bad_tag(h, tag_index, raw_tag))
3467 return next_command(h);
3468 c = h->cmd_pool + tag_index;
3469 finish_cmd(h, c, raw_tag);
3470 return next_command(h);
3473 /* process completion of a non-indexed command */
3474 static inline u32 process_nonindexed_cmd(ctlr_info_t *h, u32 raw_tag)
3476 CommandList_struct *c = NULL;
3477 __u32 busaddr_masked, tag_masked;
3479 tag_masked = cciss_tag_discard_error_bits(h, raw_tag);
3480 list_for_each_entry(c, &h->cmpQ, list) {
3481 busaddr_masked = cciss_tag_discard_error_bits(h, c->busaddr);
3482 if (busaddr_masked == tag_masked) {
3483 finish_cmd(h, c, raw_tag);
3484 return next_command(h);
3487 bad_tag(h, h->nr_cmds + 1, raw_tag);
3488 return next_command(h);
3491 /* Some controllers, like p400, will give us one interrupt
3492 * after a soft reset, even if we turned interrupts off.
3493 * Only need to check for this in the cciss_xxx_discard_completions
3496 static int ignore_bogus_interrupt(ctlr_info_t *h)
3498 if (likely(!reset_devices))
3501 if (likely(h->interrupts_enabled))
3504 dev_info(&h->pdev->dev, "Received interrupt while interrupts disabled "
3505 "(known firmware bug.) Ignoring.\n");
3510 static irqreturn_t cciss_intx_discard_completions(int irq, void *dev_id)
3512 ctlr_info_t *h = dev_id;
3513 unsigned long flags;
3516 if (ignore_bogus_interrupt(h))
3519 if (interrupt_not_for_us(h))
3521 spin_lock_irqsave(&h->lock, flags);
3522 while (interrupt_pending(h)) {
3523 raw_tag = get_next_completion(h);
3524 while (raw_tag != FIFO_EMPTY)
3525 raw_tag = next_command(h);
3527 spin_unlock_irqrestore(&h->lock, flags);
3531 static irqreturn_t cciss_msix_discard_completions(int irq, void *dev_id)
3533 ctlr_info_t *h = dev_id;
3534 unsigned long flags;
3537 if (ignore_bogus_interrupt(h))
3540 spin_lock_irqsave(&h->lock, flags);
3541 raw_tag = get_next_completion(h);
3542 while (raw_tag != FIFO_EMPTY)
3543 raw_tag = next_command(h);
3544 spin_unlock_irqrestore(&h->lock, flags);
3548 static irqreturn_t do_cciss_intx(int irq, void *dev_id)
3550 ctlr_info_t *h = dev_id;
3551 unsigned long flags;
3554 if (interrupt_not_for_us(h))
3556 spin_lock_irqsave(&h->lock, flags);
3557 while (interrupt_pending(h)) {
3558 raw_tag = get_next_completion(h);
3559 while (raw_tag != FIFO_EMPTY) {
3560 if (cciss_tag_contains_index(raw_tag))
3561 raw_tag = process_indexed_cmd(h, raw_tag);
3563 raw_tag = process_nonindexed_cmd(h, raw_tag);
3566 spin_unlock_irqrestore(&h->lock, flags);
3570 /* Add a second interrupt handler for MSI/MSI-X mode. In this mode we never
3571 * check the interrupt pending register because it is not set.
3573 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id)
3575 ctlr_info_t *h = dev_id;
3576 unsigned long flags;
3579 spin_lock_irqsave(&h->lock, flags);
3580 raw_tag = get_next_completion(h);
3581 while (raw_tag != FIFO_EMPTY) {
3582 if (cciss_tag_contains_index(raw_tag))
3583 raw_tag = process_indexed_cmd(h, raw_tag);
3585 raw_tag = process_nonindexed_cmd(h, raw_tag);
3587 spin_unlock_irqrestore(&h->lock, flags);
3592 * add_to_scan_list() - add controller to rescan queue
3593 * @h: Pointer to the controller.
3595 * Adds the controller to the rescan queue if not already on the queue.
3597 * returns 1 if added to the queue, 0 if skipped (could be on the
3598 * queue already, or the controller could be initializing or shutting
3601 static int add_to_scan_list(struct ctlr_info *h)
3603 struct ctlr_info *test_h;
3607 if (h->busy_initializing)
3610 if (!mutex_trylock(&h->busy_shutting_down))
3613 mutex_lock(&scan_mutex);
3614 list_for_each_entry(test_h, &scan_q, scan_list) {
3620 if (!found && !h->busy_scanning) {
3621 INIT_COMPLETION(h->scan_wait);
3622 list_add_tail(&h->scan_list, &scan_q);
3625 mutex_unlock(&scan_mutex);
3626 mutex_unlock(&h->busy_shutting_down);
3632 * remove_from_scan_list() - remove controller from rescan queue
3633 * @h: Pointer to the controller.
3635 * Removes the controller from the rescan queue if present. Blocks if
3636 * the controller is currently conducting a rescan. The controller
3637 * can be in one of three states:
3638 * 1. Doesn't need a scan
3639 * 2. On the scan list, but not scanning yet (we remove it)
3640 * 3. Busy scanning (and not on the list). In this case we want to wait for
3641 * the scan to complete to make sure the scanning thread for this
3642 * controller is completely idle.
3644 static void remove_from_scan_list(struct ctlr_info *h)
3646 struct ctlr_info *test_h, *tmp_h;
3648 mutex_lock(&scan_mutex);
3649 list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3650 if (test_h == h) { /* state 2. */
3651 list_del(&h->scan_list);
3652 complete_all(&h->scan_wait);
3653 mutex_unlock(&scan_mutex);
3657 if (h->busy_scanning) { /* state 3. */
3658 mutex_unlock(&scan_mutex);
3659 wait_for_completion(&h->scan_wait);
3660 } else { /* state 1, nothing to do. */
3661 mutex_unlock(&scan_mutex);
3666 * scan_thread() - kernel thread used to rescan controllers
3669 * A kernel thread used scan for drive topology changes on
3670 * controllers. The thread processes only one controller at a time
3671 * using a queue. Controllers are added to the queue using
3672 * add_to_scan_list() and removed from the queue either after done
3673 * processing or using remove_from_scan_list().
3677 static int scan_thread(void *data)
3679 struct ctlr_info *h;
3682 set_current_state(TASK_INTERRUPTIBLE);
3684 if (kthread_should_stop())
3688 mutex_lock(&scan_mutex);
3689 if (list_empty(&scan_q)) {
3690 mutex_unlock(&scan_mutex);
3694 h = list_entry(scan_q.next,
3697 list_del(&h->scan_list);
3698 h->busy_scanning = 1;
3699 mutex_unlock(&scan_mutex);
3701 rebuild_lun_table(h, 0, 0);
3702 complete_all(&h->scan_wait);
3703 mutex_lock(&scan_mutex);
3704 h->busy_scanning = 0;
3705 mutex_unlock(&scan_mutex);
3712 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3714 if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3717 switch (c->err_info->SenseInfo[12]) {
3719 dev_warn(&h->pdev->dev, "a state change "
3720 "detected, command retried\n");
3724 dev_warn(&h->pdev->dev, "LUN failure "
3725 "detected, action required\n");
3728 case REPORT_LUNS_CHANGED:
3729 dev_warn(&h->pdev->dev, "report LUN data changed\n");
3731 * Here, we could call add_to_scan_list and wake up the scan thread,
3732 * except that it's quite likely that we will get more than one
3733 * REPORT_LUNS_CHANGED condition in quick succession, which means
3734 * that those which occur after the first one will likely happen
3735 * *during* the scan_thread's rescan. And the rescan code is not
3736 * robust enough to restart in the middle, undoing what it has already
3737 * done, and it's not clear that it's even possible to do this, since
3738 * part of what it does is notify the block layer, which starts
3739 * doing it's own i/o to read partition tables and so on, and the
3740 * driver doesn't have visibility to know what might need undoing.
3741 * In any event, if possible, it is horribly complicated to get right
3742 * so we just don't do it for now.
3744 * Note: this REPORT_LUNS_CHANGED condition only occurs on the MSA2012.
3748 case POWER_OR_RESET:
3749 dev_warn(&h->pdev->dev,
3750 "a power on or device reset detected\n");
3753 case UNIT_ATTENTION_CLEARED:
3754 dev_warn(&h->pdev->dev,
3755 "unit attention cleared by another initiator\n");
3759 dev_warn(&h->pdev->dev, "unknown unit attention detected\n");
3765 * We cannot read the structure directly, for portability we must use
3767 * This is for debug only.
3769 static void print_cfg_table(ctlr_info_t *h)
3773 CfgTable_struct *tb = h->cfgtable;
3775 dev_dbg(&h->pdev->dev, "Controller Configuration information\n");
3776 dev_dbg(&h->pdev->dev, "------------------------------------\n");
3777 for (i = 0; i < 4; i++)
3778 temp_name[i] = readb(&(tb->Signature[i]));
3779 temp_name[4] = '\0';
3780 dev_dbg(&h->pdev->dev, " Signature = %s\n", temp_name);
3781 dev_dbg(&h->pdev->dev, " Spec Number = %d\n",
3782 readl(&(tb->SpecValence)));
3783 dev_dbg(&h->pdev->dev, " Transport methods supported = 0x%x\n",
3784 readl(&(tb->TransportSupport)));
3785 dev_dbg(&h->pdev->dev, " Transport methods active = 0x%x\n",
3786 readl(&(tb->TransportActive)));
3787 dev_dbg(&h->pdev->dev, " Requested transport Method = 0x%x\n",
3788 readl(&(tb->HostWrite.TransportRequest)));
3789 dev_dbg(&h->pdev->dev, " Coalesce Interrupt Delay = 0x%x\n",
3790 readl(&(tb->HostWrite.CoalIntDelay)));
3791 dev_dbg(&h->pdev->dev, " Coalesce Interrupt Count = 0x%x\n",
3792 readl(&(tb->HostWrite.CoalIntCount)));
3793 dev_dbg(&h->pdev->dev, " Max outstanding commands = 0x%d\n",
3794 readl(&(tb->CmdsOutMax)));
3795 dev_dbg(&h->pdev->dev, " Bus Types = 0x%x\n",
3796 readl(&(tb->BusTypes)));
3797 for (i = 0; i < 16; i++)
3798 temp_name[i] = readb(&(tb->ServerName[i]));
3799 temp_name[16] = '\0';
3800 dev_dbg(&h->pdev->dev, " Server Name = %s\n", temp_name);
3801 dev_dbg(&h->pdev->dev, " Heartbeat Counter = 0x%x\n\n\n",
3802 readl(&(tb->HeartBeat)));
3805 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3807 int i, offset, mem_type, bar_type;
3808 if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3811 for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3812 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3813 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3816 mem_type = pci_resource_flags(pdev, i) &
3817 PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3819 case PCI_BASE_ADDRESS_MEM_TYPE_32:
3820 case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3821 offset += 4; /* 32 bit */
3823 case PCI_BASE_ADDRESS_MEM_TYPE_64:
3826 default: /* reserved in PCI 2.2 */
3827 dev_warn(&pdev->dev,
3828 "Base address is invalid\n");
3833 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3839 /* Fill in bucket_map[], given nsgs (the max number of
3840 * scatter gather elements supported) and bucket[],
3841 * which is an array of 8 integers. The bucket[] array
3842 * contains 8 different DMA transfer sizes (in 16
3843 * byte increments) which the controller uses to fetch
3844 * commands. This function fills in bucket_map[], which
3845 * maps a given number of scatter gather elements to one of
3846 * the 8 DMA transfer sizes. The point of it is to allow the
3847 * controller to only do as much DMA as needed to fetch the
3848 * command, with the DMA transfer size encoded in the lower
3849 * bits of the command address.
3851 static void calc_bucket_map(int bucket[], int num_buckets,
3852 int nsgs, int *bucket_map)
3856 /* even a command with 0 SGs requires 4 blocks */
3857 #define MINIMUM_TRANSFER_BLOCKS 4
3858 #define NUM_BUCKETS 8
3859 /* Note, bucket_map must have nsgs+1 entries. */
3860 for (i = 0; i <= nsgs; i++) {
3861 /* Compute size of a command with i SG entries */
3862 size = i + MINIMUM_TRANSFER_BLOCKS;
3863 b = num_buckets; /* Assume the biggest bucket */
3864 /* Find the bucket that is just big enough */
3865 for (j = 0; j < 8; j++) {
3866 if (bucket[j] >= size) {
3871 /* for a command with i SG entries, use bucket b. */
3876 static void __devinit cciss_wait_for_mode_change_ack(ctlr_info_t *h)
3880 /* under certain very rare conditions, this can take awhile.
3881 * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3882 * as we enter this code.) */
3883 for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3884 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3886 usleep_range(10000, 20000);
3890 static __devinit void cciss_enter_performant_mode(ctlr_info_t *h,
3893 /* This is a bit complicated. There are 8 registers on
3894 * the controller which we write to to tell it 8 different
3895 * sizes of commands which there may be. It's a way of
3896 * reducing the DMA done to fetch each command. Encoded into
3897 * each command's tag are 3 bits which communicate to the controller
3898 * which of the eight sizes that command fits within. The size of
3899 * each command depends on how many scatter gather entries there are.
3900 * Each SG entry requires 16 bytes. The eight registers are programmed
3901 * with the number of 16-byte blocks a command of that size requires.
3902 * The smallest command possible requires 5 such 16 byte blocks.
3903 * the largest command possible requires MAXSGENTRIES + 4 16-byte
3904 * blocks. Note, this only extends to the SG entries contained
3905 * within the command block, and does not extend to chained blocks
3906 * of SG elements. bft[] contains the eight values we write to
3907 * the registers. They are not evenly distributed, but have more
3908 * sizes for small commands, and fewer sizes for larger commands.
3911 int bft[8] = { 5, 6, 8, 10, 12, 20, 28, MAXSGENTRIES + 4};
3913 * 5 = 1 s/g entry or 4k
3914 * 6 = 2 s/g entry or 8k
3915 * 8 = 4 s/g entry or 16k
3916 * 10 = 6 s/g entry or 24k
3918 unsigned long register_value;
3919 BUILD_BUG_ON(28 > MAXSGENTRIES + 4);
3921 h->reply_pool_wraparound = 1; /* spec: init to 1 */
3923 /* Controller spec: zero out this buffer. */
3924 memset(h->reply_pool, 0, h->max_commands * sizeof(__u64));
3925 h->reply_pool_head = h->reply_pool;
3927 trans_offset = readl(&(h->cfgtable->TransMethodOffset));
3928 calc_bucket_map(bft, ARRAY_SIZE(bft), h->maxsgentries,
3929 h->blockFetchTable);
3930 writel(bft[0], &h->transtable->BlockFetch0);
3931 writel(bft[1], &h->transtable->BlockFetch1);
3932 writel(bft[2], &h->transtable->BlockFetch2);
3933 writel(bft[3], &h->transtable->BlockFetch3);
3934 writel(bft[4], &h->transtable->BlockFetch4);
3935 writel(bft[5], &h->transtable->BlockFetch5);
3936 writel(bft[6], &h->transtable->BlockFetch6);
3937 writel(bft[7], &h->transtable->BlockFetch7);
3939 /* size of controller ring buffer */
3940 writel(h->max_commands, &h->transtable->RepQSize);
3941 writel(1, &h->transtable->RepQCount);
3942 writel(0, &h->transtable->RepQCtrAddrLow32);
3943 writel(0, &h->transtable->RepQCtrAddrHigh32);
3944 writel(h->reply_pool_dhandle, &h->transtable->RepQAddr0Low32);
3945 writel(0, &h->transtable->RepQAddr0High32);
3946 writel(CFGTBL_Trans_Performant | use_short_tags,
3947 &(h->cfgtable->HostWrite.TransportRequest));
3949 writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
3950 cciss_wait_for_mode_change_ack(h);
3951 register_value = readl(&(h->cfgtable->TransportActive));
3952 if (!(register_value & CFGTBL_Trans_Performant))
3953 dev_warn(&h->pdev->dev, "cciss: unable to get board into"
3954 " performant mode\n");
3957 static void __devinit cciss_put_controller_into_performant_mode(ctlr_info_t *h)
3959 __u32 trans_support;
3961 dev_dbg(&h->pdev->dev, "Trying to put board into Performant mode\n");
3962 /* Attempt to put controller into performant mode if supported */
3963 /* Does board support performant mode? */
3964 trans_support = readl(&(h->cfgtable->TransportSupport));
3965 if (!(trans_support & PERFORMANT_MODE))
3968 dev_dbg(&h->pdev->dev, "Placing controller into performant mode\n");
3969 /* Performant mode demands commands on a 32 byte boundary
3970 * pci_alloc_consistent aligns on page boundarys already.
3971 * Just need to check if divisible by 32
3973 if ((sizeof(CommandList_struct) % 32) != 0) {
3974 dev_warn(&h->pdev->dev, "%s %d %s\n",
3975 "cciss info: command size[",
3976 (int)sizeof(CommandList_struct),
3977 "] not divisible by 32, no performant mode..\n");
3981 /* Performant mode ring buffer and supporting data structures */
3982 h->reply_pool = (__u64 *)pci_alloc_consistent(
3983 h->pdev, h->max_commands * sizeof(__u64),
3984 &(h->reply_pool_dhandle));
3986 /* Need a block fetch table for performant mode */
3987 h->blockFetchTable = kmalloc(((h->maxsgentries+1) *
3988 sizeof(__u32)), GFP_KERNEL);
3990 if ((h->reply_pool == NULL) || (h->blockFetchTable == NULL))
3993 cciss_enter_performant_mode(h,
3994 trans_support & CFGTBL_Trans_use_short_tags);
3996 /* Change the access methods to the performant access methods */
3997 h->access = SA5_performant_access;
3998 h->transMethod = CFGTBL_Trans_Performant;
4002 kfree(h->blockFetchTable);
4004 pci_free_consistent(h->pdev,
4005 h->max_commands * sizeof(__u64),
4007 h->reply_pool_dhandle);
4010 } /* cciss_put_controller_into_performant_mode */
4012 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
4013 * controllers that are capable. If not, we use IO-APIC mode.
4016 static void __devinit cciss_interrupt_mode(ctlr_info_t *h)
4018 #ifdef CONFIG_PCI_MSI
4020 struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
4024 /* Some boards advertise MSI but don't really support it */
4025 if ((h->board_id == 0x40700E11) || (h->board_id == 0x40800E11) ||
4026 (h->board_id == 0x40820E11) || (h->board_id == 0x40830E11))
4027 goto default_int_mode;
4029 if (pci_find_capability(h->pdev, PCI_CAP_ID_MSIX)) {
4030 err = pci_enable_msix(h->pdev, cciss_msix_entries, 4);
4032 h->intr[0] = cciss_msix_entries[0].vector;
4033 h->intr[1] = cciss_msix_entries[1].vector;
4034 h->intr[2] = cciss_msix_entries[2].vector;
4035 h->intr[3] = cciss_msix_entries[3].vector;
4040 dev_warn(&h->pdev->dev,
4041 "only %d MSI-X vectors available\n", err);
4042 goto default_int_mode;
4044 dev_warn(&h->pdev->dev,
4045 "MSI-X init failed %d\n", err);
4046 goto default_int_mode;
4049 if (pci_find_capability(h->pdev, PCI_CAP_ID_MSI)) {
4050 if (!pci_enable_msi(h->pdev))
4053 dev_warn(&h->pdev->dev, "MSI init failed\n");
4056 #endif /* CONFIG_PCI_MSI */
4057 /* if we get here we're going to use the default interrupt mode */
4058 h->intr[PERF_MODE_INT] = h->pdev->irq;
4062 static int __devinit cciss_lookup_board_id(struct pci_dev *pdev, u32 *board_id)
4065 u32 subsystem_vendor_id, subsystem_device_id;
4067 subsystem_vendor_id = pdev->subsystem_vendor;
4068 subsystem_device_id = pdev->subsystem_device;
4069 *board_id = ((subsystem_device_id << 16) & 0xffff0000) |
4070 subsystem_vendor_id;
4072 for (i = 0; i < ARRAY_SIZE(products); i++)
4073 if (*board_id == products[i].board_id)
4075 dev_warn(&pdev->dev, "unrecognized board ID: 0x%08x, ignoring.\n",
4080 static inline bool cciss_board_disabled(ctlr_info_t *h)
4084 (void) pci_read_config_word(h->pdev, PCI_COMMAND, &command);
4085 return ((command & PCI_COMMAND_MEMORY) == 0);
4088 static int __devinit cciss_pci_find_memory_BAR(struct pci_dev *pdev,
4089 unsigned long *memory_bar)
4093 for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
4094 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM) {
4095 /* addressing mode bits already removed */
4096 *memory_bar = pci_resource_start(pdev, i);
4097 dev_dbg(&pdev->dev, "memory BAR = %lx\n",
4101 dev_warn(&pdev->dev, "no memory BAR found\n");
4105 static int __devinit cciss_wait_for_board_state(struct pci_dev *pdev,
4106 void __iomem *vaddr, int wait_for_ready)
4107 #define BOARD_READY 1
4108 #define BOARD_NOT_READY 0
4114 iterations = CCISS_BOARD_READY_ITERATIONS;
4116 iterations = CCISS_BOARD_NOT_READY_ITERATIONS;
4118 for (i = 0; i < iterations; i++) {
4119 scratchpad = readl(vaddr + SA5_SCRATCHPAD_OFFSET);
4120 if (wait_for_ready) {
4121 if (scratchpad == CCISS_FIRMWARE_READY)
4124 if (scratchpad != CCISS_FIRMWARE_READY)
4127 msleep(CCISS_BOARD_READY_POLL_INTERVAL_MSECS);
4129 dev_warn(&pdev->dev, "board not ready, timed out.\n");
4133 static int __devinit cciss_find_cfg_addrs(struct pci_dev *pdev,
4134 void __iomem *vaddr, u32 *cfg_base_addr, u64 *cfg_base_addr_index,
4137 *cfg_base_addr = readl(vaddr + SA5_CTCFG_OFFSET);
4138 *cfg_offset = readl(vaddr + SA5_CTMEM_OFFSET);
4139 *cfg_base_addr &= (u32) 0x0000ffff;
4140 *cfg_base_addr_index = find_PCI_BAR_index(pdev, *cfg_base_addr);
4141 if (*cfg_base_addr_index == -1) {
4142 dev_warn(&pdev->dev, "cannot find cfg_base_addr_index, "
4143 "*cfg_base_addr = 0x%08x\n", *cfg_base_addr);
4149 static int __devinit cciss_find_cfgtables(ctlr_info_t *h)
4153 u64 cfg_base_addr_index;
4157 rc = cciss_find_cfg_addrs(h->pdev, h->vaddr, &cfg_base_addr,
4158 &cfg_base_addr_index, &cfg_offset);
4161 h->cfgtable = remap_pci_mem(pci_resource_start(h->pdev,
4162 cfg_base_addr_index) + cfg_offset, sizeof(h->cfgtable));
4165 rc = write_driver_ver_to_cfgtable(h->cfgtable);
4168 /* Find performant mode table. */
4169 trans_offset = readl(&h->cfgtable->TransMethodOffset);
4170 h->transtable = remap_pci_mem(pci_resource_start(h->pdev,
4171 cfg_base_addr_index)+cfg_offset+trans_offset,
4172 sizeof(*h->transtable));
4178 static void __devinit cciss_get_max_perf_mode_cmds(struct ctlr_info *h)
4180 h->max_commands = readl(&(h->cfgtable->MaxPerformantModeCommands));
4182 /* Limit commands in memory limited kdump scenario. */
4183 if (reset_devices && h->max_commands > 32)
4184 h->max_commands = 32;
4186 if (h->max_commands < 16) {
4187 dev_warn(&h->pdev->dev, "Controller reports "
4188 "max supported commands of %d, an obvious lie. "
4189 "Using 16. Ensure that firmware is up to date.\n",
4191 h->max_commands = 16;
4195 /* Interrogate the hardware for some limits:
4196 * max commands, max SG elements without chaining, and with chaining,
4197 * SG chain block size, etc.
4199 static void __devinit cciss_find_board_params(ctlr_info_t *h)
4201 cciss_get_max_perf_mode_cmds(h);
4202 h->nr_cmds = h->max_commands - 4; /* Allow room for some ioctls */
4203 h->maxsgentries = readl(&(h->cfgtable->MaxSGElements));
4205 * Limit in-command s/g elements to 32 save dma'able memory.
4206 * Howvever spec says if 0, use 31
4208 h->max_cmd_sgentries = 31;
4209 if (h->maxsgentries > 512) {
4210 h->max_cmd_sgentries = 32;
4211 h->chainsize = h->maxsgentries - h->max_cmd_sgentries + 1;
4212 h->maxsgentries--; /* save one for chain pointer */
4214 h->maxsgentries = 31; /* default to traditional values */
4219 static inline bool CISS_signature_present(ctlr_info_t *h)
4221 if ((readb(&h->cfgtable->Signature[0]) != 'C') ||
4222 (readb(&h->cfgtable->Signature[1]) != 'I') ||
4223 (readb(&h->cfgtable->Signature[2]) != 'S') ||
4224 (readb(&h->cfgtable->Signature[3]) != 'S')) {
4225 dev_warn(&h->pdev->dev, "not a valid CISS config table\n");
4231 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
4232 static inline void cciss_enable_scsi_prefetch(ctlr_info_t *h)
4237 prefetch = readl(&(h->cfgtable->SCSI_Prefetch));
4239 writel(prefetch, &(h->cfgtable->SCSI_Prefetch));
4243 /* Disable DMA prefetch for the P600. Otherwise an ASIC bug may result
4244 * in a prefetch beyond physical memory.
4246 static inline void cciss_p600_dma_prefetch_quirk(ctlr_info_t *h)
4251 if (h->board_id != 0x3225103C)
4253 dma_prefetch = readl(h->vaddr + I2O_DMA1_CFG);
4254 dma_prefetch |= 0x8000;
4255 writel(dma_prefetch, h->vaddr + I2O_DMA1_CFG);
4256 pci_read_config_dword(h->pdev, PCI_COMMAND_PARITY, &dma_refetch);
4258 pci_write_config_dword(h->pdev, PCI_COMMAND_PARITY, dma_refetch);
4261 static int __devinit cciss_pci_init(ctlr_info_t *h)
4263 int prod_index, err;
4265 prod_index = cciss_lookup_board_id(h->pdev, &h->board_id);
4268 h->product_name = products[prod_index].product_name;
4269 h->access = *(products[prod_index].access);
4271 if (cciss_board_disabled(h)) {
4272 dev_warn(&h->pdev->dev, "controller appears to be disabled\n");
4275 err = pci_enable_device(h->pdev);
4277 dev_warn(&h->pdev->dev, "Unable to Enable PCI device\n");
4281 err = pci_request_regions(h->pdev, "cciss");
4283 dev_warn(&h->pdev->dev,
4284 "Cannot obtain PCI resources, aborting\n");
4288 dev_dbg(&h->pdev->dev, "irq = %x\n", h->pdev->irq);
4289 dev_dbg(&h->pdev->dev, "board_id = %x\n", h->board_id);
4291 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
4292 * else we use the IO-APIC interrupt assigned to us by system ROM.
4294 cciss_interrupt_mode(h);
4295 err = cciss_pci_find_memory_BAR(h->pdev, &h->paddr);
4297 goto err_out_free_res;
4298 h->vaddr = remap_pci_mem(h->paddr, 0x250);
4301 goto err_out_free_res;
4303 err = cciss_wait_for_board_state(h->pdev, h->vaddr, BOARD_READY);
4305 goto err_out_free_res;
4306 err = cciss_find_cfgtables(h);
4308 goto err_out_free_res;
4310 cciss_find_board_params(h);
4312 if (!CISS_signature_present(h)) {
4314 goto err_out_free_res;
4316 cciss_enable_scsi_prefetch(h);
4317 cciss_p600_dma_prefetch_quirk(h);
4318 cciss_put_controller_into_performant_mode(h);
4323 * Deliberately omit pci_disable_device(): it does something nasty to
4324 * Smart Array controllers that pci_enable_device does not undo
4327 iounmap(h->transtable);
4329 iounmap(h->cfgtable);
4332 pci_release_regions(h->pdev);
4336 /* Function to find the first free pointer into our hba[] array
4337 * Returns -1 if no free entries are left.
4339 static int alloc_cciss_hba(struct pci_dev *pdev)
4343 for (i = 0; i < MAX_CTLR; i++) {
4347 h = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
4354 dev_warn(&pdev->dev, "This driver supports a maximum"
4355 " of %d controllers.\n", MAX_CTLR);
4358 dev_warn(&pdev->dev, "out of memory.\n");
4362 static void free_hba(ctlr_info_t *h)
4366 hba[h->ctlr] = NULL;
4367 for (i = 0; i < h->highest_lun + 1; i++)
4368 if (h->gendisk[i] != NULL)
4369 put_disk(h->gendisk[i]);
4373 /* Send a message CDB to the firmware. */
4374 static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
4377 CommandListHeader_struct CommandHeader;
4378 RequestBlock_struct Request;
4379 ErrDescriptor_struct ErrorDescriptor;
4381 static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
4384 uint32_t paddr32, tag;
4385 void __iomem *vaddr;
4388 vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
4392 /* The Inbound Post Queue only accepts 32-bit physical addresses for the
4393 CCISS commands, so they must be allocated from the lower 4GiB of
4395 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
4401 cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
4407 /* This must fit, because of the 32-bit consistent DMA mask. Also,
4408 although there's no guarantee, we assume that the address is at
4409 least 4-byte aligned (most likely, it's page-aligned). */
4412 cmd->CommandHeader.ReplyQueue = 0;
4413 cmd->CommandHeader.SGList = 0;
4414 cmd->CommandHeader.SGTotal = 0;
4415 cmd->CommandHeader.Tag.lower = paddr32;
4416 cmd->CommandHeader.Tag.upper = 0;
4417 memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
4419 cmd->Request.CDBLen = 16;
4420 cmd->Request.Type.Type = TYPE_MSG;
4421 cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
4422 cmd->Request.Type.Direction = XFER_NONE;
4423 cmd->Request.Timeout = 0; /* Don't time out */
4424 cmd->Request.CDB[0] = opcode;
4425 cmd->Request.CDB[1] = type;
4426 memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
4428 cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
4429 cmd->ErrorDescriptor.Addr.upper = 0;
4430 cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
4432 writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
4434 for (i = 0; i < 10; i++) {
4435 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
4436 if ((tag & ~3) == paddr32)
4438 msleep(CCISS_POST_RESET_NOOP_TIMEOUT_MSECS);
4443 /* we leak the DMA buffer here ... no choice since the controller could
4444 still complete the command. */
4447 "controller message %02x:%02x timed out\n",
4452 pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
4455 dev_err(&pdev->dev, "controller message %02x:%02x failed\n",
4460 dev_info(&pdev->dev, "controller message %02x:%02x succeeded\n",
4465 #define cciss_noop(p) cciss_message(p, 3, 0)
4467 static int cciss_controller_hard_reset(struct pci_dev *pdev,
4468 void * __iomem vaddr, u32 use_doorbell)
4474 /* For everything after the P600, the PCI power state method
4475 * of resetting the controller doesn't work, so we have this
4476 * other way using the doorbell register.
4478 dev_info(&pdev->dev, "using doorbell to reset controller\n");
4479 writel(use_doorbell, vaddr + SA5_DOORBELL);
4480 } else { /* Try to do it the PCI power state way */
4482 /* Quoting from the Open CISS Specification: "The Power
4483 * Management Control/Status Register (CSR) controls the power
4484 * state of the device. The normal operating state is D0,
4485 * CSR=00h. The software off state is D3, CSR=03h. To reset
4486 * the controller, place the interface device in D3 then to D0,
4487 * this causes a secondary PCI reset which will reset the
4490 pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4493 "cciss_controller_hard_reset: "
4494 "PCI PM not supported\n");
4497 dev_info(&pdev->dev, "using PCI PM to reset controller\n");
4498 /* enter the D3hot power management state */
4499 pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4500 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4502 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4506 /* enter the D0 power management state */
4507 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4509 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4514 static __devinit void init_driver_version(char *driver_version, int len)
4516 memset(driver_version, 0, len);
4517 strncpy(driver_version, "cciss " DRIVER_NAME, len - 1);
4520 static __devinit int write_driver_ver_to_cfgtable(
4521 CfgTable_struct __iomem *cfgtable)
4523 char *driver_version;
4524 int i, size = sizeof(cfgtable->driver_version);
4526 driver_version = kmalloc(size, GFP_KERNEL);
4527 if (!driver_version)
4530 init_driver_version(driver_version, size);
4531 for (i = 0; i < size; i++)
4532 writeb(driver_version[i], &cfgtable->driver_version[i]);
4533 kfree(driver_version);
4537 static __devinit void read_driver_ver_from_cfgtable(
4538 CfgTable_struct __iomem *cfgtable, unsigned char *driver_ver)
4542 for (i = 0; i < sizeof(cfgtable->driver_version); i++)
4543 driver_ver[i] = readb(&cfgtable->driver_version[i]);
4546 static __devinit int controller_reset_failed(
4547 CfgTable_struct __iomem *cfgtable)
4550 char *driver_ver, *old_driver_ver;
4551 int rc, size = sizeof(cfgtable->driver_version);
4553 old_driver_ver = kmalloc(2 * size, GFP_KERNEL);
4554 if (!old_driver_ver)
4556 driver_ver = old_driver_ver + size;
4558 /* After a reset, the 32 bytes of "driver version" in the cfgtable
4559 * should have been changed, otherwise we know the reset failed.
4561 init_driver_version(old_driver_ver, size);
4562 read_driver_ver_from_cfgtable(cfgtable, driver_ver);
4563 rc = !memcmp(driver_ver, old_driver_ver, size);
4564 kfree(old_driver_ver);
4568 /* This does a hard reset of the controller using PCI power management
4569 * states or using the doorbell register. */
4570 static __devinit int cciss_kdump_hard_reset_controller(struct pci_dev *pdev)
4574 u64 cfg_base_addr_index;
4575 void __iomem *vaddr;
4576 unsigned long paddr;
4577 u32 misc_fw_support;
4579 CfgTable_struct __iomem *cfgtable;
4582 u16 command_register;
4584 /* For controllers as old a the p600, this is very nearly
4587 * pci_save_state(pci_dev);
4588 * pci_set_power_state(pci_dev, PCI_D3hot);
4589 * pci_set_power_state(pci_dev, PCI_D0);
4590 * pci_restore_state(pci_dev);
4592 * For controllers newer than the P600, the pci power state
4593 * method of resetting doesn't work so we have another way
4594 * using the doorbell register.
4597 /* Exclude 640x boards. These are two pci devices in one slot
4598 * which share a battery backed cache module. One controls the
4599 * cache, the other accesses the cache through the one that controls
4600 * it. If we reset the one controlling the cache, the other will
4601 * likely not be happy. Just forbid resetting this conjoined mess.
4603 cciss_lookup_board_id(pdev, &board_id);
4604 if (board_id == 0x409C0E11 || board_id == 0x409D0E11) {
4605 dev_warn(&pdev->dev, "Cannot reset Smart Array 640x "
4606 "due to shared cache module.");
4610 /* Save the PCI command register */
4611 pci_read_config_word(pdev, 4, &command_register);
4612 /* Turn the board off. This is so that later pci_restore_state()
4613 * won't turn the board on before the rest of config space is ready.
4615 pci_disable_device(pdev);
4616 pci_save_state(pdev);
4618 /* find the first memory BAR, so we can find the cfg table */
4619 rc = cciss_pci_find_memory_BAR(pdev, &paddr);
4622 vaddr = remap_pci_mem(paddr, 0x250);
4626 /* find cfgtable in order to check if reset via doorbell is supported */
4627 rc = cciss_find_cfg_addrs(pdev, vaddr, &cfg_base_addr,
4628 &cfg_base_addr_index, &cfg_offset);
4631 cfgtable = remap_pci_mem(pci_resource_start(pdev,
4632 cfg_base_addr_index) + cfg_offset, sizeof(*cfgtable));
4637 rc = write_driver_ver_to_cfgtable(cfgtable);
4641 /* If reset via doorbell register is supported, use that.
4642 * There are two such methods. Favor the newest method.
4644 misc_fw_support = readl(&cfgtable->misc_fw_support);
4645 use_doorbell = misc_fw_support & MISC_FW_DOORBELL_RESET2;
4647 use_doorbell = DOORBELL_CTLR_RESET2;
4649 use_doorbell = misc_fw_support & MISC_FW_DOORBELL_RESET;
4651 use_doorbell = DOORBELL_CTLR_RESET;
4654 rc = cciss_controller_hard_reset(pdev, vaddr, use_doorbell);
4656 goto unmap_cfgtable;
4657 pci_restore_state(pdev);
4658 rc = pci_enable_device(pdev);
4660 dev_warn(&pdev->dev, "failed to enable device.\n");
4661 goto unmap_cfgtable;
4663 pci_write_config_word(pdev, 4, command_register);
4665 /* Some devices (notably the HP Smart Array 5i Controller)
4666 need a little pause here */
4667 msleep(CCISS_POST_RESET_PAUSE_MSECS);
4669 /* Wait for board to become not ready, then ready. */
4670 dev_info(&pdev->dev, "Waiting for board to reset.\n");
4671 rc = cciss_wait_for_board_state(pdev, vaddr, BOARD_NOT_READY);
4673 dev_warn(&pdev->dev, "Failed waiting for board to hard reset."
4674 " Will try soft reset.\n");
4675 rc = -ENOTSUPP; /* Not expected, but try soft reset later */
4676 goto unmap_cfgtable;
4678 rc = cciss_wait_for_board_state(pdev, vaddr, BOARD_READY);
4680 dev_warn(&pdev->dev,
4681 "failed waiting for board to become ready "
4682 "after hard reset\n");
4683 goto unmap_cfgtable;
4686 rc = controller_reset_failed(vaddr);
4688 goto unmap_cfgtable;
4690 dev_warn(&pdev->dev, "Unable to successfully hard reset "
4691 "controller. Will try soft reset.\n");
4692 rc = -ENOTSUPP; /* Not expected, but try soft reset later */
4694 dev_info(&pdev->dev, "Board ready after hard reset.\n");
4705 static __devinit int cciss_init_reset_devices(struct pci_dev *pdev)
4712 /* Reset the controller with a PCI power-cycle or via doorbell */
4713 rc = cciss_kdump_hard_reset_controller(pdev);
4715 /* -ENOTSUPP here means we cannot reset the controller
4716 * but it's already (and still) up and running in
4717 * "performant mode". Or, it might be 640x, which can't reset
4718 * due to concerns about shared bbwc between 6402/6404 pair.
4720 if (rc == -ENOTSUPP)
4721 return rc; /* just try to do the kdump anyhow. */
4725 /* Now try to get the controller to respond to a no-op */
4726 dev_warn(&pdev->dev, "Waiting for controller to respond to no-op\n");
4727 for (i = 0; i < CCISS_POST_RESET_NOOP_RETRIES; i++) {
4728 if (cciss_noop(pdev) == 0)
4731 dev_warn(&pdev->dev, "no-op failed%s\n",
4732 (i < CCISS_POST_RESET_NOOP_RETRIES - 1 ?
4733 "; re-trying" : ""));
4734 msleep(CCISS_POST_RESET_NOOP_INTERVAL_MSECS);
4739 static __devinit int cciss_allocate_cmd_pool(ctlr_info_t *h)
4741 h->cmd_pool_bits = kmalloc(
4742 DIV_ROUND_UP(h->nr_cmds, BITS_PER_LONG) *
4743 sizeof(unsigned long), GFP_KERNEL);
4744 h->cmd_pool = pci_alloc_consistent(h->pdev,
4745 h->nr_cmds * sizeof(CommandList_struct),
4746 &(h->cmd_pool_dhandle));
4747 h->errinfo_pool = pci_alloc_consistent(h->pdev,
4748 h->nr_cmds * sizeof(ErrorInfo_struct),
4749 &(h->errinfo_pool_dhandle));
4750 if ((h->cmd_pool_bits == NULL)
4751 || (h->cmd_pool == NULL)
4752 || (h->errinfo_pool == NULL)) {
4753 dev_err(&h->pdev->dev, "out of memory");
4759 static __devinit int cciss_allocate_scatterlists(ctlr_info_t *h)
4763 /* zero it, so that on free we need not know how many were alloc'ed */
4764 h->scatter_list = kzalloc(h->max_commands *
4765 sizeof(struct scatterlist *), GFP_KERNEL);
4766 if (!h->scatter_list)
4769 for (i = 0; i < h->nr_cmds; i++) {
4770 h->scatter_list[i] = kmalloc(sizeof(struct scatterlist) *
4771 h->maxsgentries, GFP_KERNEL);
4772 if (h->scatter_list[i] == NULL) {
4773 dev_err(&h->pdev->dev, "could not allocate "
4781 static void cciss_free_scatterlists(ctlr_info_t *h)
4785 if (h->scatter_list) {
4786 for (i = 0; i < h->nr_cmds; i++)
4787 kfree(h->scatter_list[i]);
4788 kfree(h->scatter_list);
4792 static void cciss_free_cmd_pool(ctlr_info_t *h)
4794 kfree(h->cmd_pool_bits);
4796 pci_free_consistent(h->pdev,
4797 h->nr_cmds * sizeof(CommandList_struct),
4798 h->cmd_pool, h->cmd_pool_dhandle);
4799 if (h->errinfo_pool)
4800 pci_free_consistent(h->pdev,
4801 h->nr_cmds * sizeof(ErrorInfo_struct),
4802 h->errinfo_pool, h->errinfo_pool_dhandle);
4805 static int cciss_request_irq(ctlr_info_t *h,
4806 irqreturn_t (*msixhandler)(int, void *),
4807 irqreturn_t (*intxhandler)(int, void *))
4809 if (h->msix_vector || h->msi_vector) {
4810 if (!request_irq(h->intr[PERF_MODE_INT], msixhandler,
4811 IRQF_DISABLED, h->devname, h))
4813 dev_err(&h->pdev->dev, "Unable to get msi irq %d"
4814 " for %s\n", h->intr[PERF_MODE_INT],
4819 if (!request_irq(h->intr[PERF_MODE_INT], intxhandler,
4820 IRQF_DISABLED, h->devname, h))
4822 dev_err(&h->pdev->dev, "Unable to get irq %d for %s\n",
4823 h->intr[PERF_MODE_INT], h->devname);
4827 static int __devinit cciss_kdump_soft_reset(ctlr_info_t *h)
4829 if (cciss_send_reset(h, CTLR_LUNID, CCISS_RESET_TYPE_CONTROLLER)) {
4830 dev_warn(&h->pdev->dev, "Resetting array controller failed.\n");
4834 dev_info(&h->pdev->dev, "Waiting for board to soft reset.\n");
4835 if (cciss_wait_for_board_state(h->pdev, h->vaddr, BOARD_NOT_READY)) {
4836 dev_warn(&h->pdev->dev, "Soft reset had no effect.\n");
4840 dev_info(&h->pdev->dev, "Board reset, awaiting READY status.\n");
4841 if (cciss_wait_for_board_state(h->pdev, h->vaddr, BOARD_READY)) {
4842 dev_warn(&h->pdev->dev, "Board failed to become ready "
4843 "after soft reset.\n");
4850 static void cciss_undo_allocations_after_kdump_soft_reset(ctlr_info_t *h)
4854 free_irq(h->intr[PERF_MODE_INT], h);
4855 #ifdef CONFIG_PCI_MSI
4857 pci_disable_msix(h->pdev);
4858 else if (h->msi_vector)
4859 pci_disable_msi(h->pdev);
4860 #endif /* CONFIG_PCI_MSI */
4861 cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
4862 cciss_free_scatterlists(h);
4863 cciss_free_cmd_pool(h);
4864 kfree(h->blockFetchTable);
4866 pci_free_consistent(h->pdev, h->max_commands * sizeof(__u64),
4867 h->reply_pool, h->reply_pool_dhandle);
4869 iounmap(h->transtable);
4871 iounmap(h->cfgtable);
4874 unregister_blkdev(h->major, h->devname);
4875 cciss_destroy_hba_sysfs_entry(h);
4876 pci_release_regions(h->pdev);
4882 * This is it. Find all the controllers and register them. I really hate
4883 * stealing all these major device numbers.
4884 * returns the number of block devices registered.
4886 static int __devinit cciss_init_one(struct pci_dev *pdev,
4887 const struct pci_device_id *ent)
4892 int try_soft_reset = 0;
4893 int dac, return_code;
4894 InquiryData_struct *inq_buff;
4896 unsigned long flags;
4898 rc = cciss_init_reset_devices(pdev);
4900 if (rc != -ENOTSUPP)
4902 /* If the reset fails in a particular way (it has no way to do
4903 * a proper hard reset, so returns -ENOTSUPP) we can try to do
4904 * a soft reset once we get the controller configured up to the
4905 * point that it can accept a command.
4911 reinit_after_soft_reset:
4913 i = alloc_cciss_hba(pdev);
4919 h->busy_initializing = 1;
4920 INIT_LIST_HEAD(&h->cmpQ);
4921 INIT_LIST_HEAD(&h->reqQ);
4922 mutex_init(&h->busy_shutting_down);
4924 if (cciss_pci_init(h) != 0)
4925 goto clean_no_release_regions;
4927 sprintf(h->devname, "cciss%d", i);
4930 init_completion(&h->scan_wait);
4932 if (cciss_create_hba_sysfs_entry(h))
4935 /* configure PCI DMA stuff */
4936 if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
4938 else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
4941 dev_err(&h->pdev->dev, "no suitable DMA available\n");
4946 * register with the major number, or get a dynamic major number
4947 * by passing 0 as argument. This is done for greater than
4948 * 8 controller support.
4950 if (i < MAX_CTLR_ORIG)
4951 h->major = COMPAQ_CISS_MAJOR + i;
4952 rc = register_blkdev(h->major, h->devname);
4953 if (rc == -EBUSY || rc == -EINVAL) {
4954 dev_err(&h->pdev->dev,
4955 "Unable to get major number %d for %s "
4956 "on hba %d\n", h->major, h->devname, i);
4959 if (i >= MAX_CTLR_ORIG)
4963 /* make sure the board interrupts are off */
4964 h->access.set_intr_mask(h, CCISS_INTR_OFF);
4965 rc = cciss_request_irq(h, do_cciss_msix_intr, do_cciss_intx);
4969 dev_info(&h->pdev->dev, "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
4970 h->devname, pdev->device, pci_name(pdev),
4971 h->intr[PERF_MODE_INT], dac ? "" : " not");
4973 if (cciss_allocate_cmd_pool(h))
4976 if (cciss_allocate_scatterlists(h))
4979 h->cmd_sg_list = cciss_allocate_sg_chain_blocks(h,
4980 h->chainsize, h->nr_cmds);
4981 if (!h->cmd_sg_list && h->chainsize > 0)
4984 spin_lock_init(&h->lock);
4986 /* Initialize the pdev driver private data.
4987 have it point to h. */
4988 pci_set_drvdata(pdev, h);
4989 /* command and error info recs zeroed out before
4991 memset(h->cmd_pool_bits, 0,
4992 DIV_ROUND_UP(h->nr_cmds, BITS_PER_LONG)
4993 * sizeof(unsigned long));
4996 h->highest_lun = -1;
4997 for (j = 0; j < CISS_MAX_LUN; j++) {
4999 h->gendisk[j] = NULL;
5002 /* At this point, the controller is ready to take commands.
5003 * Now, if reset_devices and the hard reset didn't work, try
5004 * the soft reset and see if that works.
5006 if (try_soft_reset) {
5008 /* This is kind of gross. We may or may not get a completion
5009 * from the soft reset command, and if we do, then the value
5010 * from the fifo may or may not be valid. So, we wait 10 secs
5011 * after the reset throwing away any completions we get during
5012 * that time. Unregister the interrupt handler and register
5013 * fake ones to scoop up any residual completions.
5015 spin_lock_irqsave(&h->lock, flags);
5016 h->access.set_intr_mask(h, CCISS_INTR_OFF);
5017 spin_unlock_irqrestore(&h->lock, flags);
5018 free_irq(h->intr[PERF_MODE_INT], h);
5019 rc = cciss_request_irq(h, cciss_msix_discard_completions,
5020 cciss_intx_discard_completions);
5022 dev_warn(&h->pdev->dev, "Failed to request_irq after "
5027 rc = cciss_kdump_soft_reset(h);
5029 dev_warn(&h->pdev->dev, "Soft reset failed.\n");
5033 dev_info(&h->pdev->dev, "Board READY.\n");
5034 dev_info(&h->pdev->dev,
5035 "Waiting for stale completions to drain.\n");
5036 h->access.set_intr_mask(h, CCISS_INTR_ON);
5038 h->access.set_intr_mask(h, CCISS_INTR_OFF);
5040 rc = controller_reset_failed(h->cfgtable);
5042 dev_info(&h->pdev->dev,
5043 "Soft reset appears to have failed.\n");
5045 /* since the controller's reset, we have to go back and re-init
5046 * everything. Easiest to just forget what we've done and do it
5049 cciss_undo_allocations_after_kdump_soft_reset(h);
5052 /* don't go to clean4, we already unallocated */
5055 goto reinit_after_soft_reset;
5058 cciss_scsi_setup(h);
5060 /* Turn the interrupts on so we can service requests */
5061 h->access.set_intr_mask(h, CCISS_INTR_ON);
5063 /* Get the firmware version */
5064 inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
5065 if (inq_buff == NULL) {
5066 dev_err(&h->pdev->dev, "out of memory\n");
5070 return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
5071 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
5072 if (return_code == IO_OK) {
5073 h->firm_ver[0] = inq_buff->data_byte[32];
5074 h->firm_ver[1] = inq_buff->data_byte[33];
5075 h->firm_ver[2] = inq_buff->data_byte[34];
5076 h->firm_ver[3] = inq_buff->data_byte[35];
5077 } else { /* send command failed */
5078 dev_warn(&h->pdev->dev, "unable to determine firmware"
5079 " version of controller\n");
5085 h->cciss_max_sectors = 8192;
5087 rebuild_lun_table(h, 1, 0);
5088 h->busy_initializing = 0;
5092 cciss_free_cmd_pool(h);
5093 cciss_free_scatterlists(h);
5094 cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
5095 free_irq(h->intr[PERF_MODE_INT], h);
5097 unregister_blkdev(h->major, h->devname);
5099 cciss_destroy_hba_sysfs_entry(h);
5101 pci_release_regions(pdev);
5102 clean_no_release_regions:
5103 h->busy_initializing = 0;
5106 * Deliberately omit pci_disable_device(): it does something nasty to
5107 * Smart Array controllers that pci_enable_device does not undo
5109 pci_set_drvdata(pdev, NULL);
5114 static void cciss_shutdown(struct pci_dev *pdev)
5120 h = pci_get_drvdata(pdev);
5121 flush_buf = kzalloc(4, GFP_KERNEL);
5123 dev_warn(&h->pdev->dev, "cache not flushed, out of memory.\n");
5126 /* write all data in the battery backed cache to disk */
5127 memset(flush_buf, 0, 4);
5128 return_code = sendcmd_withirq(h, CCISS_CACHE_FLUSH, flush_buf,
5129 4, 0, CTLR_LUNID, TYPE_CMD);
5131 if (return_code != IO_OK)
5132 dev_warn(&h->pdev->dev, "Error flushing cache\n");
5133 h->access.set_intr_mask(h, CCISS_INTR_OFF);
5134 free_irq(h->intr[PERF_MODE_INT], h);
5137 static void __devexit cciss_remove_one(struct pci_dev *pdev)
5142 if (pci_get_drvdata(pdev) == NULL) {
5143 dev_err(&pdev->dev, "Unable to remove device\n");
5147 h = pci_get_drvdata(pdev);
5149 if (hba[i] == NULL) {
5150 dev_err(&pdev->dev, "device appears to already be removed\n");
5154 mutex_lock(&h->busy_shutting_down);
5156 remove_from_scan_list(h);
5157 remove_proc_entry(h->devname, proc_cciss);
5158 unregister_blkdev(h->major, h->devname);
5160 /* remove it from the disk list */
5161 for (j = 0; j < CISS_MAX_LUN; j++) {
5162 struct gendisk *disk = h->gendisk[j];
5164 struct request_queue *q = disk->queue;
5166 if (disk->flags & GENHD_FL_UP) {
5167 cciss_destroy_ld_sysfs_entry(h, j, 1);
5171 blk_cleanup_queue(q);
5175 #ifdef CONFIG_CISS_SCSI_TAPE
5176 cciss_unregister_scsi(h); /* unhook from SCSI subsystem */
5179 cciss_shutdown(pdev);
5181 #ifdef CONFIG_PCI_MSI
5183 pci_disable_msix(h->pdev);
5184 else if (h->msi_vector)
5185 pci_disable_msi(h->pdev);
5186 #endif /* CONFIG_PCI_MSI */
5188 iounmap(h->transtable);
5189 iounmap(h->cfgtable);
5192 cciss_free_cmd_pool(h);
5193 /* Free up sg elements */
5194 for (j = 0; j < h->nr_cmds; j++)
5195 kfree(h->scatter_list[j]);
5196 kfree(h->scatter_list);
5197 cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
5198 kfree(h->blockFetchTable);
5200 pci_free_consistent(h->pdev, h->max_commands * sizeof(__u64),
5201 h->reply_pool, h->reply_pool_dhandle);
5203 * Deliberately omit pci_disable_device(): it does something nasty to
5204 * Smart Array controllers that pci_enable_device does not undo
5206 pci_release_regions(pdev);
5207 pci_set_drvdata(pdev, NULL);
5208 cciss_destroy_hba_sysfs_entry(h);
5209 mutex_unlock(&h->busy_shutting_down);
5213 static struct pci_driver cciss_pci_driver = {
5215 .probe = cciss_init_one,
5216 .remove = __devexit_p(cciss_remove_one),
5217 .id_table = cciss_pci_device_id, /* id_table */
5218 .shutdown = cciss_shutdown,
5222 * This is it. Register the PCI driver information for the cards we control
5223 * the OS will call our registered routines when it finds one of our cards.
5225 static int __init cciss_init(void)
5230 * The hardware requires that commands are aligned on a 64-bit
5231 * boundary. Given that we use pci_alloc_consistent() to allocate an
5232 * array of them, the size must be a multiple of 8 bytes.
5234 BUILD_BUG_ON(sizeof(CommandList_struct) % COMMANDLIST_ALIGNMENT);
5235 printk(KERN_INFO DRIVER_NAME "\n");
5237 err = bus_register(&cciss_bus_type);
5241 /* Start the scan thread */
5242 cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
5243 if (IS_ERR(cciss_scan_thread)) {
5244 err = PTR_ERR(cciss_scan_thread);
5245 goto err_bus_unregister;
5248 /* Register for our PCI devices */
5249 err = pci_register_driver(&cciss_pci_driver);
5251 goto err_thread_stop;
5256 kthread_stop(cciss_scan_thread);
5258 bus_unregister(&cciss_bus_type);
5263 static void __exit cciss_cleanup(void)
5267 pci_unregister_driver(&cciss_pci_driver);
5268 /* double check that all controller entrys have been removed */
5269 for (i = 0; i < MAX_CTLR; i++) {
5270 if (hba[i] != NULL) {
5271 dev_warn(&hba[i]->pdev->dev,
5272 "had to remove controller\n");
5273 cciss_remove_one(hba[i]->pdev);
5276 kthread_stop(cciss_scan_thread);
5278 remove_proc_entry("driver/cciss", NULL);
5279 bus_unregister(&cciss_bus_type);
5282 module_init(cciss_init);
5283 module_exit(cciss_cleanup);