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/smp_lock.h>
30 #include <linux/delay.h>
31 #include <linux/major.h>
33 #include <linux/bio.h>
34 #include <linux/blkpg.h>
35 #include <linux/timer.h>
36 #include <linux/proc_fs.h>
37 #include <linux/seq_file.h>
38 #include <linux/init.h>
39 #include <linux/jiffies.h>
40 #include <linux/hdreg.h>
41 #include <linux/spinlock.h>
42 #include <linux/compat.h>
43 #include <linux/mutex.h>
44 #include <asm/uaccess.h>
47 #include <linux/dma-mapping.h>
48 #include <linux/blkdev.h>
49 #include <linux/genhd.h>
50 #include <linux/completion.h>
51 #include <scsi/scsi.h>
53 #include <scsi/scsi_ioctl.h>
54 #include <linux/cdrom.h>
55 #include <linux/scatterlist.h>
56 #include <linux/kthread.h>
58 #define CCISS_DRIVER_VERSION(maj,min,submin) ((maj<<16)|(min<<8)|(submin))
59 #define DRIVER_NAME "HP CISS Driver (v 3.6.26)"
60 #define DRIVER_VERSION CCISS_DRIVER_VERSION(3, 6, 26)
62 /* Embedded module documentation macros - see modules.h */
63 MODULE_AUTHOR("Hewlett-Packard Company");
64 MODULE_DESCRIPTION("Driver for HP Smart Array Controllers");
65 MODULE_SUPPORTED_DEVICE("HP Smart Array Controllers");
66 MODULE_VERSION("3.6.26");
67 MODULE_LICENSE("GPL");
69 static int cciss_allow_hpsa;
70 module_param(cciss_allow_hpsa, int, S_IRUGO|S_IWUSR);
71 MODULE_PARM_DESC(cciss_allow_hpsa,
72 "Prevent cciss driver from accessing hardware known to be "
73 " supported by the hpsa driver");
75 #include "cciss_cmd.h"
77 #include <linux/cciss_ioctl.h>
79 /* define the PCI info for the cards we can control */
80 static const struct pci_device_id cciss_pci_device_id[] = {
81 {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISS, 0x0E11, 0x4070},
82 {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4080},
83 {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4082},
84 {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4083},
85 {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x4091},
86 {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409A},
87 {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409B},
88 {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409C},
89 {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409D},
90 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSA, 0x103C, 0x3225},
91 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSC, 0x103C, 0x3223},
92 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSC, 0x103C, 0x3234},
93 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSC, 0x103C, 0x3235},
94 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSD, 0x103C, 0x3211},
95 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSD, 0x103C, 0x3212},
96 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSD, 0x103C, 0x3213},
97 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSD, 0x103C, 0x3214},
98 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSD, 0x103C, 0x3215},
99 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSC, 0x103C, 0x3237},
100 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSC, 0x103C, 0x323D},
101 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSE, 0x103C, 0x3241},
102 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSE, 0x103C, 0x3243},
103 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSE, 0x103C, 0x3245},
104 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSE, 0x103C, 0x3247},
105 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSE, 0x103C, 0x3249},
106 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSE, 0x103C, 0x324A},
107 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSE, 0x103C, 0x324B},
108 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSE, 0x103C, 0x3250},
109 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSE, 0x103C, 0x3251},
110 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSE, 0x103C, 0x3252},
111 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSE, 0x103C, 0x3253},
112 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSE, 0x103C, 0x3254},
116 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
118 /* board_id = Subsystem Device ID & Vendor ID
119 * product = Marketing Name for the board
120 * access = Address of the struct of function pointers
122 static struct board_type products[] = {
123 {0x40700E11, "Smart Array 5300", &SA5_access},
124 {0x40800E11, "Smart Array 5i", &SA5B_access},
125 {0x40820E11, "Smart Array 532", &SA5B_access},
126 {0x40830E11, "Smart Array 5312", &SA5B_access},
127 {0x409A0E11, "Smart Array 641", &SA5_access},
128 {0x409B0E11, "Smart Array 642", &SA5_access},
129 {0x409C0E11, "Smart Array 6400", &SA5_access},
130 {0x409D0E11, "Smart Array 6400 EM", &SA5_access},
131 {0x40910E11, "Smart Array 6i", &SA5_access},
132 {0x3225103C, "Smart Array P600", &SA5_access},
133 {0x3235103C, "Smart Array P400i", &SA5_access},
134 {0x3211103C, "Smart Array E200i", &SA5_access},
135 {0x3212103C, "Smart Array E200", &SA5_access},
136 {0x3213103C, "Smart Array E200i", &SA5_access},
137 {0x3214103C, "Smart Array E200i", &SA5_access},
138 {0x3215103C, "Smart Array E200i", &SA5_access},
139 {0x3237103C, "Smart Array E500", &SA5_access},
140 /* controllers below this line are also supported by the hpsa driver. */
141 #define HPSA_BOUNDARY 0x3223103C
142 {0x3223103C, "Smart Array P800", &SA5_access},
143 {0x3234103C, "Smart Array P400", &SA5_access},
144 {0x323D103C, "Smart Array P700m", &SA5_access},
145 {0x3241103C, "Smart Array P212", &SA5_access},
146 {0x3243103C, "Smart Array P410", &SA5_access},
147 {0x3245103C, "Smart Array P410i", &SA5_access},
148 {0x3247103C, "Smart Array P411", &SA5_access},
149 {0x3249103C, "Smart Array P812", &SA5_access},
150 {0x324A103C, "Smart Array P712m", &SA5_access},
151 {0x324B103C, "Smart Array P711m", &SA5_access},
152 {0x3250103C, "Smart Array", &SA5_access},
153 {0x3251103C, "Smart Array", &SA5_access},
154 {0x3252103C, "Smart Array", &SA5_access},
155 {0x3253103C, "Smart Array", &SA5_access},
156 {0x3254103C, "Smart Array", &SA5_access},
159 /* How long to wait (in milliseconds) for board to go into simple mode */
160 #define MAX_CONFIG_WAIT 30000
161 #define MAX_IOCTL_CONFIG_WAIT 1000
163 /*define how many times we will try a command because of bus resets */
164 #define MAX_CMD_RETRIES 3
168 /* Originally cciss driver only supports 8 major numbers */
169 #define MAX_CTLR_ORIG 8
171 static ctlr_info_t *hba[MAX_CTLR];
173 static struct task_struct *cciss_scan_thread;
174 static DEFINE_MUTEX(scan_mutex);
175 static LIST_HEAD(scan_q);
177 static void do_cciss_request(struct request_queue *q);
178 static irqreturn_t do_cciss_intx(int irq, void *dev_id);
179 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id);
180 static int cciss_open(struct block_device *bdev, fmode_t mode);
181 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode);
182 static int cciss_release(struct gendisk *disk, fmode_t mode);
183 static int do_ioctl(struct block_device *bdev, fmode_t mode,
184 unsigned int cmd, unsigned long arg);
185 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
186 unsigned int cmd, unsigned long arg);
187 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
189 static int cciss_revalidate(struct gendisk *disk);
190 static int rebuild_lun_table(ctlr_info_t *h, int first_time, int via_ioctl);
191 static int deregister_disk(ctlr_info_t *h, int drv_index,
192 int clear_all, int via_ioctl);
194 static void cciss_read_capacity(int ctlr, int logvol,
195 sector_t *total_size, unsigned int *block_size);
196 static void cciss_read_capacity_16(int ctlr, int logvol,
197 sector_t *total_size, unsigned int *block_size);
198 static void cciss_geometry_inquiry(int ctlr, int logvol,
200 unsigned int block_size, InquiryData_struct *inq_buff,
201 drive_info_struct *drv);
202 static void __devinit cciss_interrupt_mode(ctlr_info_t *);
203 static void start_io(ctlr_info_t *h);
204 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
205 __u8 page_code, unsigned char scsi3addr[],
207 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
209 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c);
211 static int add_to_scan_list(struct ctlr_info *h);
212 static int scan_thread(void *data);
213 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c);
214 static void cciss_hba_release(struct device *dev);
215 static void cciss_device_release(struct device *dev);
216 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index);
217 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index);
218 static inline u32 next_command(ctlr_info_t *h);
219 static int __devinit cciss_find_cfg_addrs(struct pci_dev *pdev,
220 void __iomem *vaddr, u32 *cfg_base_addr, u64 *cfg_base_addr_index,
222 static int __devinit cciss_pci_find_memory_BAR(struct pci_dev *pdev,
223 unsigned long *memory_bar);
226 /* performant mode helper functions */
227 static void calc_bucket_map(int *bucket, int num_buckets, int nsgs,
229 static void cciss_put_controller_into_performant_mode(ctlr_info_t *h);
231 #ifdef CONFIG_PROC_FS
232 static void cciss_procinit(int i);
234 static void cciss_procinit(int i)
237 #endif /* CONFIG_PROC_FS */
240 static int cciss_compat_ioctl(struct block_device *, fmode_t,
241 unsigned, unsigned long);
244 static const struct block_device_operations cciss_fops = {
245 .owner = THIS_MODULE,
246 .open = cciss_unlocked_open,
247 .release = cciss_release,
249 .getgeo = cciss_getgeo,
251 .compat_ioctl = cciss_compat_ioctl,
253 .revalidate_disk = cciss_revalidate,
256 /* set_performant_mode: Modify the tag for cciss performant
257 * set bit 0 for pull model, bits 3-1 for block fetch
260 static void set_performant_mode(ctlr_info_t *h, CommandList_struct *c)
262 if (likely(h->transMethod == CFGTBL_Trans_Performant))
263 c->busaddr |= 1 | (h->blockFetchTable[c->Header.SGList] << 1);
267 * Enqueuing and dequeuing functions for cmdlists.
269 static inline void addQ(struct hlist_head *list, CommandList_struct *c)
271 hlist_add_head(&c->list, list);
274 static inline void removeQ(CommandList_struct *c)
277 * After kexec/dump some commands might still
278 * be in flight, which the firmware will try
279 * to complete. Resetting the firmware doesn't work
280 * with old fw revisions, so we have to mark
281 * them off as 'stale' to prevent the driver from
284 if (WARN_ON(hlist_unhashed(&c->list))) {
285 c->cmd_type = CMD_MSG_STALE;
289 hlist_del_init(&c->list);
292 static void enqueue_cmd_and_start_io(ctlr_info_t *h,
293 CommandList_struct *c)
296 set_performant_mode(h, c);
297 spin_lock_irqsave(&h->lock, flags);
301 spin_unlock_irqrestore(&h->lock, flags);
304 static void cciss_free_sg_chain_blocks(SGDescriptor_struct **cmd_sg_list,
311 for (i = 0; i < nr_cmds; i++) {
312 kfree(cmd_sg_list[i]);
313 cmd_sg_list[i] = NULL;
318 static SGDescriptor_struct **cciss_allocate_sg_chain_blocks(
319 ctlr_info_t *h, int chainsize, int nr_cmds)
322 SGDescriptor_struct **cmd_sg_list;
327 cmd_sg_list = kmalloc(sizeof(*cmd_sg_list) * nr_cmds, GFP_KERNEL);
331 /* Build up chain blocks for each command */
332 for (j = 0; j < nr_cmds; j++) {
333 /* Need a block of chainsized s/g elements. */
334 cmd_sg_list[j] = kmalloc((chainsize *
335 sizeof(*cmd_sg_list[j])), GFP_KERNEL);
336 if (!cmd_sg_list[j]) {
337 dev_err(&h->pdev->dev, "Cannot get memory "
338 "for s/g chains.\n");
344 cciss_free_sg_chain_blocks(cmd_sg_list, nr_cmds);
348 static void cciss_unmap_sg_chain_block(ctlr_info_t *h, CommandList_struct *c)
350 SGDescriptor_struct *chain_sg;
353 if (c->Header.SGTotal <= h->max_cmd_sgentries)
356 chain_sg = &c->SG[h->max_cmd_sgentries - 1];
357 temp64.val32.lower = chain_sg->Addr.lower;
358 temp64.val32.upper = chain_sg->Addr.upper;
359 pci_unmap_single(h->pdev, temp64.val, chain_sg->Len, PCI_DMA_TODEVICE);
362 static void cciss_map_sg_chain_block(ctlr_info_t *h, CommandList_struct *c,
363 SGDescriptor_struct *chain_block, int len)
365 SGDescriptor_struct *chain_sg;
368 chain_sg = &c->SG[h->max_cmd_sgentries - 1];
369 chain_sg->Ext = CCISS_SG_CHAIN;
371 temp64.val = pci_map_single(h->pdev, chain_block, len,
373 chain_sg->Addr.lower = temp64.val32.lower;
374 chain_sg->Addr.upper = temp64.val32.upper;
377 #include "cciss_scsi.c" /* For SCSI tape support */
379 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
382 #define RAID_UNKNOWN (sizeof(raid_label) / sizeof(raid_label[0])-1)
384 #ifdef CONFIG_PROC_FS
387 * Report information about this controller.
389 #define ENG_GIG 1000000000
390 #define ENG_GIG_FACTOR (ENG_GIG/512)
391 #define ENGAGE_SCSI "engage scsi"
393 static struct proc_dir_entry *proc_cciss;
395 static void cciss_seq_show_header(struct seq_file *seq)
397 ctlr_info_t *h = seq->private;
399 seq_printf(seq, "%s: HP %s Controller\n"
400 "Board ID: 0x%08lx\n"
401 "Firmware Version: %c%c%c%c\n"
403 "Logical drives: %d\n"
404 "Current Q depth: %d\n"
405 "Current # commands on controller: %d\n"
406 "Max Q depth since init: %d\n"
407 "Max # commands on controller since init: %d\n"
408 "Max SG entries since init: %d\n",
411 (unsigned long)h->board_id,
412 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
413 h->firm_ver[3], (unsigned int)h->intr[PERF_MODE_INT],
415 h->Qdepth, h->commands_outstanding,
416 h->maxQsinceinit, h->max_outstanding, h->maxSG);
418 #ifdef CONFIG_CISS_SCSI_TAPE
419 cciss_seq_tape_report(seq, h->ctlr);
420 #endif /* CONFIG_CISS_SCSI_TAPE */
423 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
425 ctlr_info_t *h = seq->private;
426 unsigned ctlr = h->ctlr;
429 /* prevent displaying bogus info during configuration
430 * or deconfiguration of a logical volume
432 spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
433 if (h->busy_configuring) {
434 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
435 return ERR_PTR(-EBUSY);
437 h->busy_configuring = 1;
438 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
441 cciss_seq_show_header(seq);
446 static int cciss_seq_show(struct seq_file *seq, void *v)
448 sector_t vol_sz, vol_sz_frac;
449 ctlr_info_t *h = seq->private;
450 unsigned ctlr = h->ctlr;
452 drive_info_struct *drv = h->drv[*pos];
454 if (*pos > h->highest_lun)
457 if (drv == NULL) /* it's possible for h->drv[] to have holes. */
463 vol_sz = drv->nr_blocks;
464 vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
466 sector_div(vol_sz_frac, ENG_GIG_FACTOR);
468 if (drv->raid_level < 0 || drv->raid_level > RAID_UNKNOWN)
469 drv->raid_level = RAID_UNKNOWN;
470 seq_printf(seq, "cciss/c%dd%d:"
471 "\t%4u.%02uGB\tRAID %s\n",
472 ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
473 raid_label[drv->raid_level]);
477 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
479 ctlr_info_t *h = seq->private;
481 if (*pos > h->highest_lun)
488 static void cciss_seq_stop(struct seq_file *seq, void *v)
490 ctlr_info_t *h = seq->private;
492 /* Only reset h->busy_configuring if we succeeded in setting
493 * it during cciss_seq_start. */
494 if (v == ERR_PTR(-EBUSY))
497 h->busy_configuring = 0;
500 static const struct seq_operations cciss_seq_ops = {
501 .start = cciss_seq_start,
502 .show = cciss_seq_show,
503 .next = cciss_seq_next,
504 .stop = cciss_seq_stop,
507 static int cciss_seq_open(struct inode *inode, struct file *file)
509 int ret = seq_open(file, &cciss_seq_ops);
510 struct seq_file *seq = file->private_data;
513 seq->private = PDE(inode)->data;
519 cciss_proc_write(struct file *file, const char __user *buf,
520 size_t length, loff_t *ppos)
525 #ifndef CONFIG_CISS_SCSI_TAPE
529 if (!buf || length > PAGE_SIZE - 1)
532 buffer = (char *)__get_free_page(GFP_KERNEL);
537 if (copy_from_user(buffer, buf, length))
539 buffer[length] = '\0';
541 #ifdef CONFIG_CISS_SCSI_TAPE
542 if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
543 struct seq_file *seq = file->private_data;
544 ctlr_info_t *h = seq->private;
546 err = cciss_engage_scsi(h->ctlr);
550 #endif /* CONFIG_CISS_SCSI_TAPE */
552 /* might be nice to have "disengage" too, but it's not
553 safely possible. (only 1 module use count, lock issues.) */
556 free_page((unsigned long)buffer);
560 static const struct file_operations cciss_proc_fops = {
561 .owner = THIS_MODULE,
562 .open = cciss_seq_open,
565 .release = seq_release,
566 .write = cciss_proc_write,
569 static void __devinit cciss_procinit(int i)
571 struct proc_dir_entry *pde;
573 if (proc_cciss == NULL)
574 proc_cciss = proc_mkdir("driver/cciss", NULL);
577 pde = proc_create_data(hba[i]->devname, S_IWUSR | S_IRUSR | S_IRGRP |
579 &cciss_proc_fops, hba[i]);
581 #endif /* CONFIG_PROC_FS */
583 #define MAX_PRODUCT_NAME_LEN 19
585 #define to_hba(n) container_of(n, struct ctlr_info, dev)
586 #define to_drv(n) container_of(n, drive_info_struct, dev)
588 static ssize_t host_store_rescan(struct device *dev,
589 struct device_attribute *attr,
590 const char *buf, size_t count)
592 struct ctlr_info *h = to_hba(dev);
595 wake_up_process(cciss_scan_thread);
596 wait_for_completion_interruptible(&h->scan_wait);
600 static DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
602 static ssize_t dev_show_unique_id(struct device *dev,
603 struct device_attribute *attr,
606 drive_info_struct *drv = to_drv(dev);
607 struct ctlr_info *h = to_hba(drv->dev.parent);
612 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
613 if (h->busy_configuring)
616 memcpy(sn, drv->serial_no, sizeof(sn));
617 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
622 return snprintf(buf, 16 * 2 + 2,
623 "%02X%02X%02X%02X%02X%02X%02X%02X"
624 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
625 sn[0], sn[1], sn[2], sn[3],
626 sn[4], sn[5], sn[6], sn[7],
627 sn[8], sn[9], sn[10], sn[11],
628 sn[12], sn[13], sn[14], sn[15]);
630 static DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
632 static ssize_t dev_show_vendor(struct device *dev,
633 struct device_attribute *attr,
636 drive_info_struct *drv = to_drv(dev);
637 struct ctlr_info *h = to_hba(drv->dev.parent);
638 char vendor[VENDOR_LEN + 1];
642 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
643 if (h->busy_configuring)
646 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
647 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
652 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
654 static DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
656 static ssize_t dev_show_model(struct device *dev,
657 struct device_attribute *attr,
660 drive_info_struct *drv = to_drv(dev);
661 struct ctlr_info *h = to_hba(drv->dev.parent);
662 char model[MODEL_LEN + 1];
666 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
667 if (h->busy_configuring)
670 memcpy(model, drv->model, MODEL_LEN + 1);
671 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
676 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
678 static DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
680 static ssize_t dev_show_rev(struct device *dev,
681 struct device_attribute *attr,
684 drive_info_struct *drv = to_drv(dev);
685 struct ctlr_info *h = to_hba(drv->dev.parent);
686 char rev[REV_LEN + 1];
690 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
691 if (h->busy_configuring)
694 memcpy(rev, drv->rev, REV_LEN + 1);
695 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
700 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
702 static DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
704 static ssize_t cciss_show_lunid(struct device *dev,
705 struct device_attribute *attr, char *buf)
707 drive_info_struct *drv = to_drv(dev);
708 struct ctlr_info *h = to_hba(drv->dev.parent);
710 unsigned char lunid[8];
712 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
713 if (h->busy_configuring) {
714 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
718 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
721 memcpy(lunid, drv->LunID, sizeof(lunid));
722 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
723 return snprintf(buf, 20, "0x%02x%02x%02x%02x%02x%02x%02x%02x\n",
724 lunid[0], lunid[1], lunid[2], lunid[3],
725 lunid[4], lunid[5], lunid[6], lunid[7]);
727 static DEVICE_ATTR(lunid, S_IRUGO, cciss_show_lunid, NULL);
729 static ssize_t cciss_show_raid_level(struct device *dev,
730 struct device_attribute *attr, char *buf)
732 drive_info_struct *drv = to_drv(dev);
733 struct ctlr_info *h = to_hba(drv->dev.parent);
737 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
738 if (h->busy_configuring) {
739 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
742 raid = drv->raid_level;
743 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
744 if (raid < 0 || raid > RAID_UNKNOWN)
747 return snprintf(buf, strlen(raid_label[raid]) + 7, "RAID %s\n",
750 static DEVICE_ATTR(raid_level, S_IRUGO, cciss_show_raid_level, NULL);
752 static ssize_t cciss_show_usage_count(struct device *dev,
753 struct device_attribute *attr, char *buf)
755 drive_info_struct *drv = to_drv(dev);
756 struct ctlr_info *h = to_hba(drv->dev.parent);
760 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
761 if (h->busy_configuring) {
762 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
765 count = drv->usage_count;
766 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
767 return snprintf(buf, 20, "%d\n", count);
769 static DEVICE_ATTR(usage_count, S_IRUGO, cciss_show_usage_count, NULL);
771 static struct attribute *cciss_host_attrs[] = {
772 &dev_attr_rescan.attr,
776 static struct attribute_group cciss_host_attr_group = {
777 .attrs = cciss_host_attrs,
780 static const struct attribute_group *cciss_host_attr_groups[] = {
781 &cciss_host_attr_group,
785 static struct device_type cciss_host_type = {
786 .name = "cciss_host",
787 .groups = cciss_host_attr_groups,
788 .release = cciss_hba_release,
791 static struct attribute *cciss_dev_attrs[] = {
792 &dev_attr_unique_id.attr,
793 &dev_attr_model.attr,
794 &dev_attr_vendor.attr,
796 &dev_attr_lunid.attr,
797 &dev_attr_raid_level.attr,
798 &dev_attr_usage_count.attr,
802 static struct attribute_group cciss_dev_attr_group = {
803 .attrs = cciss_dev_attrs,
806 static const struct attribute_group *cciss_dev_attr_groups[] = {
807 &cciss_dev_attr_group,
811 static struct device_type cciss_dev_type = {
812 .name = "cciss_device",
813 .groups = cciss_dev_attr_groups,
814 .release = cciss_device_release,
817 static struct bus_type cciss_bus_type = {
822 * cciss_hba_release is called when the reference count
823 * of h->dev goes to zero.
825 static void cciss_hba_release(struct device *dev)
828 * nothing to do, but need this to avoid a warning
829 * about not having a release handler from lib/kref.c.
834 * Initialize sysfs entry for each controller. This sets up and registers
835 * the 'cciss#' directory for each individual controller under
836 * /sys/bus/pci/devices/<dev>/.
838 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
840 device_initialize(&h->dev);
841 h->dev.type = &cciss_host_type;
842 h->dev.bus = &cciss_bus_type;
843 dev_set_name(&h->dev, "%s", h->devname);
844 h->dev.parent = &h->pdev->dev;
846 return device_add(&h->dev);
850 * Remove sysfs entries for an hba.
852 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
855 put_device(&h->dev); /* final put. */
858 /* cciss_device_release is called when the reference count
859 * of h->drv[x]dev goes to zero.
861 static void cciss_device_release(struct device *dev)
863 drive_info_struct *drv = to_drv(dev);
868 * Initialize sysfs for each logical drive. This sets up and registers
869 * the 'c#d#' directory for each individual logical drive under
870 * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
871 * /sys/block/cciss!c#d# to this entry.
873 static long cciss_create_ld_sysfs_entry(struct ctlr_info *h,
878 if (h->drv[drv_index]->device_initialized)
881 dev = &h->drv[drv_index]->dev;
882 device_initialize(dev);
883 dev->type = &cciss_dev_type;
884 dev->bus = &cciss_bus_type;
885 dev_set_name(dev, "c%dd%d", h->ctlr, drv_index);
886 dev->parent = &h->dev;
887 h->drv[drv_index]->device_initialized = 1;
888 return device_add(dev);
892 * Remove sysfs entries for a logical drive.
894 static void cciss_destroy_ld_sysfs_entry(struct ctlr_info *h, int drv_index,
897 struct device *dev = &h->drv[drv_index]->dev;
899 /* special case for c*d0, we only destroy it on controller exit */
900 if (drv_index == 0 && !ctlr_exiting)
904 put_device(dev); /* the "final" put. */
905 h->drv[drv_index] = NULL;
909 * For operations that cannot sleep, a command block is allocated at init,
910 * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
911 * which ones are free or in use. For operations that can wait for kmalloc
912 * to possible sleep, this routine can be called with get_from_pool set to 0.
913 * cmd_free() MUST be called with a got_from_pool set to 0 if cmd_alloc was.
915 static CommandList_struct *cmd_alloc(ctlr_info_t *h, int get_from_pool)
917 CommandList_struct *c;
920 dma_addr_t cmd_dma_handle, err_dma_handle;
922 if (!get_from_pool) {
923 c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
924 sizeof(CommandList_struct), &cmd_dma_handle);
927 memset(c, 0, sizeof(CommandList_struct));
931 c->err_info = (ErrorInfo_struct *)
932 pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
935 if (c->err_info == NULL) {
936 pci_free_consistent(h->pdev,
937 sizeof(CommandList_struct), c, cmd_dma_handle);
940 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
941 } else { /* get it out of the controllers pool */
944 i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
947 } while (test_and_set_bit
948 (i & (BITS_PER_LONG - 1),
949 h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
951 printk(KERN_DEBUG "cciss: using command buffer %d\n", i);
954 memset(c, 0, sizeof(CommandList_struct));
955 cmd_dma_handle = h->cmd_pool_dhandle
956 + i * sizeof(CommandList_struct);
957 c->err_info = h->errinfo_pool + i;
958 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
959 err_dma_handle = h->errinfo_pool_dhandle
960 + i * sizeof(ErrorInfo_struct);
966 INIT_HLIST_NODE(&c->list);
967 c->busaddr = (__u32) cmd_dma_handle;
968 temp64.val = (__u64) err_dma_handle;
969 c->ErrDesc.Addr.lower = temp64.val32.lower;
970 c->ErrDesc.Addr.upper = temp64.val32.upper;
971 c->ErrDesc.Len = sizeof(ErrorInfo_struct);
978 * Frees a command block that was previously allocated with cmd_alloc().
980 static void cmd_free(ctlr_info_t *h, CommandList_struct *c, int got_from_pool)
985 if (!got_from_pool) {
986 temp64.val32.lower = c->ErrDesc.Addr.lower;
987 temp64.val32.upper = c->ErrDesc.Addr.upper;
988 pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
989 c->err_info, (dma_addr_t) temp64.val);
990 pci_free_consistent(h->pdev, sizeof(CommandList_struct),
991 c, (dma_addr_t) c->busaddr);
994 clear_bit(i & (BITS_PER_LONG - 1),
995 h->cmd_pool_bits + (i / BITS_PER_LONG));
1000 static inline ctlr_info_t *get_host(struct gendisk *disk)
1002 return disk->queue->queuedata;
1005 static inline drive_info_struct *get_drv(struct gendisk *disk)
1007 return disk->private_data;
1011 * Open. Make sure the device is really there.
1013 static int cciss_open(struct block_device *bdev, fmode_t mode)
1015 ctlr_info_t *host = get_host(bdev->bd_disk);
1016 drive_info_struct *drv = get_drv(bdev->bd_disk);
1019 printk(KERN_DEBUG "cciss_open %s\n", bdev->bd_disk->disk_name);
1020 #endif /* CCISS_DEBUG */
1022 if (drv->busy_configuring)
1025 * Root is allowed to open raw volume zero even if it's not configured
1026 * so array config can still work. Root is also allowed to open any
1027 * volume that has a LUN ID, so it can issue IOCTL to reread the
1028 * disk information. I don't think I really like this
1029 * but I'm already using way to many device nodes to claim another one
1030 * for "raw controller".
1032 if (drv->heads == 0) {
1033 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
1034 /* if not node 0 make sure it is a partition = 0 */
1035 if (MINOR(bdev->bd_dev) & 0x0f) {
1037 /* if it is, make sure we have a LUN ID */
1038 } else if (memcmp(drv->LunID, CTLR_LUNID,
1039 sizeof(drv->LunID))) {
1043 if (!capable(CAP_SYS_ADMIN))
1047 host->usage_count++;
1051 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode)
1056 ret = cciss_open(bdev, mode);
1063 * Close. Sync first.
1065 static int cciss_release(struct gendisk *disk, fmode_t mode)
1068 drive_info_struct *drv;
1071 host = get_host(disk);
1072 drv = get_drv(disk);
1075 printk(KERN_DEBUG "cciss_release %s\n", disk->disk_name);
1076 #endif /* CCISS_DEBUG */
1079 host->usage_count--;
1084 static int do_ioctl(struct block_device *bdev, fmode_t mode,
1085 unsigned cmd, unsigned long arg)
1089 ret = cciss_ioctl(bdev, mode, cmd, arg);
1094 #ifdef CONFIG_COMPAT
1096 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1097 unsigned cmd, unsigned long arg);
1098 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1099 unsigned cmd, unsigned long arg);
1101 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
1102 unsigned cmd, unsigned long arg)
1105 case CCISS_GETPCIINFO:
1106 case CCISS_GETINTINFO:
1107 case CCISS_SETINTINFO:
1108 case CCISS_GETNODENAME:
1109 case CCISS_SETNODENAME:
1110 case CCISS_GETHEARTBEAT:
1111 case CCISS_GETBUSTYPES:
1112 case CCISS_GETFIRMVER:
1113 case CCISS_GETDRIVVER:
1114 case CCISS_REVALIDVOLS:
1115 case CCISS_DEREGDISK:
1116 case CCISS_REGNEWDISK:
1118 case CCISS_RESCANDISK:
1119 case CCISS_GETLUNINFO:
1120 return do_ioctl(bdev, mode, cmd, arg);
1122 case CCISS_PASSTHRU32:
1123 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
1124 case CCISS_BIG_PASSTHRU32:
1125 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
1128 return -ENOIOCTLCMD;
1132 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1133 unsigned cmd, unsigned long arg)
1135 IOCTL32_Command_struct __user *arg32 =
1136 (IOCTL32_Command_struct __user *) arg;
1137 IOCTL_Command_struct arg64;
1138 IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
1144 copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1145 sizeof(arg64.LUN_info));
1147 copy_from_user(&arg64.Request, &arg32->Request,
1148 sizeof(arg64.Request));
1150 copy_from_user(&arg64.error_info, &arg32->error_info,
1151 sizeof(arg64.error_info));
1152 err |= get_user(arg64.buf_size, &arg32->buf_size);
1153 err |= get_user(cp, &arg32->buf);
1154 arg64.buf = compat_ptr(cp);
1155 err |= copy_to_user(p, &arg64, sizeof(arg64));
1160 err = do_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
1164 copy_in_user(&arg32->error_info, &p->error_info,
1165 sizeof(arg32->error_info));
1171 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1172 unsigned cmd, unsigned long arg)
1174 BIG_IOCTL32_Command_struct __user *arg32 =
1175 (BIG_IOCTL32_Command_struct __user *) arg;
1176 BIG_IOCTL_Command_struct arg64;
1177 BIG_IOCTL_Command_struct __user *p =
1178 compat_alloc_user_space(sizeof(arg64));
1184 copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1185 sizeof(arg64.LUN_info));
1187 copy_from_user(&arg64.Request, &arg32->Request,
1188 sizeof(arg64.Request));
1190 copy_from_user(&arg64.error_info, &arg32->error_info,
1191 sizeof(arg64.error_info));
1192 err |= get_user(arg64.buf_size, &arg32->buf_size);
1193 err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1194 err |= get_user(cp, &arg32->buf);
1195 arg64.buf = compat_ptr(cp);
1196 err |= copy_to_user(p, &arg64, sizeof(arg64));
1201 err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1205 copy_in_user(&arg32->error_info, &p->error_info,
1206 sizeof(arg32->error_info));
1213 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1215 drive_info_struct *drv = get_drv(bdev->bd_disk);
1217 if (!drv->cylinders)
1220 geo->heads = drv->heads;
1221 geo->sectors = drv->sectors;
1222 geo->cylinders = drv->cylinders;
1226 static void check_ioctl_unit_attention(ctlr_info_t *host, CommandList_struct *c)
1228 if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1229 c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1230 (void)check_for_unit_attention(host, c);
1235 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1236 unsigned int cmd, unsigned long arg)
1238 struct gendisk *disk = bdev->bd_disk;
1239 ctlr_info_t *host = get_host(disk);
1240 drive_info_struct *drv = get_drv(disk);
1241 int ctlr = host->ctlr;
1242 void __user *argp = (void __user *)arg;
1245 printk(KERN_DEBUG "cciss_ioctl: Called with cmd=%x %lx\n", cmd, arg);
1246 #endif /* CCISS_DEBUG */
1249 case CCISS_GETPCIINFO:
1251 cciss_pci_info_struct pciinfo;
1255 pciinfo.domain = pci_domain_nr(host->pdev->bus);
1256 pciinfo.bus = host->pdev->bus->number;
1257 pciinfo.dev_fn = host->pdev->devfn;
1258 pciinfo.board_id = host->board_id;
1260 (argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1264 case CCISS_GETINTINFO:
1266 cciss_coalint_struct intinfo;
1270 readl(&host->cfgtable->HostWrite.CoalIntDelay);
1272 readl(&host->cfgtable->HostWrite.CoalIntCount);
1274 (argp, &intinfo, sizeof(cciss_coalint_struct)))
1278 case CCISS_SETINTINFO:
1280 cciss_coalint_struct intinfo;
1281 unsigned long flags;
1286 if (!capable(CAP_SYS_ADMIN))
1289 (&intinfo, argp, sizeof(cciss_coalint_struct)))
1291 if ((intinfo.delay == 0) && (intinfo.count == 0))
1293 // printk("cciss_ioctl: delay and count cannot be 0\n");
1296 spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1297 /* Update the field, and then ring the doorbell */
1298 writel(intinfo.delay,
1299 &(host->cfgtable->HostWrite.CoalIntDelay));
1300 writel(intinfo.count,
1301 &(host->cfgtable->HostWrite.CoalIntCount));
1302 writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1304 for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1305 if (!(readl(host->vaddr + SA5_DOORBELL)
1306 & CFGTBL_ChangeReq))
1308 /* delay and try again */
1311 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1312 if (i >= MAX_IOCTL_CONFIG_WAIT)
1316 case CCISS_GETNODENAME:
1318 NodeName_type NodeName;
1323 for (i = 0; i < 16; i++)
1325 readb(&host->cfgtable->ServerName[i]);
1326 if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1330 case CCISS_SETNODENAME:
1332 NodeName_type NodeName;
1333 unsigned long flags;
1338 if (!capable(CAP_SYS_ADMIN))
1342 (NodeName, argp, sizeof(NodeName_type)))
1345 spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1347 /* Update the field, and then ring the doorbell */
1348 for (i = 0; i < 16; i++)
1350 &host->cfgtable->ServerName[i]);
1352 writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1354 for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1355 if (!(readl(host->vaddr + SA5_DOORBELL)
1356 & CFGTBL_ChangeReq))
1358 /* delay and try again */
1361 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1362 if (i >= MAX_IOCTL_CONFIG_WAIT)
1367 case CCISS_GETHEARTBEAT:
1369 Heartbeat_type heartbeat;
1373 heartbeat = readl(&host->cfgtable->HeartBeat);
1375 (argp, &heartbeat, sizeof(Heartbeat_type)))
1379 case CCISS_GETBUSTYPES:
1381 BusTypes_type BusTypes;
1385 BusTypes = readl(&host->cfgtable->BusTypes);
1387 (argp, &BusTypes, sizeof(BusTypes_type)))
1391 case CCISS_GETFIRMVER:
1393 FirmwareVer_type firmware;
1397 memcpy(firmware, host->firm_ver, 4);
1400 (argp, firmware, sizeof(FirmwareVer_type)))
1404 case CCISS_GETDRIVVER:
1406 DriverVer_type DriverVer = DRIVER_VERSION;
1412 (argp, &DriverVer, sizeof(DriverVer_type)))
1417 case CCISS_DEREGDISK:
1419 case CCISS_REVALIDVOLS:
1420 return rebuild_lun_table(host, 0, 1);
1422 case CCISS_GETLUNINFO:{
1423 LogvolInfo_struct luninfo;
1425 memcpy(&luninfo.LunID, drv->LunID,
1426 sizeof(luninfo.LunID));
1427 luninfo.num_opens = drv->usage_count;
1428 luninfo.num_parts = 0;
1429 if (copy_to_user(argp, &luninfo,
1430 sizeof(LogvolInfo_struct)))
1434 case CCISS_PASSTHRU:
1436 IOCTL_Command_struct iocommand;
1437 CommandList_struct *c;
1440 DECLARE_COMPLETION_ONSTACK(wait);
1445 if (!capable(CAP_SYS_RAWIO))
1449 (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1451 if ((iocommand.buf_size < 1) &&
1452 (iocommand.Request.Type.Direction != XFER_NONE)) {
1455 #if 0 /* 'buf_size' member is 16-bits, and always smaller than kmalloc limit */
1456 /* Check kmalloc limits */
1457 if (iocommand.buf_size > 128000)
1460 if (iocommand.buf_size > 0) {
1461 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1465 if (iocommand.Request.Type.Direction == XFER_WRITE) {
1466 /* Copy the data into the buffer we created */
1468 (buff, iocommand.buf, iocommand.buf_size)) {
1473 memset(buff, 0, iocommand.buf_size);
1475 if ((c = cmd_alloc(host, 0)) == NULL) {
1479 /* Fill in the command type */
1480 c->cmd_type = CMD_IOCTL_PEND;
1481 /* Fill in Command Header */
1482 c->Header.ReplyQueue = 0; /* unused in simple mode */
1483 if (iocommand.buf_size > 0) /* buffer to fill */
1485 c->Header.SGList = 1;
1486 c->Header.SGTotal = 1;
1487 } else /* no buffers to fill */
1489 c->Header.SGList = 0;
1490 c->Header.SGTotal = 0;
1492 c->Header.LUN = iocommand.LUN_info;
1493 /* use the kernel address the cmd block for tag */
1494 c->Header.Tag.lower = c->busaddr;
1496 /* Fill in Request block */
1497 c->Request = iocommand.Request;
1499 /* Fill in the scatter gather information */
1500 if (iocommand.buf_size > 0) {
1501 temp64.val = pci_map_single(host->pdev, buff,
1503 PCI_DMA_BIDIRECTIONAL);
1504 c->SG[0].Addr.lower = temp64.val32.lower;
1505 c->SG[0].Addr.upper = temp64.val32.upper;
1506 c->SG[0].Len = iocommand.buf_size;
1507 c->SG[0].Ext = 0; /* we are not chaining */
1511 enqueue_cmd_and_start_io(host, c);
1512 wait_for_completion(&wait);
1514 /* unlock the buffers from DMA */
1515 temp64.val32.lower = c->SG[0].Addr.lower;
1516 temp64.val32.upper = c->SG[0].Addr.upper;
1517 pci_unmap_single(host->pdev, (dma_addr_t) temp64.val,
1519 PCI_DMA_BIDIRECTIONAL);
1521 check_ioctl_unit_attention(host, c);
1523 /* Copy the error information out */
1524 iocommand.error_info = *(c->err_info);
1526 (argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1528 cmd_free(host, c, 0);
1532 if (iocommand.Request.Type.Direction == XFER_READ) {
1533 /* Copy the data out of the buffer we created */
1535 (iocommand.buf, buff, iocommand.buf_size)) {
1537 cmd_free(host, c, 0);
1542 cmd_free(host, c, 0);
1545 case CCISS_BIG_PASSTHRU:{
1546 BIG_IOCTL_Command_struct *ioc;
1547 CommandList_struct *c;
1548 unsigned char **buff = NULL;
1549 int *buff_size = NULL;
1554 DECLARE_COMPLETION_ONSTACK(wait);
1557 BYTE __user *data_ptr;
1561 if (!capable(CAP_SYS_RAWIO))
1563 ioc = (BIG_IOCTL_Command_struct *)
1564 kmalloc(sizeof(*ioc), GFP_KERNEL);
1569 if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1573 if ((ioc->buf_size < 1) &&
1574 (ioc->Request.Type.Direction != XFER_NONE)) {
1578 /* Check kmalloc limits using all SGs */
1579 if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1583 if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1588 kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1593 buff_size = kmalloc(MAXSGENTRIES * sizeof(int),
1599 left = ioc->buf_size;
1600 data_ptr = ioc->buf;
1603 ioc->malloc_size) ? ioc->
1605 buff_size[sg_used] = sz;
1606 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1607 if (buff[sg_used] == NULL) {
1611 if (ioc->Request.Type.Direction == XFER_WRITE) {
1613 (buff[sg_used], data_ptr, sz)) {
1618 memset(buff[sg_used], 0, sz);
1624 if ((c = cmd_alloc(host, 0)) == NULL) {
1628 c->cmd_type = CMD_IOCTL_PEND;
1629 c->Header.ReplyQueue = 0;
1631 if (ioc->buf_size > 0) {
1632 c->Header.SGList = sg_used;
1633 c->Header.SGTotal = sg_used;
1635 c->Header.SGList = 0;
1636 c->Header.SGTotal = 0;
1638 c->Header.LUN = ioc->LUN_info;
1639 c->Header.Tag.lower = c->busaddr;
1641 c->Request = ioc->Request;
1642 if (ioc->buf_size > 0) {
1643 for (i = 0; i < sg_used; i++) {
1645 pci_map_single(host->pdev, buff[i],
1647 PCI_DMA_BIDIRECTIONAL);
1648 c->SG[i].Addr.lower =
1650 c->SG[i].Addr.upper =
1652 c->SG[i].Len = buff_size[i];
1653 c->SG[i].Ext = 0; /* we are not chaining */
1657 enqueue_cmd_and_start_io(host, c);
1658 wait_for_completion(&wait);
1659 /* unlock the buffers from DMA */
1660 for (i = 0; i < sg_used; i++) {
1661 temp64.val32.lower = c->SG[i].Addr.lower;
1662 temp64.val32.upper = c->SG[i].Addr.upper;
1663 pci_unmap_single(host->pdev,
1664 (dma_addr_t) temp64.val, buff_size[i],
1665 PCI_DMA_BIDIRECTIONAL);
1667 check_ioctl_unit_attention(host, c);
1668 /* Copy the error information out */
1669 ioc->error_info = *(c->err_info);
1670 if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1671 cmd_free(host, c, 0);
1675 if (ioc->Request.Type.Direction == XFER_READ) {
1676 /* Copy the data out of the buffer we created */
1677 BYTE __user *ptr = ioc->buf;
1678 for (i = 0; i < sg_used; i++) {
1680 (ptr, buff[i], buff_size[i])) {
1681 cmd_free(host, c, 0);
1685 ptr += buff_size[i];
1688 cmd_free(host, c, 0);
1692 for (i = 0; i < sg_used; i++)
1701 /* scsi_cmd_ioctl handles these, below, though some are not */
1702 /* very meaningful for cciss. SG_IO is the main one people want. */
1704 case SG_GET_VERSION_NUM:
1705 case SG_SET_TIMEOUT:
1706 case SG_GET_TIMEOUT:
1707 case SG_GET_RESERVED_SIZE:
1708 case SG_SET_RESERVED_SIZE:
1709 case SG_EMULATED_HOST:
1711 case SCSI_IOCTL_SEND_COMMAND:
1712 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
1714 /* scsi_cmd_ioctl would normally handle these, below, but */
1715 /* they aren't a good fit for cciss, as CD-ROMs are */
1716 /* not supported, and we don't have any bus/target/lun */
1717 /* which we present to the kernel. */
1719 case CDROM_SEND_PACKET:
1720 case CDROMCLOSETRAY:
1722 case SCSI_IOCTL_GET_IDLUN:
1723 case SCSI_IOCTL_GET_BUS_NUMBER:
1729 static void cciss_check_queues(ctlr_info_t *h)
1731 int start_queue = h->next_to_run;
1734 /* check to see if we have maxed out the number of commands that can
1735 * be placed on the queue. If so then exit. We do this check here
1736 * in case the interrupt we serviced was from an ioctl and did not
1737 * free any new commands.
1739 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1742 /* We have room on the queue for more commands. Now we need to queue
1743 * them up. We will also keep track of the next queue to run so
1744 * that every queue gets a chance to be started first.
1746 for (i = 0; i < h->highest_lun + 1; i++) {
1747 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1748 /* make sure the disk has been added and the drive is real
1749 * because this can be called from the middle of init_one.
1751 if (!h->drv[curr_queue])
1753 if (!(h->drv[curr_queue]->queue) ||
1754 !(h->drv[curr_queue]->heads))
1756 blk_start_queue(h->gendisk[curr_queue]->queue);
1758 /* check to see if we have maxed out the number of commands
1759 * that can be placed on the queue.
1761 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1762 if (curr_queue == start_queue) {
1764 (start_queue + 1) % (h->highest_lun + 1);
1767 h->next_to_run = curr_queue;
1774 static void cciss_softirq_done(struct request *rq)
1776 CommandList_struct *cmd = rq->completion_data;
1777 ctlr_info_t *h = hba[cmd->ctlr];
1778 SGDescriptor_struct *curr_sg = cmd->SG;
1780 unsigned long flags;
1784 if (cmd->Request.Type.Direction == XFER_READ)
1785 ddir = PCI_DMA_FROMDEVICE;
1787 ddir = PCI_DMA_TODEVICE;
1789 /* command did not need to be retried */
1790 /* unmap the DMA mapping for all the scatter gather elements */
1791 for (i = 0; i < cmd->Header.SGList; i++) {
1792 if (curr_sg[sg_index].Ext == CCISS_SG_CHAIN) {
1793 cciss_unmap_sg_chain_block(h, cmd);
1794 /* Point to the next block */
1795 curr_sg = h->cmd_sg_list[cmd->cmdindex];
1798 temp64.val32.lower = curr_sg[sg_index].Addr.lower;
1799 temp64.val32.upper = curr_sg[sg_index].Addr.upper;
1800 pci_unmap_page(h->pdev, temp64.val, curr_sg[sg_index].Len,
1806 printk("Done with %p\n", rq);
1807 #endif /* CCISS_DEBUG */
1809 /* set the residual count for pc requests */
1810 if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
1811 rq->resid_len = cmd->err_info->ResidualCnt;
1813 blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1815 spin_lock_irqsave(&h->lock, flags);
1816 cmd_free(h, cmd, 1);
1817 cciss_check_queues(h);
1818 spin_unlock_irqrestore(&h->lock, flags);
1821 static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1822 unsigned char scsi3addr[], uint32_t log_unit)
1824 memcpy(scsi3addr, h->drv[log_unit]->LunID,
1825 sizeof(h->drv[log_unit]->LunID));
1828 /* This function gets the SCSI vendor, model, and revision of a logical drive
1829 * via the inquiry page 0. Model, vendor, and rev are set to empty strings if
1830 * they cannot be read.
1832 static void cciss_get_device_descr(int ctlr, int logvol,
1833 char *vendor, char *model, char *rev)
1836 InquiryData_struct *inq_buf;
1837 unsigned char scsi3addr[8];
1843 inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1847 log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1848 rc = sendcmd_withirq(CISS_INQUIRY, ctlr, inq_buf, sizeof(*inq_buf), 0,
1849 scsi3addr, TYPE_CMD);
1851 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1852 vendor[VENDOR_LEN] = '\0';
1853 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1854 model[MODEL_LEN] = '\0';
1855 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1856 rev[REV_LEN] = '\0';
1863 /* This function gets the serial number of a logical drive via
1864 * inquiry page 0x83. Serial no. is 16 bytes. If the serial
1865 * number cannot be had, for whatever reason, 16 bytes of 0xff
1866 * are returned instead.
1868 static void cciss_get_serial_no(int ctlr, int logvol,
1869 unsigned char *serial_no, int buflen)
1871 #define PAGE_83_INQ_BYTES 64
1874 unsigned char scsi3addr[8];
1878 memset(serial_no, 0xff, buflen);
1879 buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1882 memset(serial_no, 0, buflen);
1883 log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1884 rc = sendcmd_withirq(CISS_INQUIRY, ctlr, buf,
1885 PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1887 memcpy(serial_no, &buf[8], buflen);
1893 * cciss_add_disk sets up the block device queue for a logical drive
1895 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1898 disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1900 goto init_queue_failure;
1901 sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1902 disk->major = h->major;
1903 disk->first_minor = drv_index << NWD_SHIFT;
1904 disk->fops = &cciss_fops;
1905 if (cciss_create_ld_sysfs_entry(h, drv_index))
1907 disk->private_data = h->drv[drv_index];
1908 disk->driverfs_dev = &h->drv[drv_index]->dev;
1910 /* Set up queue information */
1911 blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1913 /* This is a hardware imposed limit. */
1914 blk_queue_max_segments(disk->queue, h->maxsgentries);
1916 blk_queue_max_hw_sectors(disk->queue, h->cciss_max_sectors);
1918 blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1920 disk->queue->queuedata = h;
1922 blk_queue_logical_block_size(disk->queue,
1923 h->drv[drv_index]->block_size);
1925 /* Make sure all queue data is written out before */
1926 /* setting h->drv[drv_index]->queue, as setting this */
1927 /* allows the interrupt handler to start the queue */
1929 h->drv[drv_index]->queue = disk->queue;
1934 blk_cleanup_queue(disk->queue);
1940 /* This function will check the usage_count of the drive to be updated/added.
1941 * If the usage_count is zero and it is a heretofore unknown drive, or,
1942 * the drive's capacity, geometry, or serial number has changed,
1943 * then the drive information will be updated and the disk will be
1944 * re-registered with the kernel. If these conditions don't hold,
1945 * then it will be left alone for the next reboot. The exception to this
1946 * is disk 0 which will always be left registered with the kernel since it
1947 * is also the controller node. Any changes to disk 0 will show up on
1950 static void cciss_update_drive_info(int ctlr, int drv_index, int first_time,
1953 ctlr_info_t *h = hba[ctlr];
1954 struct gendisk *disk;
1955 InquiryData_struct *inq_buff = NULL;
1956 unsigned int block_size;
1957 sector_t total_size;
1958 unsigned long flags = 0;
1960 drive_info_struct *drvinfo;
1962 /* Get information about the disk and modify the driver structure */
1963 inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1964 drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
1965 if (inq_buff == NULL || drvinfo == NULL)
1968 /* testing to see if 16-byte CDBs are already being used */
1969 if (h->cciss_read == CCISS_READ_16) {
1970 cciss_read_capacity_16(h->ctlr, drv_index,
1971 &total_size, &block_size);
1974 cciss_read_capacity(ctlr, drv_index, &total_size, &block_size);
1975 /* if read_capacity returns all F's this volume is >2TB */
1976 /* in size so we switch to 16-byte CDB's for all */
1977 /* read/write ops */
1978 if (total_size == 0xFFFFFFFFULL) {
1979 cciss_read_capacity_16(ctlr, drv_index,
1980 &total_size, &block_size);
1981 h->cciss_read = CCISS_READ_16;
1982 h->cciss_write = CCISS_WRITE_16;
1984 h->cciss_read = CCISS_READ_10;
1985 h->cciss_write = CCISS_WRITE_10;
1989 cciss_geometry_inquiry(ctlr, drv_index, total_size, block_size,
1991 drvinfo->block_size = block_size;
1992 drvinfo->nr_blocks = total_size + 1;
1994 cciss_get_device_descr(ctlr, drv_index, drvinfo->vendor,
1995 drvinfo->model, drvinfo->rev);
1996 cciss_get_serial_no(ctlr, drv_index, drvinfo->serial_no,
1997 sizeof(drvinfo->serial_no));
1998 /* Save the lunid in case we deregister the disk, below. */
1999 memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
2000 sizeof(drvinfo->LunID));
2002 /* Is it the same disk we already know, and nothing's changed? */
2003 if (h->drv[drv_index]->raid_level != -1 &&
2004 ((memcmp(drvinfo->serial_no,
2005 h->drv[drv_index]->serial_no, 16) == 0) &&
2006 drvinfo->block_size == h->drv[drv_index]->block_size &&
2007 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
2008 drvinfo->heads == h->drv[drv_index]->heads &&
2009 drvinfo->sectors == h->drv[drv_index]->sectors &&
2010 drvinfo->cylinders == h->drv[drv_index]->cylinders))
2011 /* The disk is unchanged, nothing to update */
2014 /* If we get here it's not the same disk, or something's changed,
2015 * so we need to * deregister it, and re-register it, if it's not
2017 * If the disk already exists then deregister it before proceeding
2018 * (unless it's the first disk (for the controller node).
2020 if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
2021 printk(KERN_WARNING "disk %d has changed.\n", drv_index);
2022 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2023 h->drv[drv_index]->busy_configuring = 1;
2024 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2026 /* deregister_disk sets h->drv[drv_index]->queue = NULL
2027 * which keeps the interrupt handler from starting
2030 ret = deregister_disk(h, drv_index, 0, via_ioctl);
2033 /* If the disk is in use return */
2037 /* Save the new information from cciss_geometry_inquiry
2038 * and serial number inquiry. If the disk was deregistered
2039 * above, then h->drv[drv_index] will be NULL.
2041 if (h->drv[drv_index] == NULL) {
2042 drvinfo->device_initialized = 0;
2043 h->drv[drv_index] = drvinfo;
2044 drvinfo = NULL; /* so it won't be freed below. */
2046 /* special case for cxd0 */
2047 h->drv[drv_index]->block_size = drvinfo->block_size;
2048 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
2049 h->drv[drv_index]->heads = drvinfo->heads;
2050 h->drv[drv_index]->sectors = drvinfo->sectors;
2051 h->drv[drv_index]->cylinders = drvinfo->cylinders;
2052 h->drv[drv_index]->raid_level = drvinfo->raid_level;
2053 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
2054 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
2056 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
2057 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
2061 disk = h->gendisk[drv_index];
2062 set_capacity(disk, h->drv[drv_index]->nr_blocks);
2064 /* If it's not disk 0 (drv_index != 0)
2065 * or if it was disk 0, but there was previously
2066 * no actual corresponding configured logical drive
2067 * (raid_leve == -1) then we want to update the
2068 * logical drive's information.
2070 if (drv_index || first_time) {
2071 if (cciss_add_disk(h, disk, drv_index) != 0) {
2072 cciss_free_gendisk(h, drv_index);
2073 cciss_free_drive_info(h, drv_index);
2074 printk(KERN_WARNING "cciss:%d could not update "
2075 "disk %d\n", h->ctlr, drv_index);
2085 printk(KERN_ERR "cciss: out of memory\n");
2089 /* This function will find the first index of the controllers drive array
2090 * that has a null drv pointer and allocate the drive info struct and
2091 * will return that index This is where new drives will be added.
2092 * If the index to be returned is greater than the highest_lun index for
2093 * the controller then highest_lun is set * to this new index.
2094 * If there are no available indexes or if tha allocation fails, then -1
2095 * is returned. * "controller_node" is used to know if this is a real
2096 * logical drive, or just the controller node, which determines if this
2097 * counts towards highest_lun.
2099 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
2102 drive_info_struct *drv;
2104 /* Search for an empty slot for our drive info */
2105 for (i = 0; i < CISS_MAX_LUN; i++) {
2107 /* if not cxd0 case, and it's occupied, skip it. */
2108 if (h->drv[i] && i != 0)
2111 * If it's cxd0 case, and drv is alloc'ed already, and a
2112 * disk is configured there, skip it.
2114 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2118 * We've found an empty slot. Update highest_lun
2119 * provided this isn't just the fake cxd0 controller node.
2121 if (i > h->highest_lun && !controller_node)
2124 /* If adding a real disk at cxd0, and it's already alloc'ed */
2125 if (i == 0 && h->drv[i] != NULL)
2129 * Found an empty slot, not already alloc'ed. Allocate it.
2130 * Mark it with raid_level == -1, so we know it's new later on.
2132 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2135 drv->raid_level = -1; /* so we know it's new */
2142 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2144 kfree(h->drv[drv_index]);
2145 h->drv[drv_index] = NULL;
2148 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2150 put_disk(h->gendisk[drv_index]);
2151 h->gendisk[drv_index] = NULL;
2154 /* cciss_add_gendisk finds a free hba[]->drv structure
2155 * and allocates a gendisk if needed, and sets the lunid
2156 * in the drvinfo structure. It returns the index into
2157 * the ->drv[] array, or -1 if none are free.
2158 * is_controller_node indicates whether highest_lun should
2159 * count this disk, or if it's only being added to provide
2160 * a means to talk to the controller in case no logical
2161 * drives have yet been configured.
2163 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2164 int controller_node)
2168 drv_index = cciss_alloc_drive_info(h, controller_node);
2169 if (drv_index == -1)
2172 /*Check if the gendisk needs to be allocated */
2173 if (!h->gendisk[drv_index]) {
2174 h->gendisk[drv_index] =
2175 alloc_disk(1 << NWD_SHIFT);
2176 if (!h->gendisk[drv_index]) {
2177 printk(KERN_ERR "cciss%d: could not "
2178 "allocate a new disk %d\n",
2179 h->ctlr, drv_index);
2180 goto err_free_drive_info;
2183 memcpy(h->drv[drv_index]->LunID, lunid,
2184 sizeof(h->drv[drv_index]->LunID));
2185 if (cciss_create_ld_sysfs_entry(h, drv_index))
2187 /* Don't need to mark this busy because nobody */
2188 /* else knows about this disk yet to contend */
2189 /* for access to it. */
2190 h->drv[drv_index]->busy_configuring = 0;
2195 cciss_free_gendisk(h, drv_index);
2196 err_free_drive_info:
2197 cciss_free_drive_info(h, drv_index);
2201 /* This is for the special case of a controller which
2202 * has no logical drives. In this case, we still need
2203 * to register a disk so the controller can be accessed
2204 * by the Array Config Utility.
2206 static void cciss_add_controller_node(ctlr_info_t *h)
2208 struct gendisk *disk;
2211 if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2214 drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2215 if (drv_index == -1)
2217 h->drv[drv_index]->block_size = 512;
2218 h->drv[drv_index]->nr_blocks = 0;
2219 h->drv[drv_index]->heads = 0;
2220 h->drv[drv_index]->sectors = 0;
2221 h->drv[drv_index]->cylinders = 0;
2222 h->drv[drv_index]->raid_level = -1;
2223 memset(h->drv[drv_index]->serial_no, 0, 16);
2224 disk = h->gendisk[drv_index];
2225 if (cciss_add_disk(h, disk, drv_index) == 0)
2227 cciss_free_gendisk(h, drv_index);
2228 cciss_free_drive_info(h, drv_index);
2230 printk(KERN_WARNING "cciss%d: could not "
2231 "add disk 0.\n", h->ctlr);
2235 /* This function will add and remove logical drives from the Logical
2236 * drive array of the controller and maintain persistency of ordering
2237 * so that mount points are preserved until the next reboot. This allows
2238 * for the removal of logical drives in the middle of the drive array
2239 * without a re-ordering of those drives.
2241 * h = The controller to perform the operations on
2243 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2248 ReportLunData_struct *ld_buff = NULL;
2254 unsigned char lunid[8] = CTLR_LUNID;
2255 unsigned long flags;
2257 if (!capable(CAP_SYS_RAWIO))
2260 /* Set busy_configuring flag for this operation */
2261 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2262 if (h->busy_configuring) {
2263 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2266 h->busy_configuring = 1;
2267 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2269 ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2270 if (ld_buff == NULL)
2273 return_code = sendcmd_withirq(CISS_REPORT_LOG, ctlr, ld_buff,
2274 sizeof(ReportLunData_struct),
2275 0, CTLR_LUNID, TYPE_CMD);
2277 if (return_code == IO_OK)
2278 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2279 else { /* reading number of logical volumes failed */
2280 printk(KERN_WARNING "cciss: report logical volume"
2281 " command failed\n");
2286 num_luns = listlength / 8; /* 8 bytes per entry */
2287 if (num_luns > CISS_MAX_LUN) {
2288 num_luns = CISS_MAX_LUN;
2289 printk(KERN_WARNING "cciss: more luns configured"
2290 " on controller than can be handled by"
2295 cciss_add_controller_node(h);
2297 /* Compare controller drive array to driver's drive array
2298 * to see if any drives are missing on the controller due
2299 * to action of Array Config Utility (user deletes drive)
2300 * and deregister logical drives which have disappeared.
2302 for (i = 0; i <= h->highest_lun; i++) {
2306 /* skip holes in the array from already deleted drives */
2307 if (h->drv[i] == NULL)
2310 for (j = 0; j < num_luns; j++) {
2311 memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2312 if (memcmp(h->drv[i]->LunID, lunid,
2313 sizeof(lunid)) == 0) {
2319 /* Deregister it from the OS, it's gone. */
2320 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2321 h->drv[i]->busy_configuring = 1;
2322 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2323 return_code = deregister_disk(h, i, 1, via_ioctl);
2324 if (h->drv[i] != NULL)
2325 h->drv[i]->busy_configuring = 0;
2329 /* Compare controller drive array to driver's drive array.
2330 * Check for updates in the drive information and any new drives
2331 * on the controller due to ACU adding logical drives, or changing
2332 * a logical drive's size, etc. Reregister any new/changed drives
2334 for (i = 0; i < num_luns; i++) {
2339 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2340 /* Find if the LUN is already in the drive array
2341 * of the driver. If so then update its info
2342 * if not in use. If it does not exist then find
2343 * the first free index and add it.
2345 for (j = 0; j <= h->highest_lun; j++) {
2346 if (h->drv[j] != NULL &&
2347 memcmp(h->drv[j]->LunID, lunid,
2348 sizeof(h->drv[j]->LunID)) == 0) {
2355 /* check if the drive was found already in the array */
2357 drv_index = cciss_add_gendisk(h, lunid, 0);
2358 if (drv_index == -1)
2361 cciss_update_drive_info(ctlr, drv_index, first_time,
2367 h->busy_configuring = 0;
2368 /* We return -1 here to tell the ACU that we have registered/updated
2369 * all of the drives that we can and to keep it from calling us
2374 printk(KERN_ERR "cciss: out of memory\n");
2375 h->busy_configuring = 0;
2379 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2381 /* zero out the disk size info */
2382 drive_info->nr_blocks = 0;
2383 drive_info->block_size = 0;
2384 drive_info->heads = 0;
2385 drive_info->sectors = 0;
2386 drive_info->cylinders = 0;
2387 drive_info->raid_level = -1;
2388 memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2389 memset(drive_info->model, 0, sizeof(drive_info->model));
2390 memset(drive_info->rev, 0, sizeof(drive_info->rev));
2391 memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2393 * don't clear the LUNID though, we need to remember which
2398 /* This function will deregister the disk and it's queue from the
2399 * kernel. It must be called with the controller lock held and the
2400 * drv structures busy_configuring flag set. It's parameters are:
2402 * disk = This is the disk to be deregistered
2403 * drv = This is the drive_info_struct associated with the disk to be
2404 * deregistered. It contains information about the disk used
2406 * clear_all = This flag determines whether or not the disk information
2407 * is going to be completely cleared out and the highest_lun
2408 * reset. Sometimes we want to clear out information about
2409 * the disk in preparation for re-adding it. In this case
2410 * the highest_lun should be left unchanged and the LunID
2411 * should not be cleared.
2413 * This indicates whether we've reached this path via ioctl.
2414 * This affects the maximum usage count allowed for c0d0 to be messed with.
2415 * If this path is reached via ioctl(), then the max_usage_count will
2416 * be 1, as the process calling ioctl() has got to have the device open.
2417 * If we get here via sysfs, then the max usage count will be zero.
2419 static int deregister_disk(ctlr_info_t *h, int drv_index,
2420 int clear_all, int via_ioctl)
2423 struct gendisk *disk;
2424 drive_info_struct *drv;
2425 int recalculate_highest_lun;
2427 if (!capable(CAP_SYS_RAWIO))
2430 drv = h->drv[drv_index];
2431 disk = h->gendisk[drv_index];
2433 /* make sure logical volume is NOT is use */
2434 if (clear_all || (h->gendisk[0] == disk)) {
2435 if (drv->usage_count > via_ioctl)
2437 } else if (drv->usage_count > 0)
2440 recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2442 /* invalidate the devices and deregister the disk. If it is disk
2443 * zero do not deregister it but just zero out it's values. This
2444 * allows us to delete disk zero but keep the controller registered.
2446 if (h->gendisk[0] != disk) {
2447 struct request_queue *q = disk->queue;
2448 if (disk->flags & GENHD_FL_UP) {
2449 cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2453 blk_cleanup_queue(q);
2454 /* If clear_all is set then we are deleting the logical
2455 * drive, not just refreshing its info. For drives
2456 * other than disk 0 we will call put_disk. We do not
2457 * do this for disk 0 as we need it to be able to
2458 * configure the controller.
2461 /* This isn't pretty, but we need to find the
2462 * disk in our array and NULL our the pointer.
2463 * This is so that we will call alloc_disk if
2464 * this index is used again later.
2466 for (i=0; i < CISS_MAX_LUN; i++){
2467 if (h->gendisk[i] == disk) {
2468 h->gendisk[i] = NULL;
2475 set_capacity(disk, 0);
2476 cciss_clear_drive_info(drv);
2481 /* if it was the last disk, find the new hightest lun */
2482 if (clear_all && recalculate_highest_lun) {
2483 int newhighest = -1;
2484 for (i = 0; i <= h->highest_lun; i++) {
2485 /* if the disk has size > 0, it is available */
2486 if (h->drv[i] && h->drv[i]->heads)
2489 h->highest_lun = newhighest;
2494 static int fill_cmd(CommandList_struct *c, __u8 cmd, int ctlr, void *buff,
2495 size_t size, __u8 page_code, unsigned char *scsi3addr,
2498 ctlr_info_t *h = hba[ctlr];
2499 u64bit buff_dma_handle;
2502 c->cmd_type = CMD_IOCTL_PEND;
2503 c->Header.ReplyQueue = 0;
2505 c->Header.SGList = 1;
2506 c->Header.SGTotal = 1;
2508 c->Header.SGList = 0;
2509 c->Header.SGTotal = 0;
2511 c->Header.Tag.lower = c->busaddr;
2512 memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2514 c->Request.Type.Type = cmd_type;
2515 if (cmd_type == TYPE_CMD) {
2518 /* are we trying to read a vital product page */
2519 if (page_code != 0) {
2520 c->Request.CDB[1] = 0x01;
2521 c->Request.CDB[2] = page_code;
2523 c->Request.CDBLen = 6;
2524 c->Request.Type.Attribute = ATTR_SIMPLE;
2525 c->Request.Type.Direction = XFER_READ;
2526 c->Request.Timeout = 0;
2527 c->Request.CDB[0] = CISS_INQUIRY;
2528 c->Request.CDB[4] = size & 0xFF;
2530 case CISS_REPORT_LOG:
2531 case CISS_REPORT_PHYS:
2532 /* Talking to controller so It's a physical command
2533 mode = 00 target = 0. Nothing to write.
2535 c->Request.CDBLen = 12;
2536 c->Request.Type.Attribute = ATTR_SIMPLE;
2537 c->Request.Type.Direction = XFER_READ;
2538 c->Request.Timeout = 0;
2539 c->Request.CDB[0] = cmd;
2540 c->Request.CDB[6] = (size >> 24) & 0xFF; /* MSB */
2541 c->Request.CDB[7] = (size >> 16) & 0xFF;
2542 c->Request.CDB[8] = (size >> 8) & 0xFF;
2543 c->Request.CDB[9] = size & 0xFF;
2546 case CCISS_READ_CAPACITY:
2547 c->Request.CDBLen = 10;
2548 c->Request.Type.Attribute = ATTR_SIMPLE;
2549 c->Request.Type.Direction = XFER_READ;
2550 c->Request.Timeout = 0;
2551 c->Request.CDB[0] = cmd;
2553 case CCISS_READ_CAPACITY_16:
2554 c->Request.CDBLen = 16;
2555 c->Request.Type.Attribute = ATTR_SIMPLE;
2556 c->Request.Type.Direction = XFER_READ;
2557 c->Request.Timeout = 0;
2558 c->Request.CDB[0] = cmd;
2559 c->Request.CDB[1] = 0x10;
2560 c->Request.CDB[10] = (size >> 24) & 0xFF;
2561 c->Request.CDB[11] = (size >> 16) & 0xFF;
2562 c->Request.CDB[12] = (size >> 8) & 0xFF;
2563 c->Request.CDB[13] = size & 0xFF;
2564 c->Request.Timeout = 0;
2565 c->Request.CDB[0] = cmd;
2567 case CCISS_CACHE_FLUSH:
2568 c->Request.CDBLen = 12;
2569 c->Request.Type.Attribute = ATTR_SIMPLE;
2570 c->Request.Type.Direction = XFER_WRITE;
2571 c->Request.Timeout = 0;
2572 c->Request.CDB[0] = BMIC_WRITE;
2573 c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2575 case TEST_UNIT_READY:
2576 c->Request.CDBLen = 6;
2577 c->Request.Type.Attribute = ATTR_SIMPLE;
2578 c->Request.Type.Direction = XFER_NONE;
2579 c->Request.Timeout = 0;
2583 "cciss%d: Unknown Command 0x%c\n", ctlr, cmd);
2586 } else if (cmd_type == TYPE_MSG) {
2588 case 0: /* ABORT message */
2589 c->Request.CDBLen = 12;
2590 c->Request.Type.Attribute = ATTR_SIMPLE;
2591 c->Request.Type.Direction = XFER_WRITE;
2592 c->Request.Timeout = 0;
2593 c->Request.CDB[0] = cmd; /* abort */
2594 c->Request.CDB[1] = 0; /* abort a command */
2595 /* buff contains the tag of the command to abort */
2596 memcpy(&c->Request.CDB[4], buff, 8);
2598 case 1: /* RESET message */
2599 c->Request.CDBLen = 16;
2600 c->Request.Type.Attribute = ATTR_SIMPLE;
2601 c->Request.Type.Direction = XFER_NONE;
2602 c->Request.Timeout = 0;
2603 memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2604 c->Request.CDB[0] = cmd; /* reset */
2605 c->Request.CDB[1] = 0x03; /* reset a target */
2607 case 3: /* No-Op message */
2608 c->Request.CDBLen = 1;
2609 c->Request.Type.Attribute = ATTR_SIMPLE;
2610 c->Request.Type.Direction = XFER_WRITE;
2611 c->Request.Timeout = 0;
2612 c->Request.CDB[0] = cmd;
2616 "cciss%d: unknown message type %d\n", ctlr, cmd);
2621 "cciss%d: unknown command type %d\n", ctlr, cmd_type);
2624 /* Fill in the scatter gather information */
2626 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2628 PCI_DMA_BIDIRECTIONAL);
2629 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2630 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2631 c->SG[0].Len = size;
2632 c->SG[0].Ext = 0; /* we are not chaining */
2637 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2639 switch (c->err_info->ScsiStatus) {
2642 case SAM_STAT_CHECK_CONDITION:
2643 switch (0xf & c->err_info->SenseInfo[2]) {
2644 case 0: return IO_OK; /* no sense */
2645 case 1: return IO_OK; /* recovered error */
2647 if (check_for_unit_attention(h, c))
2648 return IO_NEEDS_RETRY;
2649 printk(KERN_WARNING "cciss%d: cmd 0x%02x "
2650 "check condition, sense key = 0x%02x\n",
2651 h->ctlr, c->Request.CDB[0],
2652 c->err_info->SenseInfo[2]);
2656 printk(KERN_WARNING "cciss%d: cmd 0x%02x"
2657 "scsi status = 0x%02x\n", h->ctlr,
2658 c->Request.CDB[0], c->err_info->ScsiStatus);
2664 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2666 int return_status = IO_OK;
2668 if (c->err_info->CommandStatus == CMD_SUCCESS)
2671 switch (c->err_info->CommandStatus) {
2672 case CMD_TARGET_STATUS:
2673 return_status = check_target_status(h, c);
2675 case CMD_DATA_UNDERRUN:
2676 case CMD_DATA_OVERRUN:
2677 /* expected for inquiry and report lun commands */
2680 printk(KERN_WARNING "cciss: cmd 0x%02x is "
2681 "reported invalid\n", c->Request.CDB[0]);
2682 return_status = IO_ERROR;
2684 case CMD_PROTOCOL_ERR:
2685 printk(KERN_WARNING "cciss: cmd 0x%02x has "
2686 "protocol error \n", c->Request.CDB[0]);
2687 return_status = IO_ERROR;
2689 case CMD_HARDWARE_ERR:
2690 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2691 " hardware error\n", c->Request.CDB[0]);
2692 return_status = IO_ERROR;
2694 case CMD_CONNECTION_LOST:
2695 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2696 "connection lost\n", c->Request.CDB[0]);
2697 return_status = IO_ERROR;
2700 printk(KERN_WARNING "cciss: cmd 0x%02x was "
2701 "aborted\n", c->Request.CDB[0]);
2702 return_status = IO_ERROR;
2704 case CMD_ABORT_FAILED:
2705 printk(KERN_WARNING "cciss: cmd 0x%02x reports "
2706 "abort failed\n", c->Request.CDB[0]);
2707 return_status = IO_ERROR;
2709 case CMD_UNSOLICITED_ABORT:
2711 "cciss%d: unsolicited abort 0x%02x\n", h->ctlr,
2713 return_status = IO_NEEDS_RETRY;
2716 printk(KERN_WARNING "cciss: cmd 0x%02x returned "
2717 "unknown status %x\n", c->Request.CDB[0],
2718 c->err_info->CommandStatus);
2719 return_status = IO_ERROR;
2721 return return_status;
2724 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2727 DECLARE_COMPLETION_ONSTACK(wait);
2728 u64bit buff_dma_handle;
2729 int return_status = IO_OK;
2733 enqueue_cmd_and_start_io(h, c);
2735 wait_for_completion(&wait);
2737 if (c->err_info->CommandStatus == 0 || !attempt_retry)
2740 return_status = process_sendcmd_error(h, c);
2742 if (return_status == IO_NEEDS_RETRY &&
2743 c->retry_count < MAX_CMD_RETRIES) {
2744 printk(KERN_WARNING "cciss%d: retrying 0x%02x\n", h->ctlr,
2747 /* erase the old error information */
2748 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2749 return_status = IO_OK;
2750 INIT_COMPLETION(wait);
2755 /* unlock the buffers from DMA */
2756 buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2757 buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2758 pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2759 c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2760 return return_status;
2763 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
2764 __u8 page_code, unsigned char scsi3addr[],
2767 ctlr_info_t *h = hba[ctlr];
2768 CommandList_struct *c;
2771 c = cmd_alloc(h, 0);
2774 return_status = fill_cmd(c, cmd, ctlr, buff, size, page_code,
2775 scsi3addr, cmd_type);
2776 if (return_status == IO_OK)
2777 return_status = sendcmd_withirq_core(h, c, 1);
2780 return return_status;
2783 static void cciss_geometry_inquiry(int ctlr, int logvol,
2784 sector_t total_size,
2785 unsigned int block_size,
2786 InquiryData_struct *inq_buff,
2787 drive_info_struct *drv)
2791 unsigned char scsi3addr[8];
2793 memset(inq_buff, 0, sizeof(InquiryData_struct));
2794 log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2795 return_code = sendcmd_withirq(CISS_INQUIRY, ctlr, inq_buff,
2796 sizeof(*inq_buff), 0xC1, scsi3addr, TYPE_CMD);
2797 if (return_code == IO_OK) {
2798 if (inq_buff->data_byte[8] == 0xFF) {
2800 "cciss: reading geometry failed, volume "
2801 "does not support reading geometry\n");
2803 drv->sectors = 32; /* Sectors per track */
2804 drv->cylinders = total_size + 1;
2805 drv->raid_level = RAID_UNKNOWN;
2807 drv->heads = inq_buff->data_byte[6];
2808 drv->sectors = inq_buff->data_byte[7];
2809 drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2810 drv->cylinders += inq_buff->data_byte[5];
2811 drv->raid_level = inq_buff->data_byte[8];
2813 drv->block_size = block_size;
2814 drv->nr_blocks = total_size + 1;
2815 t = drv->heads * drv->sectors;
2817 sector_t real_size = total_size + 1;
2818 unsigned long rem = sector_div(real_size, t);
2821 drv->cylinders = real_size;
2823 } else { /* Get geometry failed */
2824 printk(KERN_WARNING "cciss: reading geometry failed\n");
2829 cciss_read_capacity(int ctlr, int logvol, sector_t *total_size,
2830 unsigned int *block_size)
2832 ReadCapdata_struct *buf;
2834 unsigned char scsi3addr[8];
2836 buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2838 printk(KERN_WARNING "cciss: out of memory\n");
2842 log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2843 return_code = sendcmd_withirq(CCISS_READ_CAPACITY, ctlr, buf,
2844 sizeof(ReadCapdata_struct), 0, scsi3addr, TYPE_CMD);
2845 if (return_code == IO_OK) {
2846 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2847 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2848 } else { /* read capacity command failed */
2849 printk(KERN_WARNING "cciss: read capacity failed\n");
2851 *block_size = BLOCK_SIZE;
2856 static void cciss_read_capacity_16(int ctlr, int logvol,
2857 sector_t *total_size, unsigned int *block_size)
2859 ReadCapdata_struct_16 *buf;
2861 unsigned char scsi3addr[8];
2863 buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2865 printk(KERN_WARNING "cciss: out of memory\n");
2869 log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2870 return_code = sendcmd_withirq(CCISS_READ_CAPACITY_16,
2871 ctlr, buf, sizeof(ReadCapdata_struct_16),
2872 0, scsi3addr, TYPE_CMD);
2873 if (return_code == IO_OK) {
2874 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2875 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2876 } else { /* read capacity command failed */
2877 printk(KERN_WARNING "cciss: read capacity failed\n");
2879 *block_size = BLOCK_SIZE;
2881 printk(KERN_INFO " blocks= %llu block_size= %d\n",
2882 (unsigned long long)*total_size+1, *block_size);
2886 static int cciss_revalidate(struct gendisk *disk)
2888 ctlr_info_t *h = get_host(disk);
2889 drive_info_struct *drv = get_drv(disk);
2892 unsigned int block_size;
2893 sector_t total_size;
2894 InquiryData_struct *inq_buff = NULL;
2896 for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
2897 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2898 sizeof(drv->LunID)) == 0) {
2907 inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2908 if (inq_buff == NULL) {
2909 printk(KERN_WARNING "cciss: out of memory\n");
2912 if (h->cciss_read == CCISS_READ_10) {
2913 cciss_read_capacity(h->ctlr, logvol,
2914 &total_size, &block_size);
2916 cciss_read_capacity_16(h->ctlr, logvol,
2917 &total_size, &block_size);
2919 cciss_geometry_inquiry(h->ctlr, logvol, total_size, block_size,
2922 blk_queue_logical_block_size(drv->queue, drv->block_size);
2923 set_capacity(disk, drv->nr_blocks);
2930 * Map (physical) PCI mem into (virtual) kernel space
2932 static void __iomem *remap_pci_mem(ulong base, ulong size)
2934 ulong page_base = ((ulong) base) & PAGE_MASK;
2935 ulong page_offs = ((ulong) base) - page_base;
2936 void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2938 return page_remapped ? (page_remapped + page_offs) : NULL;
2942 * Takes jobs of the Q and sends them to the hardware, then puts it on
2943 * the Q to wait for completion.
2945 static void start_io(ctlr_info_t *h)
2947 CommandList_struct *c;
2949 while (!hlist_empty(&h->reqQ)) {
2950 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
2951 /* can't do anything if fifo is full */
2952 if ((h->access.fifo_full(h))) {
2953 printk(KERN_WARNING "cciss: fifo full\n");
2957 /* Get the first entry from the Request Q */
2961 /* Tell the controller execute command */
2962 h->access.submit_command(h, c);
2964 /* Put job onto the completed Q */
2969 /* Assumes that CCISS_LOCK(h->ctlr) is held. */
2970 /* Zeros out the error record and then resends the command back */
2971 /* to the controller */
2972 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2974 /* erase the old error information */
2975 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2977 /* add it to software queue and then send it to the controller */
2980 if (h->Qdepth > h->maxQsinceinit)
2981 h->maxQsinceinit = h->Qdepth;
2986 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
2987 unsigned int msg_byte, unsigned int host_byte,
2988 unsigned int driver_byte)
2990 /* inverse of macros in scsi.h */
2991 return (scsi_status_byte & 0xff) |
2992 ((msg_byte & 0xff) << 8) |
2993 ((host_byte & 0xff) << 16) |
2994 ((driver_byte & 0xff) << 24);
2997 static inline int evaluate_target_status(ctlr_info_t *h,
2998 CommandList_struct *cmd, int *retry_cmd)
3000 unsigned char sense_key;
3001 unsigned char status_byte, msg_byte, host_byte, driver_byte;
3005 /* If we get in here, it means we got "target status", that is, scsi status */
3006 status_byte = cmd->err_info->ScsiStatus;
3007 driver_byte = DRIVER_OK;
3008 msg_byte = cmd->err_info->CommandStatus; /* correct? seems too device specific */
3010 if (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC)
3011 host_byte = DID_PASSTHROUGH;
3015 error_value = make_status_bytes(status_byte, msg_byte,
3016 host_byte, driver_byte);
3018 if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
3019 if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC)
3020 printk(KERN_WARNING "cciss: cmd %p "
3021 "has SCSI Status 0x%x\n",
3022 cmd, cmd->err_info->ScsiStatus);
3026 /* check the sense key */
3027 sense_key = 0xf & cmd->err_info->SenseInfo[2];
3028 /* no status or recovered error */
3029 if (((sense_key == 0x0) || (sense_key == 0x1)) &&
3030 (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC))
3033 if (check_for_unit_attention(h, cmd)) {
3034 *retry_cmd = !(cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC);
3038 /* Not SG_IO or similar? */
3039 if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC) {
3040 if (error_value != 0)
3041 printk(KERN_WARNING "cciss: cmd %p has CHECK CONDITION"
3042 " sense key = 0x%x\n", cmd, sense_key);
3046 /* SG_IO or similar, copy sense data back */
3047 if (cmd->rq->sense) {
3048 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
3049 cmd->rq->sense_len = cmd->err_info->SenseLen;
3050 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
3051 cmd->rq->sense_len);
3053 cmd->rq->sense_len = 0;
3058 /* checks the status of the job and calls complete buffers to mark all
3059 * buffers for the completed job. Note that this function does not need
3060 * to hold the hba/queue lock.
3062 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
3066 struct request *rq = cmd->rq;
3071 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
3073 if (cmd->err_info->CommandStatus == 0) /* no error has occurred */
3074 goto after_error_processing;
3076 switch (cmd->err_info->CommandStatus) {
3077 case CMD_TARGET_STATUS:
3078 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
3080 case CMD_DATA_UNDERRUN:
3081 if (cmd->rq->cmd_type == REQ_TYPE_FS) {
3082 printk(KERN_WARNING "cciss: cmd %p has"
3083 " completed with data underrun "
3085 cmd->rq->resid_len = cmd->err_info->ResidualCnt;
3088 case CMD_DATA_OVERRUN:
3089 if (cmd->rq->cmd_type == REQ_TYPE_FS)
3090 printk(KERN_WARNING "cciss: cmd %p has"
3091 " completed with data overrun "
3095 printk(KERN_WARNING "cciss: cmd %p is "
3096 "reported invalid\n", cmd);
3097 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3098 cmd->err_info->CommandStatus, DRIVER_OK,
3099 (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3100 DID_PASSTHROUGH : DID_ERROR);
3102 case CMD_PROTOCOL_ERR:
3103 printk(KERN_WARNING "cciss: cmd %p has "
3104 "protocol error \n", cmd);
3105 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3106 cmd->err_info->CommandStatus, DRIVER_OK,
3107 (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3108 DID_PASSTHROUGH : DID_ERROR);
3110 case CMD_HARDWARE_ERR:
3111 printk(KERN_WARNING "cciss: cmd %p had "
3112 " hardware error\n", cmd);
3113 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3114 cmd->err_info->CommandStatus, DRIVER_OK,
3115 (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3116 DID_PASSTHROUGH : DID_ERROR);
3118 case CMD_CONNECTION_LOST:
3119 printk(KERN_WARNING "cciss: cmd %p had "
3120 "connection lost\n", cmd);
3121 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3122 cmd->err_info->CommandStatus, DRIVER_OK,
3123 (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3124 DID_PASSTHROUGH : DID_ERROR);
3127 printk(KERN_WARNING "cciss: cmd %p was "
3129 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3130 cmd->err_info->CommandStatus, DRIVER_OK,
3131 (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3132 DID_PASSTHROUGH : DID_ABORT);
3134 case CMD_ABORT_FAILED:
3135 printk(KERN_WARNING "cciss: cmd %p reports "
3136 "abort failed\n", cmd);
3137 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3138 cmd->err_info->CommandStatus, DRIVER_OK,
3139 (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3140 DID_PASSTHROUGH : DID_ERROR);
3142 case CMD_UNSOLICITED_ABORT:
3143 printk(KERN_WARNING "cciss%d: unsolicited "
3144 "abort %p\n", h->ctlr, cmd);
3145 if (cmd->retry_count < MAX_CMD_RETRIES) {
3148 "cciss%d: retrying %p\n", h->ctlr, cmd);
3152 "cciss%d: %p retried too "
3153 "many times\n", h->ctlr, cmd);
3154 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3155 cmd->err_info->CommandStatus, DRIVER_OK,
3156 (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3157 DID_PASSTHROUGH : DID_ABORT);
3160 printk(KERN_WARNING "cciss: cmd %p timedout\n", cmd);
3161 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3162 cmd->err_info->CommandStatus, DRIVER_OK,
3163 (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3164 DID_PASSTHROUGH : DID_ERROR);
3167 printk(KERN_WARNING "cciss: cmd %p returned "
3168 "unknown status %x\n", cmd,
3169 cmd->err_info->CommandStatus);
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);
3176 after_error_processing:
3178 /* We need to return this command */
3180 resend_cciss_cmd(h, cmd);
3183 cmd->rq->completion_data = cmd;
3184 blk_complete_request(cmd->rq);
3187 static inline u32 cciss_tag_contains_index(u32 tag)
3189 #define DIRECT_LOOKUP_BIT 0x10
3190 return tag & DIRECT_LOOKUP_BIT;
3193 static inline u32 cciss_tag_to_index(u32 tag)
3195 #define DIRECT_LOOKUP_SHIFT 5
3196 return tag >> DIRECT_LOOKUP_SHIFT;
3199 static inline u32 cciss_tag_discard_error_bits(u32 tag)
3201 #define CCISS_ERROR_BITS 0x03
3202 return tag & ~CCISS_ERROR_BITS;
3205 static inline void cciss_mark_tag_indexed(u32 *tag)
3207 *tag |= DIRECT_LOOKUP_BIT;
3210 static inline void cciss_set_tag_index(u32 *tag, u32 index)
3212 *tag |= (index << DIRECT_LOOKUP_SHIFT);
3216 * Get a request and submit it to the controller.
3218 static void do_cciss_request(struct request_queue *q)
3220 ctlr_info_t *h = q->queuedata;
3221 CommandList_struct *c;
3224 struct request *creq;
3226 struct scatterlist *tmp_sg;
3227 SGDescriptor_struct *curr_sg;
3228 drive_info_struct *drv;
3233 /* We call start_io here in case there is a command waiting on the
3234 * queue that has not been sent.
3236 if (blk_queue_plugged(q))
3240 creq = blk_peek_request(q);
3244 BUG_ON(creq->nr_phys_segments > h->maxsgentries);
3246 if ((c = cmd_alloc(h, 1)) == NULL)
3249 blk_start_request(creq);
3251 tmp_sg = h->scatter_list[c->cmdindex];
3252 spin_unlock_irq(q->queue_lock);
3254 c->cmd_type = CMD_RWREQ;
3257 /* fill in the request */
3258 drv = creq->rq_disk->private_data;
3259 c->Header.ReplyQueue = 0; /* unused in simple mode */
3260 /* got command from pool, so use the command block index instead */
3261 /* for direct lookups. */
3262 /* The first 2 bits are reserved for controller error reporting. */
3263 cciss_set_tag_index(&c->Header.Tag.lower, c->cmdindex);
3264 cciss_mark_tag_indexed(&c->Header.Tag.lower);
3265 memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3266 c->Request.CDBLen = 10; /* 12 byte commands not in FW yet; */
3267 c->Request.Type.Type = TYPE_CMD; /* It is a command. */
3268 c->Request.Type.Attribute = ATTR_SIMPLE;
3269 c->Request.Type.Direction =
3270 (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3271 c->Request.Timeout = 0; /* Don't time out */
3273 (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3274 start_blk = blk_rq_pos(creq);
3276 printk(KERN_DEBUG "ciss: sector =%d nr_sectors=%d\n",
3277 (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3278 #endif /* CCISS_DEBUG */
3280 sg_init_table(tmp_sg, h->maxsgentries);
3281 seg = blk_rq_map_sg(q, creq, tmp_sg);
3283 /* get the DMA records for the setup */
3284 if (c->Request.Type.Direction == XFER_READ)
3285 dir = PCI_DMA_FROMDEVICE;
3287 dir = PCI_DMA_TODEVICE;
3293 for (i = 0; i < seg; i++) {
3294 if (((sg_index+1) == (h->max_cmd_sgentries)) &&
3295 !chained && ((seg - i) > 1)) {
3296 /* Point to next chain block. */
3297 curr_sg = h->cmd_sg_list[c->cmdindex];
3301 curr_sg[sg_index].Len = tmp_sg[i].length;
3302 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3304 tmp_sg[i].length, dir);
3305 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3306 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3307 curr_sg[sg_index].Ext = 0; /* we are not chaining */
3311 cciss_map_sg_chain_block(h, c, h->cmd_sg_list[c->cmdindex],
3312 (seg - (h->max_cmd_sgentries - 1)) *
3313 sizeof(SGDescriptor_struct));
3315 /* track how many SG entries we are using */
3320 printk(KERN_DEBUG "cciss: Submitting %ld sectors in %d segments "
3322 blk_rq_sectors(creq), seg, chained);
3323 #endif /* CCISS_DEBUG */
3325 c->Header.SGTotal = seg + chained;
3326 if (seg <= h->max_cmd_sgentries)
3327 c->Header.SGList = c->Header.SGTotal;
3329 c->Header.SGList = h->max_cmd_sgentries;
3330 set_performant_mode(h, c);
3332 if (likely(creq->cmd_type == REQ_TYPE_FS)) {
3333 if(h->cciss_read == CCISS_READ_10) {
3334 c->Request.CDB[1] = 0;
3335 c->Request.CDB[2] = (start_blk >> 24) & 0xff; /* MSB */
3336 c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3337 c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3338 c->Request.CDB[5] = start_blk & 0xff;
3339 c->Request.CDB[6] = 0; /* (sect >> 24) & 0xff; MSB */
3340 c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3341 c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3342 c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3344 u32 upper32 = upper_32_bits(start_blk);
3346 c->Request.CDBLen = 16;
3347 c->Request.CDB[1]= 0;
3348 c->Request.CDB[2]= (upper32 >> 24) & 0xff; /* MSB */
3349 c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3350 c->Request.CDB[4]= (upper32 >> 8) & 0xff;
3351 c->Request.CDB[5]= upper32 & 0xff;
3352 c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3353 c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3354 c->Request.CDB[8]= (start_blk >> 8) & 0xff;
3355 c->Request.CDB[9]= start_blk & 0xff;
3356 c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3357 c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3358 c->Request.CDB[12]= (blk_rq_sectors(creq) >> 8) & 0xff;
3359 c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3360 c->Request.CDB[14] = c->Request.CDB[15] = 0;
3362 } else if (creq->cmd_type == REQ_TYPE_BLOCK_PC) {
3363 c->Request.CDBLen = creq->cmd_len;
3364 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3366 printk(KERN_WARNING "cciss%d: bad request type %d\n", h->ctlr, creq->cmd_type);
3370 spin_lock_irq(q->queue_lock);
3374 if (h->Qdepth > h->maxQsinceinit)
3375 h->maxQsinceinit = h->Qdepth;
3381 /* We will already have the driver lock here so not need
3387 static inline unsigned long get_next_completion(ctlr_info_t *h)
3389 return h->access.command_completed(h);
3392 static inline int interrupt_pending(ctlr_info_t *h)
3394 return h->access.intr_pending(h);
3397 static inline long interrupt_not_for_us(ctlr_info_t *h)
3399 return !(h->msi_vector || h->msix_vector) &&
3400 ((h->access.intr_pending(h) == 0) ||
3401 (h->interrupts_enabled == 0));
3404 static inline int bad_tag(ctlr_info_t *h, u32 tag_index,
3407 if (unlikely(tag_index >= h->nr_cmds)) {
3408 dev_warn(&h->pdev->dev, "bad tag 0x%08x ignored.\n", raw_tag);
3414 static inline void finish_cmd(ctlr_info_t *h, CommandList_struct *c,
3418 if (likely(c->cmd_type == CMD_RWREQ))
3419 complete_command(h, c, 0);
3420 else if (c->cmd_type == CMD_IOCTL_PEND)
3421 complete(c->waiting);
3422 #ifdef CONFIG_CISS_SCSI_TAPE
3423 else if (c->cmd_type == CMD_SCSI)
3424 complete_scsi_command(c, 0, raw_tag);
3428 static inline u32 next_command(ctlr_info_t *h)
3432 if (unlikely(h->transMethod != CFGTBL_Trans_Performant))
3433 return h->access.command_completed(h);
3435 if ((*(h->reply_pool_head) & 1) == (h->reply_pool_wraparound)) {
3436 a = *(h->reply_pool_head); /* Next cmd in ring buffer */
3437 (h->reply_pool_head)++;
3438 h->commands_outstanding--;
3442 /* Check for wraparound */
3443 if (h->reply_pool_head == (h->reply_pool + h->max_commands)) {
3444 h->reply_pool_head = h->reply_pool;
3445 h->reply_pool_wraparound ^= 1;
3450 /* process completion of an indexed ("direct lookup") command */
3451 static inline u32 process_indexed_cmd(ctlr_info_t *h, u32 raw_tag)
3454 CommandList_struct *c;
3456 tag_index = cciss_tag_to_index(raw_tag);
3457 if (bad_tag(h, tag_index, raw_tag))
3458 return next_command(h);
3459 c = h->cmd_pool + tag_index;
3460 finish_cmd(h, c, raw_tag);
3461 return next_command(h);
3464 /* process completion of a non-indexed command */
3465 static inline u32 process_nonindexed_cmd(ctlr_info_t *h, u32 raw_tag)
3468 CommandList_struct *c = NULL;
3469 struct hlist_node *tmp;
3470 __u32 busaddr_masked, tag_masked;
3472 tag = cciss_tag_discard_error_bits(raw_tag);
3473 hlist_for_each_entry(c, tmp, &h->cmpQ, list) {
3474 busaddr_masked = cciss_tag_discard_error_bits(c->busaddr);
3475 tag_masked = cciss_tag_discard_error_bits(tag);
3476 if (busaddr_masked == tag_masked) {
3477 finish_cmd(h, c, raw_tag);
3478 return next_command(h);
3481 bad_tag(h, h->nr_cmds + 1, raw_tag);
3482 return next_command(h);
3485 static irqreturn_t do_cciss_intx(int irq, void *dev_id)
3487 ctlr_info_t *h = dev_id;
3488 unsigned long flags;
3491 if (interrupt_not_for_us(h))
3494 * If there are completed commands in the completion queue,
3495 * we had better do something about it.
3497 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
3498 while (interrupt_pending(h)) {
3499 raw_tag = get_next_completion(h);
3500 while (raw_tag != FIFO_EMPTY) {
3501 if (cciss_tag_contains_index(raw_tag))
3502 raw_tag = process_indexed_cmd(h, raw_tag);
3504 raw_tag = process_nonindexed_cmd(h, raw_tag);
3508 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
3512 /* Add a second interrupt handler for MSI/MSI-X mode. In this mode we never
3513 * check the interrupt pending register because it is not set.
3515 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id)
3517 ctlr_info_t *h = dev_id;
3518 unsigned long flags;
3521 if (interrupt_not_for_us(h))
3524 * If there are completed commands in the completion queue,
3525 * we had better do something about it.
3527 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
3528 raw_tag = get_next_completion(h);
3529 while (raw_tag != FIFO_EMPTY) {
3530 if (cciss_tag_contains_index(raw_tag))
3531 raw_tag = process_indexed_cmd(h, raw_tag);
3533 raw_tag = process_nonindexed_cmd(h, raw_tag);
3536 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
3541 * add_to_scan_list() - add controller to rescan queue
3542 * @h: Pointer to the controller.
3544 * Adds the controller to the rescan queue if not already on the queue.
3546 * returns 1 if added to the queue, 0 if skipped (could be on the
3547 * queue already, or the controller could be initializing or shutting
3550 static int add_to_scan_list(struct ctlr_info *h)
3552 struct ctlr_info *test_h;
3556 if (h->busy_initializing)
3559 if (!mutex_trylock(&h->busy_shutting_down))
3562 mutex_lock(&scan_mutex);
3563 list_for_each_entry(test_h, &scan_q, scan_list) {
3569 if (!found && !h->busy_scanning) {
3570 INIT_COMPLETION(h->scan_wait);
3571 list_add_tail(&h->scan_list, &scan_q);
3574 mutex_unlock(&scan_mutex);
3575 mutex_unlock(&h->busy_shutting_down);
3581 * remove_from_scan_list() - remove controller from rescan queue
3582 * @h: Pointer to the controller.
3584 * Removes the controller from the rescan queue if present. Blocks if
3585 * the controller is currently conducting a rescan. The controller
3586 * can be in one of three states:
3587 * 1. Doesn't need a scan
3588 * 2. On the scan list, but not scanning yet (we remove it)
3589 * 3. Busy scanning (and not on the list). In this case we want to wait for
3590 * the scan to complete to make sure the scanning thread for this
3591 * controller is completely idle.
3593 static void remove_from_scan_list(struct ctlr_info *h)
3595 struct ctlr_info *test_h, *tmp_h;
3597 mutex_lock(&scan_mutex);
3598 list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3599 if (test_h == h) { /* state 2. */
3600 list_del(&h->scan_list);
3601 complete_all(&h->scan_wait);
3602 mutex_unlock(&scan_mutex);
3606 if (h->busy_scanning) { /* state 3. */
3607 mutex_unlock(&scan_mutex);
3608 wait_for_completion(&h->scan_wait);
3609 } else { /* state 1, nothing to do. */
3610 mutex_unlock(&scan_mutex);
3615 * scan_thread() - kernel thread used to rescan controllers
3618 * A kernel thread used scan for drive topology changes on
3619 * controllers. The thread processes only one controller at a time
3620 * using a queue. Controllers are added to the queue using
3621 * add_to_scan_list() and removed from the queue either after done
3622 * processing or using remove_from_scan_list().
3626 static int scan_thread(void *data)
3628 struct ctlr_info *h;
3631 set_current_state(TASK_INTERRUPTIBLE);
3633 if (kthread_should_stop())
3637 mutex_lock(&scan_mutex);
3638 if (list_empty(&scan_q)) {
3639 mutex_unlock(&scan_mutex);
3643 h = list_entry(scan_q.next,
3646 list_del(&h->scan_list);
3647 h->busy_scanning = 1;
3648 mutex_unlock(&scan_mutex);
3650 rebuild_lun_table(h, 0, 0);
3651 complete_all(&h->scan_wait);
3652 mutex_lock(&scan_mutex);
3653 h->busy_scanning = 0;
3654 mutex_unlock(&scan_mutex);
3661 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3663 if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3666 switch (c->err_info->SenseInfo[12]) {
3668 printk(KERN_WARNING "cciss%d: a state change "
3669 "detected, command retried\n", h->ctlr);
3673 printk(KERN_WARNING "cciss%d: LUN failure "
3674 "detected, action required\n", h->ctlr);
3677 case REPORT_LUNS_CHANGED:
3678 printk(KERN_WARNING "cciss%d: report LUN data "
3679 "changed\n", h->ctlr);
3681 * Here, we could call add_to_scan_list and wake up the scan thread,
3682 * except that it's quite likely that we will get more than one
3683 * REPORT_LUNS_CHANGED condition in quick succession, which means
3684 * that those which occur after the first one will likely happen
3685 * *during* the scan_thread's rescan. And the rescan code is not
3686 * robust enough to restart in the middle, undoing what it has already
3687 * done, and it's not clear that it's even possible to do this, since
3688 * part of what it does is notify the block layer, which starts
3689 * doing it's own i/o to read partition tables and so on, and the
3690 * driver doesn't have visibility to know what might need undoing.
3691 * In any event, if possible, it is horribly complicated to get right
3692 * so we just don't do it for now.
3694 * Note: this REPORT_LUNS_CHANGED condition only occurs on the MSA2012.
3698 case POWER_OR_RESET:
3699 printk(KERN_WARNING "cciss%d: a power on "
3700 "or device reset detected\n", h->ctlr);
3703 case UNIT_ATTENTION_CLEARED:
3704 printk(KERN_WARNING "cciss%d: unit attention "
3705 "cleared by another initiator\n", h->ctlr);
3709 printk(KERN_WARNING "cciss%d: unknown "
3710 "unit attention detected\n", h->ctlr);
3716 * We cannot read the structure directly, for portability we must use
3718 * This is for debug only.
3720 static void print_cfg_table(CfgTable_struct *tb)
3726 printk("Controller Configuration information\n");
3727 printk("------------------------------------\n");
3728 for (i = 0; i < 4; i++)
3729 temp_name[i] = readb(&(tb->Signature[i]));
3730 temp_name[4] = '\0';
3731 printk(" Signature = %s\n", temp_name);
3732 printk(" Spec Number = %d\n", readl(&(tb->SpecValence)));
3733 printk(" Transport methods supported = 0x%x\n",
3734 readl(&(tb->TransportSupport)));
3735 printk(" Transport methods active = 0x%x\n",
3736 readl(&(tb->TransportActive)));
3737 printk(" Requested transport Method = 0x%x\n",
3738 readl(&(tb->HostWrite.TransportRequest)));
3739 printk(" Coalesce Interrupt Delay = 0x%x\n",
3740 readl(&(tb->HostWrite.CoalIntDelay)));
3741 printk(" Coalesce Interrupt Count = 0x%x\n",
3742 readl(&(tb->HostWrite.CoalIntCount)));
3743 printk(" Max outstanding commands = 0x%d\n",
3744 readl(&(tb->CmdsOutMax)));
3745 printk(" Bus Types = 0x%x\n", readl(&(tb->BusTypes)));
3746 for (i = 0; i < 16; i++)
3747 temp_name[i] = readb(&(tb->ServerName[i]));
3748 temp_name[16] = '\0';
3749 printk(" Server Name = %s\n", temp_name);
3750 printk(" Heartbeat Counter = 0x%x\n\n\n", readl(&(tb->HeartBeat)));
3751 #endif /* CCISS_DEBUG */
3754 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3756 int i, offset, mem_type, bar_type;
3757 if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3760 for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3761 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3762 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3765 mem_type = pci_resource_flags(pdev, i) &
3766 PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3768 case PCI_BASE_ADDRESS_MEM_TYPE_32:
3769 case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3770 offset += 4; /* 32 bit */
3772 case PCI_BASE_ADDRESS_MEM_TYPE_64:
3775 default: /* reserved in PCI 2.2 */
3777 "Base address is invalid\n");
3782 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3788 /* Fill in bucket_map[], given nsgs (the max number of
3789 * scatter gather elements supported) and bucket[],
3790 * which is an array of 8 integers. The bucket[] array
3791 * contains 8 different DMA transfer sizes (in 16
3792 * byte increments) which the controller uses to fetch
3793 * commands. This function fills in bucket_map[], which
3794 * maps a given number of scatter gather elements to one of
3795 * the 8 DMA transfer sizes. The point of it is to allow the
3796 * controller to only do as much DMA as needed to fetch the
3797 * command, with the DMA transfer size encoded in the lower
3798 * bits of the command address.
3800 static void calc_bucket_map(int bucket[], int num_buckets,
3801 int nsgs, int *bucket_map)
3805 /* even a command with 0 SGs requires 4 blocks */
3806 #define MINIMUM_TRANSFER_BLOCKS 4
3807 #define NUM_BUCKETS 8
3808 /* Note, bucket_map must have nsgs+1 entries. */
3809 for (i = 0; i <= nsgs; i++) {
3810 /* Compute size of a command with i SG entries */
3811 size = i + MINIMUM_TRANSFER_BLOCKS;
3812 b = num_buckets; /* Assume the biggest bucket */
3813 /* Find the bucket that is just big enough */
3814 for (j = 0; j < 8; j++) {
3815 if (bucket[j] >= size) {
3820 /* for a command with i SG entries, use bucket b. */
3825 static void __devinit cciss_wait_for_mode_change_ack(ctlr_info_t *h)
3829 /* under certain very rare conditions, this can take awhile.
3830 * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3831 * as we enter this code.) */
3832 for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3833 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3839 static __devinit void cciss_enter_performant_mode(ctlr_info_t *h)
3841 /* This is a bit complicated. There are 8 registers on
3842 * the controller which we write to to tell it 8 different
3843 * sizes of commands which there may be. It's a way of
3844 * reducing the DMA done to fetch each command. Encoded into
3845 * each command's tag are 3 bits which communicate to the controller
3846 * which of the eight sizes that command fits within. The size of
3847 * each command depends on how many scatter gather entries there are.
3848 * Each SG entry requires 16 bytes. The eight registers are programmed
3849 * with the number of 16-byte blocks a command of that size requires.
3850 * The smallest command possible requires 5 such 16 byte blocks.
3851 * the largest command possible requires MAXSGENTRIES + 4 16-byte
3852 * blocks. Note, this only extends to the SG entries contained
3853 * within the command block, and does not extend to chained blocks
3854 * of SG elements. bft[] contains the eight values we write to
3855 * the registers. They are not evenly distributed, but have more
3856 * sizes for small commands, and fewer sizes for larger commands.
3859 int bft[8] = { 5, 6, 8, 10, 12, 20, 28, MAXSGENTRIES + 4};
3861 * 5 = 1 s/g entry or 4k
3862 * 6 = 2 s/g entry or 8k
3863 * 8 = 4 s/g entry or 16k
3864 * 10 = 6 s/g entry or 24k
3866 unsigned long register_value;
3867 BUILD_BUG_ON(28 > MAXSGENTRIES + 4);
3869 h->reply_pool_wraparound = 1; /* spec: init to 1 */
3871 /* Controller spec: zero out this buffer. */
3872 memset(h->reply_pool, 0, h->max_commands * sizeof(__u64));
3873 h->reply_pool_head = h->reply_pool;
3875 trans_offset = readl(&(h->cfgtable->TransMethodOffset));
3876 calc_bucket_map(bft, ARRAY_SIZE(bft), h->maxsgentries,
3877 h->blockFetchTable);
3878 writel(bft[0], &h->transtable->BlockFetch0);
3879 writel(bft[1], &h->transtable->BlockFetch1);
3880 writel(bft[2], &h->transtable->BlockFetch2);
3881 writel(bft[3], &h->transtable->BlockFetch3);
3882 writel(bft[4], &h->transtable->BlockFetch4);
3883 writel(bft[5], &h->transtable->BlockFetch5);
3884 writel(bft[6], &h->transtable->BlockFetch6);
3885 writel(bft[7], &h->transtable->BlockFetch7);
3887 /* size of controller ring buffer */
3888 writel(h->max_commands, &h->transtable->RepQSize);
3889 writel(1, &h->transtable->RepQCount);
3890 writel(0, &h->transtable->RepQCtrAddrLow32);
3891 writel(0, &h->transtable->RepQCtrAddrHigh32);
3892 writel(h->reply_pool_dhandle, &h->transtable->RepQAddr0Low32);
3893 writel(0, &h->transtable->RepQAddr0High32);
3894 writel(CFGTBL_Trans_Performant,
3895 &(h->cfgtable->HostWrite.TransportRequest));
3897 writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
3898 cciss_wait_for_mode_change_ack(h);
3899 register_value = readl(&(h->cfgtable->TransportActive));
3900 if (!(register_value & CFGTBL_Trans_Performant))
3901 printk(KERN_WARNING "cciss: unable to get board into"
3902 " performant mode\n");
3905 static void __devinit cciss_put_controller_into_performant_mode(ctlr_info_t *h)
3907 __u32 trans_support;
3909 dev_dbg(&h->pdev->dev, "Trying to put board into Performant mode\n");
3910 /* Attempt to put controller into performant mode if supported */
3911 /* Does board support performant mode? */
3912 trans_support = readl(&(h->cfgtable->TransportSupport));
3913 if (!(trans_support & PERFORMANT_MODE))
3916 printk(KERN_WARNING "cciss%d: Placing controller into "
3917 "performant mode\n", h->ctlr);
3918 /* Performant mode demands commands on a 32 byte boundary
3919 * pci_alloc_consistent aligns on page boundarys already.
3920 * Just need to check if divisible by 32
3922 if ((sizeof(CommandList_struct) % 32) != 0) {
3923 printk(KERN_WARNING "%s %d %s\n",
3924 "cciss info: command size[",
3925 (int)sizeof(CommandList_struct),
3926 "] not divisible by 32, no performant mode..\n");
3930 /* Performant mode ring buffer and supporting data structures */
3931 h->reply_pool = (__u64 *)pci_alloc_consistent(
3932 h->pdev, h->max_commands * sizeof(__u64),
3933 &(h->reply_pool_dhandle));
3935 /* Need a block fetch table for performant mode */
3936 h->blockFetchTable = kmalloc(((h->maxsgentries+1) *
3937 sizeof(__u32)), GFP_KERNEL);
3939 if ((h->reply_pool == NULL) || (h->blockFetchTable == NULL))
3942 cciss_enter_performant_mode(h);
3944 /* Change the access methods to the performant access methods */
3945 h->access = SA5_performant_access;
3946 h->transMethod = CFGTBL_Trans_Performant;
3950 kfree(h->blockFetchTable);
3952 pci_free_consistent(h->pdev,
3953 h->max_commands * sizeof(__u64),
3955 h->reply_pool_dhandle);
3958 } /* cciss_put_controller_into_performant_mode */
3960 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
3961 * controllers that are capable. If not, we use IO-APIC mode.
3964 static void __devinit cciss_interrupt_mode(ctlr_info_t *c)
3966 #ifdef CONFIG_PCI_MSI
3968 struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
3972 /* Some boards advertise MSI but don't really support it */
3973 if ((c->board_id == 0x40700E11) || (c->board_id == 0x40800E11) ||
3974 (c->board_id == 0x40820E11) || (c->board_id == 0x40830E11))
3975 goto default_int_mode;
3977 if (pci_find_capability(c->pdev, PCI_CAP_ID_MSIX)) {
3978 err = pci_enable_msix(c->pdev, cciss_msix_entries, 4);
3980 c->intr[0] = cciss_msix_entries[0].vector;
3981 c->intr[1] = cciss_msix_entries[1].vector;
3982 c->intr[2] = cciss_msix_entries[2].vector;
3983 c->intr[3] = cciss_msix_entries[3].vector;
3988 printk(KERN_WARNING "cciss: only %d MSI-X vectors "
3989 "available\n", err);
3990 goto default_int_mode;
3992 printk(KERN_WARNING "cciss: MSI-X init failed %d\n",
3994 goto default_int_mode;
3997 if (pci_find_capability(c->pdev, PCI_CAP_ID_MSI)) {
3998 if (!pci_enable_msi(c->pdev)) {
4001 printk(KERN_WARNING "cciss: MSI init failed\n");
4005 #endif /* CONFIG_PCI_MSI */
4006 /* if we get here we're going to use the default interrupt mode */
4007 c->intr[PERF_MODE_INT] = c->pdev->irq;
4011 static int __devinit cciss_lookup_board_id(struct pci_dev *pdev, u32 *board_id)
4014 u32 subsystem_vendor_id, subsystem_device_id;
4016 subsystem_vendor_id = pdev->subsystem_vendor;
4017 subsystem_device_id = pdev->subsystem_device;
4018 *board_id = ((subsystem_device_id << 16) & 0xffff0000) |
4019 subsystem_vendor_id;
4021 for (i = 0; i < ARRAY_SIZE(products); i++) {
4022 /* Stand aside for hpsa driver on request */
4023 if (cciss_allow_hpsa && products[i].board_id == HPSA_BOUNDARY)
4025 if (*board_id == products[i].board_id)
4028 dev_warn(&pdev->dev, "unrecognized board ID: 0x%08x, ignoring.\n",
4033 static inline bool cciss_board_disabled(ctlr_info_t *h)
4037 (void) pci_read_config_word(h->pdev, PCI_COMMAND, &command);
4038 return ((command & PCI_COMMAND_MEMORY) == 0);
4041 static int __devinit cciss_pci_find_memory_BAR(struct pci_dev *pdev,
4042 unsigned long *memory_bar)
4046 for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
4047 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM) {
4048 /* addressing mode bits already removed */
4049 *memory_bar = pci_resource_start(pdev, i);
4050 dev_dbg(&pdev->dev, "memory BAR = %lx\n",
4054 dev_warn(&pdev->dev, "no memory BAR found\n");
4058 static int __devinit cciss_wait_for_board_ready(ctlr_info_t *h)
4063 for (i = 0; i < CCISS_BOARD_READY_ITERATIONS; i++) {
4064 scratchpad = readl(h->vaddr + SA5_SCRATCHPAD_OFFSET);
4065 if (scratchpad == CCISS_FIRMWARE_READY)
4067 msleep(CCISS_BOARD_READY_POLL_INTERVAL_MSECS);
4069 dev_warn(&h->pdev->dev, "board not ready, timed out.\n");
4073 static int __devinit cciss_find_cfg_addrs(struct pci_dev *pdev,
4074 void __iomem *vaddr, u32 *cfg_base_addr, u64 *cfg_base_addr_index,
4077 *cfg_base_addr = readl(vaddr + SA5_CTCFG_OFFSET);
4078 *cfg_offset = readl(vaddr + SA5_CTMEM_OFFSET);
4079 *cfg_base_addr &= (u32) 0x0000ffff;
4080 *cfg_base_addr_index = find_PCI_BAR_index(pdev, *cfg_base_addr);
4081 if (*cfg_base_addr_index == -1) {
4082 dev_warn(&pdev->dev, "cannot find cfg_base_addr_index, "
4083 "*cfg_base_addr = 0x%08x\n", *cfg_base_addr);
4089 static int __devinit cciss_find_cfgtables(ctlr_info_t *h)
4093 u64 cfg_base_addr_index;
4097 rc = cciss_find_cfg_addrs(h->pdev, h->vaddr, &cfg_base_addr,
4098 &cfg_base_addr_index, &cfg_offset);
4101 h->cfgtable = remap_pci_mem(pci_resource_start(h->pdev,
4102 cfg_base_addr_index) + cfg_offset, sizeof(h->cfgtable));
4105 /* Find performant mode table. */
4106 trans_offset = readl(&h->cfgtable->TransMethodOffset);
4107 h->transtable = remap_pci_mem(pci_resource_start(h->pdev,
4108 cfg_base_addr_index)+cfg_offset+trans_offset,
4109 sizeof(*h->transtable));
4115 static void __devinit cciss_get_max_perf_mode_cmds(struct ctlr_info *h)
4117 h->max_commands = readl(&(h->cfgtable->MaxPerformantModeCommands));
4118 if (h->max_commands < 16) {
4119 dev_warn(&h->pdev->dev, "Controller reports "
4120 "max supported commands of %d, an obvious lie. "
4121 "Using 16. Ensure that firmware is up to date.\n",
4123 h->max_commands = 16;
4127 /* Interrogate the hardware for some limits:
4128 * max commands, max SG elements without chaining, and with chaining,
4129 * SG chain block size, etc.
4131 static void __devinit cciss_find_board_params(ctlr_info_t *h)
4133 cciss_get_max_perf_mode_cmds(h);
4134 h->nr_cmds = h->max_commands - 4; /* Allow room for some ioctls */
4135 h->maxsgentries = readl(&(h->cfgtable->MaxSGElements));
4137 * Limit in-command s/g elements to 32 save dma'able memory.
4138 * Howvever spec says if 0, use 31
4140 h->max_cmd_sgentries = 31;
4141 if (h->maxsgentries > 512) {
4142 h->max_cmd_sgentries = 32;
4143 h->chainsize = h->maxsgentries - h->max_cmd_sgentries + 1;
4144 h->maxsgentries--; /* save one for chain pointer */
4146 h->maxsgentries = 31; /* default to traditional values */
4151 static inline bool CISS_signature_present(ctlr_info_t *h)
4153 if ((readb(&h->cfgtable->Signature[0]) != 'C') ||
4154 (readb(&h->cfgtable->Signature[1]) != 'I') ||
4155 (readb(&h->cfgtable->Signature[2]) != 'S') ||
4156 (readb(&h->cfgtable->Signature[3]) != 'S')) {
4157 dev_warn(&h->pdev->dev, "not a valid CISS config table\n");
4163 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
4164 static inline void cciss_enable_scsi_prefetch(ctlr_info_t *h)
4169 prefetch = readl(&(h->cfgtable->SCSI_Prefetch));
4171 writel(prefetch, &(h->cfgtable->SCSI_Prefetch));
4175 /* Disable DMA prefetch for the P600. Otherwise an ASIC bug may result
4176 * in a prefetch beyond physical memory.
4178 static inline void cciss_p600_dma_prefetch_quirk(ctlr_info_t *h)
4183 if (h->board_id != 0x3225103C)
4185 dma_prefetch = readl(h->vaddr + I2O_DMA1_CFG);
4186 dma_prefetch |= 0x8000;
4187 writel(dma_prefetch, h->vaddr + I2O_DMA1_CFG);
4188 pci_read_config_dword(h->pdev, PCI_COMMAND_PARITY, &dma_refetch);
4190 pci_write_config_dword(h->pdev, PCI_COMMAND_PARITY, dma_refetch);
4193 static int __devinit cciss_pci_init(ctlr_info_t *c)
4195 int prod_index, err;
4197 prod_index = cciss_lookup_board_id(c->pdev, &c->board_id);
4200 c->product_name = products[prod_index].product_name;
4201 c->access = *(products[prod_index].access);
4203 if (cciss_board_disabled(c)) {
4205 "cciss: controller appears to be disabled\n");
4208 err = pci_enable_device(c->pdev);
4210 printk(KERN_ERR "cciss: Unable to Enable PCI device\n");
4214 err = pci_request_regions(c->pdev, "cciss");
4216 printk(KERN_ERR "cciss: Cannot obtain PCI resources, "
4222 printk(KERN_INFO "command = %x\n", command);
4223 printk(KERN_INFO "irq = %x\n", c->pdev->irq);
4224 printk(KERN_INFO "board_id = %x\n", c->board_id);
4225 #endif /* CCISS_DEBUG */
4227 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
4228 * else we use the IO-APIC interrupt assigned to us by system ROM.
4230 cciss_interrupt_mode(c);
4231 err = cciss_pci_find_memory_BAR(c->pdev, &c->paddr);
4233 goto err_out_free_res;
4234 c->vaddr = remap_pci_mem(c->paddr, 0x250);
4237 goto err_out_free_res;
4239 err = cciss_wait_for_board_ready(c);
4241 goto err_out_free_res;
4242 err = cciss_find_cfgtables(c);
4244 goto err_out_free_res;
4245 print_cfg_table(c->cfgtable);
4246 cciss_find_board_params(c);
4248 if (!CISS_signature_present(c)) {
4250 goto err_out_free_res;
4252 cciss_enable_scsi_prefetch(c);
4253 cciss_p600_dma_prefetch_quirk(c);
4254 cciss_put_controller_into_performant_mode(c);
4259 * Deliberately omit pci_disable_device(): it does something nasty to
4260 * Smart Array controllers that pci_enable_device does not undo
4263 iounmap(c->transtable);
4265 iounmap(c->cfgtable);
4268 pci_release_regions(c->pdev);
4272 /* Function to find the first free pointer into our hba[] array
4273 * Returns -1 if no free entries are left.
4275 static int alloc_cciss_hba(void)
4279 for (i = 0; i < MAX_CTLR; i++) {
4283 p = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
4290 printk(KERN_WARNING "cciss: This driver supports a maximum"
4291 " of %d controllers.\n", MAX_CTLR);
4294 printk(KERN_ERR "cciss: out of memory.\n");
4298 static void free_hba(int n)
4300 ctlr_info_t *h = hba[n];
4304 for (i = 0; i < h->highest_lun + 1; i++)
4305 if (h->gendisk[i] != NULL)
4306 put_disk(h->gendisk[i]);
4310 /* Send a message CDB to the firmware. */
4311 static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
4314 CommandListHeader_struct CommandHeader;
4315 RequestBlock_struct Request;
4316 ErrDescriptor_struct ErrorDescriptor;
4318 static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
4321 uint32_t paddr32, tag;
4322 void __iomem *vaddr;
4325 vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
4329 /* The Inbound Post Queue only accepts 32-bit physical addresses for the
4330 CCISS commands, so they must be allocated from the lower 4GiB of
4332 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
4338 cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
4344 /* This must fit, because of the 32-bit consistent DMA mask. Also,
4345 although there's no guarantee, we assume that the address is at
4346 least 4-byte aligned (most likely, it's page-aligned). */
4349 cmd->CommandHeader.ReplyQueue = 0;
4350 cmd->CommandHeader.SGList = 0;
4351 cmd->CommandHeader.SGTotal = 0;
4352 cmd->CommandHeader.Tag.lower = paddr32;
4353 cmd->CommandHeader.Tag.upper = 0;
4354 memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
4356 cmd->Request.CDBLen = 16;
4357 cmd->Request.Type.Type = TYPE_MSG;
4358 cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
4359 cmd->Request.Type.Direction = XFER_NONE;
4360 cmd->Request.Timeout = 0; /* Don't time out */
4361 cmd->Request.CDB[0] = opcode;
4362 cmd->Request.CDB[1] = type;
4363 memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
4365 cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
4366 cmd->ErrorDescriptor.Addr.upper = 0;
4367 cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
4369 writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
4371 for (i = 0; i < 10; i++) {
4372 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
4373 if ((tag & ~3) == paddr32)
4375 schedule_timeout_uninterruptible(HZ);
4380 /* we leak the DMA buffer here ... no choice since the controller could
4381 still complete the command. */
4383 printk(KERN_ERR "cciss: controller message %02x:%02x timed out\n",
4388 pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
4391 printk(KERN_ERR "cciss: controller message %02x:%02x failed\n",
4396 printk(KERN_INFO "cciss: controller message %02x:%02x succeeded\n",
4401 #define cciss_soft_reset_controller(p) cciss_message(p, 1, 0)
4402 #define cciss_noop(p) cciss_message(p, 3, 0)
4404 static __devinit int cciss_reset_msi(struct pci_dev *pdev)
4406 /* the #defines are stolen from drivers/pci/msi.h. */
4407 #define msi_control_reg(base) (base + PCI_MSI_FLAGS)
4408 #define PCI_MSIX_FLAGS_ENABLE (1 << 15)
4413 pos = pci_find_capability(pdev, PCI_CAP_ID_MSI);
4415 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4416 if (control & PCI_MSI_FLAGS_ENABLE) {
4417 printk(KERN_INFO "cciss: resetting MSI\n");
4418 pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSI_FLAGS_ENABLE);
4422 pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
4424 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4425 if (control & PCI_MSIX_FLAGS_ENABLE) {
4426 printk(KERN_INFO "cciss: resetting MSI-X\n");
4427 pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSIX_FLAGS_ENABLE);
4434 static int cciss_controller_hard_reset(struct pci_dev *pdev,
4435 void * __iomem vaddr, bool use_doorbell)
4441 /* For everything after the P600, the PCI power state method
4442 * of resetting the controller doesn't work, so we have this
4443 * other way using the doorbell register.
4445 dev_info(&pdev->dev, "using doorbell to reset controller\n");
4446 writel(DOORBELL_CTLR_RESET, vaddr + SA5_DOORBELL);
4448 } else { /* Try to do it the PCI power state way */
4450 /* Quoting from the Open CISS Specification: "The Power
4451 * Management Control/Status Register (CSR) controls the power
4452 * state of the device. The normal operating state is D0,
4453 * CSR=00h. The software off state is D3, CSR=03h. To reset
4454 * the controller, place the interface device in D3 then to D0,
4455 * this causes a secondary PCI reset which will reset the
4458 pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4461 "cciss_controller_hard_reset: "
4462 "PCI PM not supported\n");
4465 dev_info(&pdev->dev, "using PCI PM to reset controller\n");
4466 /* enter the D3hot power management state */
4467 pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4468 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4470 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4474 /* enter the D0 power management state */
4475 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4477 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4484 /* This does a hard reset of the controller using PCI power management
4485 * states or using the doorbell register. */
4486 static __devinit int cciss_kdump_hard_reset_controller(struct pci_dev *pdev)
4488 u16 saved_config_space[32];
4491 u64 cfg_base_addr_index;
4492 void __iomem *vaddr;
4493 unsigned long paddr;
4494 u32 misc_fw_support, active_transport;
4496 CfgTable_struct __iomem *cfgtable;
4500 /* For controllers as old a the p600, this is very nearly
4503 * pci_save_state(pci_dev);
4504 * pci_set_power_state(pci_dev, PCI_D3hot);
4505 * pci_set_power_state(pci_dev, PCI_D0);
4506 * pci_restore_state(pci_dev);
4508 * but we can't use these nice canned kernel routines on
4509 * kexec, because they also check the MSI/MSI-X state in PCI
4510 * configuration space and do the wrong thing when it is
4511 * set/cleared. Also, the pci_save/restore_state functions
4512 * violate the ordering requirements for restoring the
4513 * configuration space from the CCISS document (see the
4514 * comment below). So we roll our own ....
4516 * For controllers newer than the P600, the pci power state
4517 * method of resetting doesn't work so we have another way
4518 * using the doorbell register.
4521 /* Exclude 640x boards. These are two pci devices in one slot
4522 * which share a battery backed cache module. One controls the
4523 * cache, the other accesses the cache through the one that controls
4524 * it. If we reset the one controlling the cache, the other will
4525 * likely not be happy. Just forbid resetting this conjoined mess.
4527 cciss_lookup_board_id(pdev, &board_id);
4528 if (board_id == 0x409C0E11 || board_id == 0x409D0E11) {
4529 dev_warn(&pdev->dev, "Cannot reset Smart Array 640x "
4530 "due to shared cache module.");
4534 for (i = 0; i < 32; i++)
4535 pci_read_config_word(pdev, 2*i, &saved_config_space[i]);
4537 /* find the first memory BAR, so we can find the cfg table */
4538 rc = cciss_pci_find_memory_BAR(pdev, &paddr);
4541 vaddr = remap_pci_mem(paddr, 0x250);
4545 /* find cfgtable in order to check if reset via doorbell is supported */
4546 rc = cciss_find_cfg_addrs(pdev, vaddr, &cfg_base_addr,
4547 &cfg_base_addr_index, &cfg_offset);
4550 cfgtable = remap_pci_mem(pci_resource_start(pdev,
4551 cfg_base_addr_index) + cfg_offset, sizeof(*cfgtable));
4557 /* If reset via doorbell register is supported, use that. */
4558 misc_fw_support = readl(&cfgtable->misc_fw_support);
4559 use_doorbell = misc_fw_support & MISC_FW_DOORBELL_RESET;
4561 rc = cciss_controller_hard_reset(pdev, vaddr, use_doorbell);
4563 goto unmap_cfgtable;
4565 /* Restore the PCI configuration space. The Open CISS
4566 * Specification says, "Restore the PCI Configuration
4567 * Registers, offsets 00h through 60h. It is important to
4568 * restore the command register, 16-bits at offset 04h,
4569 * last. Do not restore the configuration status register,
4570 * 16-bits at offset 06h." Note that the offset is 2*i.
4572 for (i = 0; i < 32; i++) {
4573 if (i == 2 || i == 3)
4575 pci_write_config_word(pdev, 2*i, saved_config_space[i]);
4578 pci_write_config_word(pdev, 4, saved_config_space[2]);
4580 /* Some devices (notably the HP Smart Array 5i Controller)
4581 need a little pause here */
4582 msleep(CCISS_POST_RESET_PAUSE_MSECS);
4584 /* Controller should be in simple mode at this point. If it's not,
4585 * It means we're on one of those controllers which doesn't support
4586 * the doorbell reset method and on which the PCI power management reset
4587 * method doesn't work (P800, for example.)
4588 * In those cases, don't try to proceed, as it generally doesn't work.
4590 active_transport = readl(&cfgtable->TransportActive);
4591 if (active_transport & PERFORMANT_MODE) {
4592 dev_warn(&pdev->dev, "Unable to successfully reset controller,"
4593 " Ignoring controller.\n");
4605 static __devinit int cciss_init_reset_devices(struct pci_dev *pdev)
4612 /* Reset the controller with a PCI power-cycle or via doorbell */
4613 rc = cciss_kdump_hard_reset_controller(pdev);
4615 /* -ENOTSUPP here means we cannot reset the controller
4616 * but it's already (and still) up and running in
4617 * "performant mode". Or, it might be 640x, which can't reset
4618 * due to concerns about shared bbwc between 6402/6404 pair.
4620 if (rc == -ENOTSUPP)
4621 return 0; /* just try to do the kdump anyhow. */
4624 if (cciss_reset_msi(pdev))
4627 /* Now try to get the controller to respond to a no-op */
4628 for (i = 0; i < CCISS_POST_RESET_NOOP_RETRIES; i++) {
4629 if (cciss_noop(pdev) == 0)
4632 dev_warn(&pdev->dev, "no-op failed%s\n",
4633 (i < CCISS_POST_RESET_NOOP_RETRIES - 1 ?
4634 "; re-trying" : ""));
4635 msleep(CCISS_POST_RESET_NOOP_INTERVAL_MSECS);
4641 * This is it. Find all the controllers and register them. I really hate
4642 * stealing all these major device numbers.
4643 * returns the number of block devices registered.
4645 static int __devinit cciss_init_one(struct pci_dev *pdev,
4646 const struct pci_device_id *ent)
4652 int dac, return_code;
4653 InquiryData_struct *inq_buff;
4655 rc = cciss_init_reset_devices(pdev);
4658 i = alloc_cciss_hba();
4662 hba[i]->pdev = pdev;
4663 hba[i]->busy_initializing = 1;
4664 INIT_HLIST_HEAD(&hba[i]->cmpQ);
4665 INIT_HLIST_HEAD(&hba[i]->reqQ);
4666 mutex_init(&hba[i]->busy_shutting_down);
4668 if (cciss_pci_init(hba[i]) != 0)
4669 goto clean_no_release_regions;
4671 sprintf(hba[i]->devname, "cciss%d", i);
4674 init_completion(&hba[i]->scan_wait);
4676 if (cciss_create_hba_sysfs_entry(hba[i]))
4679 /* configure PCI DMA stuff */
4680 if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
4682 else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
4685 printk(KERN_ERR "cciss: no suitable DMA available\n");
4690 * register with the major number, or get a dynamic major number
4691 * by passing 0 as argument. This is done for greater than
4692 * 8 controller support.
4694 if (i < MAX_CTLR_ORIG)
4695 hba[i]->major = COMPAQ_CISS_MAJOR + i;
4696 rc = register_blkdev(hba[i]->major, hba[i]->devname);
4697 if (rc == -EBUSY || rc == -EINVAL) {
4699 "cciss: Unable to get major number %d for %s "
4700 "on hba %d\n", hba[i]->major, hba[i]->devname, i);
4703 if (i >= MAX_CTLR_ORIG)
4707 /* make sure the board interrupts are off */
4708 hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_OFF);
4709 if (hba[i]->msi_vector || hba[i]->msix_vector) {
4710 if (request_irq(hba[i]->intr[PERF_MODE_INT],
4712 IRQF_DISABLED, hba[i]->devname, hba[i])) {
4713 printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
4714 hba[i]->intr[PERF_MODE_INT], hba[i]->devname);
4718 if (request_irq(hba[i]->intr[PERF_MODE_INT], do_cciss_intx,
4719 IRQF_DISABLED, hba[i]->devname, hba[i])) {
4720 printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
4721 hba[i]->intr[PERF_MODE_INT], hba[i]->devname);
4726 printk(KERN_INFO "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
4727 hba[i]->devname, pdev->device, pci_name(pdev),
4728 hba[i]->intr[PERF_MODE_INT], dac ? "" : " not");
4730 hba[i]->cmd_pool_bits =
4731 kmalloc(DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4732 * sizeof(unsigned long), GFP_KERNEL);
4733 hba[i]->cmd_pool = (CommandList_struct *)
4734 pci_alloc_consistent(hba[i]->pdev,
4735 hba[i]->nr_cmds * sizeof(CommandList_struct),
4736 &(hba[i]->cmd_pool_dhandle));
4737 hba[i]->errinfo_pool = (ErrorInfo_struct *)
4738 pci_alloc_consistent(hba[i]->pdev,
4739 hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4740 &(hba[i]->errinfo_pool_dhandle));
4741 if ((hba[i]->cmd_pool_bits == NULL)
4742 || (hba[i]->cmd_pool == NULL)
4743 || (hba[i]->errinfo_pool == NULL)) {
4744 printk(KERN_ERR "cciss: out of memory");
4748 /* Need space for temp scatter list */
4749 hba[i]->scatter_list = kmalloc(hba[i]->max_commands *
4750 sizeof(struct scatterlist *),
4752 for (k = 0; k < hba[i]->nr_cmds; k++) {
4753 hba[i]->scatter_list[k] = kmalloc(sizeof(struct scatterlist) *
4754 hba[i]->maxsgentries,
4756 if (hba[i]->scatter_list[k] == NULL) {
4757 printk(KERN_ERR "cciss%d: could not allocate "
4762 hba[i]->cmd_sg_list = cciss_allocate_sg_chain_blocks(hba[i],
4763 hba[i]->chainsize, hba[i]->nr_cmds);
4764 if (!hba[i]->cmd_sg_list && hba[i]->chainsize > 0)
4767 spin_lock_init(&hba[i]->lock);
4769 /* Initialize the pdev driver private data.
4770 have it point to hba[i]. */
4771 pci_set_drvdata(pdev, hba[i]);
4772 /* command and error info recs zeroed out before
4774 memset(hba[i]->cmd_pool_bits, 0,
4775 DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4776 * sizeof(unsigned long));
4778 hba[i]->num_luns = 0;
4779 hba[i]->highest_lun = -1;
4780 for (j = 0; j < CISS_MAX_LUN; j++) {
4781 hba[i]->drv[j] = NULL;
4782 hba[i]->gendisk[j] = NULL;
4785 cciss_scsi_setup(i);
4787 /* Turn the interrupts on so we can service requests */
4788 hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_ON);
4790 /* Get the firmware version */
4791 inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
4792 if (inq_buff == NULL) {
4793 printk(KERN_ERR "cciss: out of memory\n");
4797 return_code = sendcmd_withirq(CISS_INQUIRY, i, inq_buff,
4798 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
4799 if (return_code == IO_OK) {
4800 hba[i]->firm_ver[0] = inq_buff->data_byte[32];
4801 hba[i]->firm_ver[1] = inq_buff->data_byte[33];
4802 hba[i]->firm_ver[2] = inq_buff->data_byte[34];
4803 hba[i]->firm_ver[3] = inq_buff->data_byte[35];
4804 } else { /* send command failed */
4805 printk(KERN_WARNING "cciss: unable to determine firmware"
4806 " version of controller\n");
4812 hba[i]->cciss_max_sectors = 8192;
4814 rebuild_lun_table(hba[i], 1, 0);
4815 hba[i]->busy_initializing = 0;
4819 kfree(hba[i]->cmd_pool_bits);
4820 /* Free up sg elements */
4821 for (k = 0; k < hba[i]->nr_cmds; k++)
4822 kfree(hba[i]->scatter_list[k]);
4823 kfree(hba[i]->scatter_list);
4824 cciss_free_sg_chain_blocks(hba[i]->cmd_sg_list, hba[i]->nr_cmds);
4825 if (hba[i]->cmd_pool)
4826 pci_free_consistent(hba[i]->pdev,
4827 hba[i]->nr_cmds * sizeof(CommandList_struct),
4828 hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4829 if (hba[i]->errinfo_pool)
4830 pci_free_consistent(hba[i]->pdev,
4831 hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4832 hba[i]->errinfo_pool,
4833 hba[i]->errinfo_pool_dhandle);
4834 free_irq(hba[i]->intr[PERF_MODE_INT], hba[i]);
4836 unregister_blkdev(hba[i]->major, hba[i]->devname);
4838 cciss_destroy_hba_sysfs_entry(hba[i]);
4840 pci_release_regions(pdev);
4841 clean_no_release_regions:
4842 hba[i]->busy_initializing = 0;
4845 * Deliberately omit pci_disable_device(): it does something nasty to
4846 * Smart Array controllers that pci_enable_device does not undo
4848 pci_set_drvdata(pdev, NULL);
4853 static void cciss_shutdown(struct pci_dev *pdev)
4859 h = pci_get_drvdata(pdev);
4860 flush_buf = kzalloc(4, GFP_KERNEL);
4863 "cciss:%d cache not flushed, out of memory.\n",
4867 /* write all data in the battery backed cache to disk */
4868 memset(flush_buf, 0, 4);
4869 return_code = sendcmd_withirq(CCISS_CACHE_FLUSH, h->ctlr, flush_buf,
4870 4, 0, CTLR_LUNID, TYPE_CMD);
4872 if (return_code != IO_OK)
4873 printk(KERN_WARNING "cciss%d: Error flushing cache\n",
4875 h->access.set_intr_mask(h, CCISS_INTR_OFF);
4876 free_irq(h->intr[PERF_MODE_INT], h);
4879 static void __devexit cciss_remove_one(struct pci_dev *pdev)
4881 ctlr_info_t *tmp_ptr;
4884 if (pci_get_drvdata(pdev) == NULL) {
4885 printk(KERN_ERR "cciss: Unable to remove device \n");
4889 tmp_ptr = pci_get_drvdata(pdev);
4891 if (hba[i] == NULL) {
4892 printk(KERN_ERR "cciss: device appears to "
4893 "already be removed \n");
4897 mutex_lock(&hba[i]->busy_shutting_down);
4899 remove_from_scan_list(hba[i]);
4900 remove_proc_entry(hba[i]->devname, proc_cciss);
4901 unregister_blkdev(hba[i]->major, hba[i]->devname);
4903 /* remove it from the disk list */
4904 for (j = 0; j < CISS_MAX_LUN; j++) {
4905 struct gendisk *disk = hba[i]->gendisk[j];
4907 struct request_queue *q = disk->queue;
4909 if (disk->flags & GENHD_FL_UP) {
4910 cciss_destroy_ld_sysfs_entry(hba[i], j, 1);
4914 blk_cleanup_queue(q);
4918 #ifdef CONFIG_CISS_SCSI_TAPE
4919 cciss_unregister_scsi(i); /* unhook from SCSI subsystem */
4922 cciss_shutdown(pdev);
4924 #ifdef CONFIG_PCI_MSI
4925 if (hba[i]->msix_vector)
4926 pci_disable_msix(hba[i]->pdev);
4927 else if (hba[i]->msi_vector)
4928 pci_disable_msi(hba[i]->pdev);
4929 #endif /* CONFIG_PCI_MSI */
4931 iounmap(hba[i]->transtable);
4932 iounmap(hba[i]->cfgtable);
4933 iounmap(hba[i]->vaddr);
4935 pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(CommandList_struct),
4936 hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4937 pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4938 hba[i]->errinfo_pool, hba[i]->errinfo_pool_dhandle);
4939 kfree(hba[i]->cmd_pool_bits);
4940 /* Free up sg elements */
4941 for (j = 0; j < hba[i]->nr_cmds; j++)
4942 kfree(hba[i]->scatter_list[j]);
4943 kfree(hba[i]->scatter_list);
4944 cciss_free_sg_chain_blocks(hba[i]->cmd_sg_list, hba[i]->nr_cmds);
4946 * Deliberately omit pci_disable_device(): it does something nasty to
4947 * Smart Array controllers that pci_enable_device does not undo
4949 pci_release_regions(pdev);
4950 pci_set_drvdata(pdev, NULL);
4951 cciss_destroy_hba_sysfs_entry(hba[i]);
4952 mutex_unlock(&hba[i]->busy_shutting_down);
4956 static struct pci_driver cciss_pci_driver = {
4958 .probe = cciss_init_one,
4959 .remove = __devexit_p(cciss_remove_one),
4960 .id_table = cciss_pci_device_id, /* id_table */
4961 .shutdown = cciss_shutdown,
4965 * This is it. Register the PCI driver information for the cards we control
4966 * the OS will call our registered routines when it finds one of our cards.
4968 static int __init cciss_init(void)
4973 * The hardware requires that commands are aligned on a 64-bit
4974 * boundary. Given that we use pci_alloc_consistent() to allocate an
4975 * array of them, the size must be a multiple of 8 bytes.
4977 BUILD_BUG_ON(sizeof(CommandList_struct) % COMMANDLIST_ALIGNMENT);
4978 printk(KERN_INFO DRIVER_NAME "\n");
4980 err = bus_register(&cciss_bus_type);
4984 /* Start the scan thread */
4985 cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
4986 if (IS_ERR(cciss_scan_thread)) {
4987 err = PTR_ERR(cciss_scan_thread);
4988 goto err_bus_unregister;
4991 /* Register for our PCI devices */
4992 err = pci_register_driver(&cciss_pci_driver);
4994 goto err_thread_stop;
4999 kthread_stop(cciss_scan_thread);
5001 bus_unregister(&cciss_bus_type);
5006 static void __exit cciss_cleanup(void)
5010 pci_unregister_driver(&cciss_pci_driver);
5011 /* double check that all controller entrys have been removed */
5012 for (i = 0; i < MAX_CTLR; i++) {
5013 if (hba[i] != NULL) {
5014 printk(KERN_WARNING "cciss: had to remove"
5015 " controller %d\n", i);
5016 cciss_remove_one(hba[i]->pdev);
5019 kthread_stop(cciss_scan_thread);
5020 remove_proc_entry("driver/cciss", NULL);
5021 bus_unregister(&cciss_bus_type);
5024 module_init(cciss_init);
5025 module_exit(cciss_cleanup);