1/*
2 *    Disk Array driver for HP Smart Array controllers.
3 *    (C) Copyright 2000, 2007 Hewlett-Packard Development Company, L.P.
4 *
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.
8 *
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.
13 *
14 *    You should have received a copy of the GNU General Public License
15 *    along with this program; if not, write to the Free Software
16 *    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
17 *    02111-1307, USA.
18 *
19 *    Questions/Comments/Bugfixes to iss_storagedev@hp.com
20 *
21 */
22
23#include <linux/module.h>
24#include <linux/interrupt.h>
25#include <linux/types.h>
26#include <linux/pci.h>
27#include <linux/pci-aspm.h>
28#include <linux/kernel.h>
29#include <linux/slab.h>
30#include <linux/delay.h>
31#include <linux/major.h>
32#include <linux/fs.h>
33#include <linux/bio.h>
34#include <linux/blkpg.h>
35#include <linux/timer.h>
36#include <linux/proc_fs.h>
37#include <linux/seq_file.h>
38#include <linux/init.h>
39#include <linux/jiffies.h>
40#include <linux/hdreg.h>
41#include <linux/spinlock.h>
42#include <linux/compat.h>
43#include <linux/mutex.h>
44#include <asm/uaccess.h>
45#include <asm/io.h>
46
47#include <linux/dma-mapping.h>
48#include <linux/blkdev.h>
49#include <linux/genhd.h>
50#include <linux/completion.h>
51#include <scsi/scsi.h>
52#include <scsi/sg.h>
53#include <scsi/scsi_ioctl.h>
54#include <linux/cdrom.h>
55#include <linux/scatterlist.h>
56#include <linux/kthread.h>
57
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)
61
62/* Embedded module documentation macros - see modules.h */
63MODULE_AUTHOR("Hewlett-Packard Company");
64MODULE_DESCRIPTION("Driver for HP Smart Array Controllers");
65MODULE_SUPPORTED_DEVICE("HP Smart Array Controllers");
66MODULE_VERSION("3.6.26");
67MODULE_LICENSE("GPL");
68static int cciss_tape_cmds = 6;
69module_param(cciss_tape_cmds, int, 0644);
70MODULE_PARM_DESC(cciss_tape_cmds,
71	"number of commands to allocate for tape devices (default: 6)");
72static int cciss_simple_mode;
73module_param(cciss_simple_mode, int, S_IRUGO|S_IWUSR);
74MODULE_PARM_DESC(cciss_simple_mode,
75	"Use 'simple mode' rather than 'performant mode'");
76
77static DEFINE_MUTEX(cciss_mutex);
78static struct proc_dir_entry *proc_cciss;
79
80#include "cciss_cmd.h"
81#include "cciss.h"
82#include <linux/cciss_ioctl.h>
83
84/* define the PCI info for the cards we can control */
85static const struct pci_device_id cciss_pci_device_id[] = {
86	{PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISS,  0x0E11, 0x4070},
87	{PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4080},
88	{PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4082},
89	{PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4083},
90	{PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x4091},
91	{PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409A},
92	{PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409B},
93	{PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409C},
94	{PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409D},
95	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSA,     0x103C, 0x3225},
96	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3223},
97	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3234},
98	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3235},
99	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3211},
100	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3212},
101	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3213},
102	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3214},
103	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3215},
104	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3237},
105	{PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x323D},
106	{0,}
107};
108
109MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
110
111/*  board_id = Subsystem Device ID & Vendor ID
112 *  product = Marketing Name for the board
113 *  access = Address of the struct of function pointers
114 */
115static struct board_type products[] = {
116	{0x40700E11, "Smart Array 5300", &SA5_access},
117	{0x40800E11, "Smart Array 5i", &SA5B_access},
118	{0x40820E11, "Smart Array 532", &SA5B_access},
119	{0x40830E11, "Smart Array 5312", &SA5B_access},
120	{0x409A0E11, "Smart Array 641", &SA5_access},
121	{0x409B0E11, "Smart Array 642", &SA5_access},
122	{0x409C0E11, "Smart Array 6400", &SA5_access},
123	{0x409D0E11, "Smart Array 6400 EM", &SA5_access},
124	{0x40910E11, "Smart Array 6i", &SA5_access},
125	{0x3225103C, "Smart Array P600", &SA5_access},
126	{0x3223103C, "Smart Array P800", &SA5_access},
127	{0x3234103C, "Smart Array P400", &SA5_access},
128	{0x3235103C, "Smart Array P400i", &SA5_access},
129	{0x3211103C, "Smart Array E200i", &SA5_access},
130	{0x3212103C, "Smart Array E200", &SA5_access},
131	{0x3213103C, "Smart Array E200i", &SA5_access},
132	{0x3214103C, "Smart Array E200i", &SA5_access},
133	{0x3215103C, "Smart Array E200i", &SA5_access},
134	{0x3237103C, "Smart Array E500", &SA5_access},
135	{0x3223103C, "Smart Array P800", &SA5_access},
136	{0x3234103C, "Smart Array P400", &SA5_access},
137	{0x323D103C, "Smart Array P700m", &SA5_access},
138};
139
140/* How long to wait (in milliseconds) for board to go into simple mode */
141#define MAX_CONFIG_WAIT 30000
142#define MAX_IOCTL_CONFIG_WAIT 1000
143
144/*define how many times we will try a command because of bus resets */
145#define MAX_CMD_RETRIES 3
146
147#define MAX_CTLR	32
148
149/* Originally cciss driver only supports 8 major numbers */
150#define MAX_CTLR_ORIG 	8
151
152static ctlr_info_t *hba[MAX_CTLR];
153
154static struct task_struct *cciss_scan_thread;
155static DEFINE_MUTEX(scan_mutex);
156static LIST_HEAD(scan_q);
157
158static void do_cciss_request(struct request_queue *q);
159static irqreturn_t do_cciss_intx(int irq, void *dev_id);
160static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id);
161static int cciss_open(struct block_device *bdev, fmode_t mode);
162static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode);
163static int cciss_release(struct gendisk *disk, fmode_t mode);
164static int do_ioctl(struct block_device *bdev, fmode_t mode,
165		    unsigned int cmd, unsigned long arg);
166static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
167		       unsigned int cmd, unsigned long arg);
168static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
169
170static int cciss_revalidate(struct gendisk *disk);
171static int rebuild_lun_table(ctlr_info_t *h, int first_time, int via_ioctl);
172static int deregister_disk(ctlr_info_t *h, int drv_index,
173			   int clear_all, int via_ioctl);
174
175static void cciss_read_capacity(ctlr_info_t *h, int logvol,
176			sector_t *total_size, unsigned int *block_size);
177static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
178			sector_t *total_size, unsigned int *block_size);
179static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
180			sector_t total_size,
181			unsigned int block_size, InquiryData_struct *inq_buff,
182				   drive_info_struct *drv);
183static void __devinit cciss_interrupt_mode(ctlr_info_t *);
184static int __devinit cciss_enter_simple_mode(struct ctlr_info *h);
185static void start_io(ctlr_info_t *h);
186static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
187			__u8 page_code, unsigned char scsi3addr[],
188			int cmd_type);
189static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
190	int attempt_retry);
191static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c);
192
193static int add_to_scan_list(struct ctlr_info *h);
194static int scan_thread(void *data);
195static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c);
196static void cciss_hba_release(struct device *dev);
197static void cciss_device_release(struct device *dev);
198static void cciss_free_gendisk(ctlr_info_t *h, int drv_index);
199static void cciss_free_drive_info(ctlr_info_t *h, int drv_index);
200static inline u32 next_command(ctlr_info_t *h);
201static int __devinit cciss_find_cfg_addrs(struct pci_dev *pdev,
202	void __iomem *vaddr, u32 *cfg_base_addr, u64 *cfg_base_addr_index,
203	u64 *cfg_offset);
204static int __devinit cciss_pci_find_memory_BAR(struct pci_dev *pdev,
205	unsigned long *memory_bar);
206static inline u32 cciss_tag_discard_error_bits(ctlr_info_t *h, u32 tag);
207static __devinit int write_driver_ver_to_cfgtable(
208	CfgTable_struct __iomem *cfgtable);
209
210/* performant mode helper functions */
211static void  calc_bucket_map(int *bucket, int num_buckets, int nsgs,
212				int *bucket_map);
213static void cciss_put_controller_into_performant_mode(ctlr_info_t *h);
214
215#ifdef CONFIG_PROC_FS
216static void cciss_procinit(ctlr_info_t *h);
217#else
218static void cciss_procinit(ctlr_info_t *h)
219{
220}
221#endif				/* CONFIG_PROC_FS */
222
223#ifdef CONFIG_COMPAT
224static int cciss_compat_ioctl(struct block_device *, fmode_t,
225			      unsigned, unsigned long);
226#endif
227
228static const struct block_device_operations cciss_fops = {
229	.owner = THIS_MODULE,
230	.open = cciss_unlocked_open,
231	.release = cciss_release,
232	.ioctl = do_ioctl,
233	.getgeo = cciss_getgeo,
234#ifdef CONFIG_COMPAT
235	.compat_ioctl = cciss_compat_ioctl,
236#endif
237	.revalidate_disk = cciss_revalidate,
238};
239
240/* set_performant_mode: Modify the tag for cciss performant
241 * set bit 0 for pull model, bits 3-1 for block fetch
242 * register number
243 */
244static void set_performant_mode(ctlr_info_t *h, CommandList_struct *c)
245{
246	if (likely(h->transMethod & CFGTBL_Trans_Performant))
247		c->busaddr |= 1 | (h->blockFetchTable[c->Header.SGList] << 1);
248}
249
250/*
251 * Enqueuing and dequeuing functions for cmdlists.
252 */
253static inline void addQ(struct list_head *list, CommandList_struct *c)
254{
255	list_add_tail(&c->list, list);
256}
257
258static inline void removeQ(CommandList_struct *c)
259{
260	/*
261	 * After kexec/dump some commands might still
262	 * be in flight, which the firmware will try
263	 * to complete. Resetting the firmware doesn't work
264	 * with old fw revisions, so we have to mark
265	 * them off as 'stale' to prevent the driver from
266	 * falling over.
267	 */
268	if (WARN_ON(list_empty(&c->list))) {
269		c->cmd_type = CMD_MSG_STALE;
270		return;
271	}
272
273	list_del_init(&c->list);
274}
275
276static void enqueue_cmd_and_start_io(ctlr_info_t *h,
277	CommandList_struct *c)
278{
279	unsigned long flags;
280	set_performant_mode(h, c);
281	spin_lock_irqsave(&h->lock, flags);
282	addQ(&h->reqQ, c);
283	h->Qdepth++;
284	if (h->Qdepth > h->maxQsinceinit)
285		h->maxQsinceinit = h->Qdepth;
286	start_io(h);
287	spin_unlock_irqrestore(&h->lock, flags);
288}
289
290static void cciss_free_sg_chain_blocks(SGDescriptor_struct **cmd_sg_list,
291	int nr_cmds)
292{
293	int i;
294
295	if (!cmd_sg_list)
296		return;
297	for (i = 0; i < nr_cmds; i++) {
298		kfree(cmd_sg_list[i]);
299		cmd_sg_list[i] = NULL;
300	}
301	kfree(cmd_sg_list);
302}
303
304static SGDescriptor_struct **cciss_allocate_sg_chain_blocks(
305	ctlr_info_t *h, int chainsize, int nr_cmds)
306{
307	int j;
308	SGDescriptor_struct **cmd_sg_list;
309
310	if (chainsize <= 0)
311		return NULL;
312
313	cmd_sg_list = kmalloc(sizeof(*cmd_sg_list) * nr_cmds, GFP_KERNEL);
314	if (!cmd_sg_list)
315		return NULL;
316
317	/* Build up chain blocks for each command */
318	for (j = 0; j < nr_cmds; j++) {
319		/* Need a block of chainsized s/g elements. */
320		cmd_sg_list[j] = kmalloc((chainsize *
321			sizeof(*cmd_sg_list[j])), GFP_KERNEL);
322		if (!cmd_sg_list[j]) {
323			dev_err(&h->pdev->dev, "Cannot get memory "
324				"for s/g chains.\n");
325			goto clean;
326		}
327	}
328	return cmd_sg_list;
329clean:
330	cciss_free_sg_chain_blocks(cmd_sg_list, nr_cmds);
331	return NULL;
332}
333
334static void cciss_unmap_sg_chain_block(ctlr_info_t *h, CommandList_struct *c)
335{
336	SGDescriptor_struct *chain_sg;
337	u64bit temp64;
338
339	if (c->Header.SGTotal <= h->max_cmd_sgentries)
340		return;
341
342	chain_sg = &c->SG[h->max_cmd_sgentries - 1];
343	temp64.val32.lower = chain_sg->Addr.lower;
344	temp64.val32.upper = chain_sg->Addr.upper;
345	pci_unmap_single(h->pdev, temp64.val, chain_sg->Len, PCI_DMA_TODEVICE);
346}
347
348static void cciss_map_sg_chain_block(ctlr_info_t *h, CommandList_struct *c,
349	SGDescriptor_struct *chain_block, int len)
350{
351	SGDescriptor_struct *chain_sg;
352	u64bit temp64;
353
354	chain_sg = &c->SG[h->max_cmd_sgentries - 1];
355	chain_sg->Ext = CCISS_SG_CHAIN;
356	chain_sg->Len = len;
357	temp64.val = pci_map_single(h->pdev, chain_block, len,
358				PCI_DMA_TODEVICE);
359	chain_sg->Addr.lower = temp64.val32.lower;
360	chain_sg->Addr.upper = temp64.val32.upper;
361}
362
363#include "cciss_scsi.c"		/* For SCSI tape support */
364
365static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
366	"UNKNOWN"
367};
368#define RAID_UNKNOWN (ARRAY_SIZE(raid_label)-1)
369
370#ifdef CONFIG_PROC_FS
371
372/*
373 * Report information about this controller.
374 */
375#define ENG_GIG 1000000000
376#define ENG_GIG_FACTOR (ENG_GIG/512)
377#define ENGAGE_SCSI	"engage scsi"
378
379static void cciss_seq_show_header(struct seq_file *seq)
380{
381	ctlr_info_t *h = seq->private;
382
383	seq_printf(seq, "%s: HP %s Controller\n"
384		"Board ID: 0x%08lx\n"
385		"Firmware Version: %c%c%c%c\n"
386		"IRQ: %d\n"
387		"Logical drives: %d\n"
388		"Current Q depth: %d\n"
389		"Current # commands on controller: %d\n"
390		"Max Q depth since init: %d\n"
391		"Max # commands on controller since init: %d\n"
392		"Max SG entries since init: %d\n",
393		h->devname,
394		h->product_name,
395		(unsigned long)h->board_id,
396		h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
397		h->firm_ver[3], (unsigned int)h->intr[h->intr_mode],
398		h->num_luns,
399		h->Qdepth, h->commands_outstanding,
400		h->maxQsinceinit, h->max_outstanding, h->maxSG);
401
402#ifdef CONFIG_CISS_SCSI_TAPE
403	cciss_seq_tape_report(seq, h);
404#endif /* CONFIG_CISS_SCSI_TAPE */
405}
406
407static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
408{
409	ctlr_info_t *h = seq->private;
410	unsigned long flags;
411
412	/* prevent displaying bogus info during configuration
413	 * or deconfiguration of a logical volume
414	 */
415	spin_lock_irqsave(&h->lock, flags);
416	if (h->busy_configuring) {
417		spin_unlock_irqrestore(&h->lock, flags);
418		return ERR_PTR(-EBUSY);
419	}
420	h->busy_configuring = 1;
421	spin_unlock_irqrestore(&h->lock, flags);
422
423	if (*pos == 0)
424		cciss_seq_show_header(seq);
425
426	return pos;
427}
428
429static int cciss_seq_show(struct seq_file *seq, void *v)
430{
431	sector_t vol_sz, vol_sz_frac;
432	ctlr_info_t *h = seq->private;
433	unsigned ctlr = h->ctlr;
434	loff_t *pos = v;
435	drive_info_struct *drv = h->drv[*pos];
436
437	if (*pos > h->highest_lun)
438		return 0;
439
440	if (drv == NULL) /* it's possible for h->drv[] to have holes. */
441		return 0;
442
443	if (drv->heads == 0)
444		return 0;
445
446	vol_sz = drv->nr_blocks;
447	vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
448	vol_sz_frac *= 100;
449	sector_div(vol_sz_frac, ENG_GIG_FACTOR);
450
451	if (drv->raid_level < 0 || drv->raid_level > RAID_UNKNOWN)
452		drv->raid_level = RAID_UNKNOWN;
453	seq_printf(seq, "cciss/c%dd%d:"
454			"\t%4u.%02uGB\tRAID %s\n",
455			ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
456			raid_label[drv->raid_level]);
457	return 0;
458}
459
460static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
461{
462	ctlr_info_t *h = seq->private;
463
464	if (*pos > h->highest_lun)
465		return NULL;
466	*pos += 1;
467
468	return pos;
469}
470
471static void cciss_seq_stop(struct seq_file *seq, void *v)
472{
473	ctlr_info_t *h = seq->private;
474
475	/* Only reset h->busy_configuring if we succeeded in setting
476	 * it during cciss_seq_start. */
477	if (v == ERR_PTR(-EBUSY))
478		return;
479
480	h->busy_configuring = 0;
481}
482
483static const struct seq_operations cciss_seq_ops = {
484	.start = cciss_seq_start,
485	.show  = cciss_seq_show,
486	.next  = cciss_seq_next,
487	.stop  = cciss_seq_stop,
488};
489
490static int cciss_seq_open(struct inode *inode, struct file *file)
491{
492	int ret = seq_open(file, &cciss_seq_ops);
493	struct seq_file *seq = file->private_data;
494
495	if (!ret)
496		seq->private = PDE(inode)->data;
497
498	return ret;
499}
500
501static ssize_t
502cciss_proc_write(struct file *file, const char __user *buf,
503		 size_t length, loff_t *ppos)
504{
505	int err;
506	char *buffer;
507
508#ifndef CONFIG_CISS_SCSI_TAPE
509	return -EINVAL;
510#endif
511
512	if (!buf || length > PAGE_SIZE - 1)
513		return -EINVAL;
514
515	buffer = (char *)__get_free_page(GFP_KERNEL);
516	if (!buffer)
517		return -ENOMEM;
518
519	err = -EFAULT;
520	if (copy_from_user(buffer, buf, length))
521		goto out;
522	buffer[length] = '\0';
523
524#ifdef CONFIG_CISS_SCSI_TAPE
525	if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
526		struct seq_file *seq = file->private_data;
527		ctlr_info_t *h = seq->private;
528
529		err = cciss_engage_scsi(h);
530		if (err == 0)
531			err = length;
532	} else
533#endif /* CONFIG_CISS_SCSI_TAPE */
534		err = -EINVAL;
535	/* might be nice to have "disengage" too, but it's not
536	   safely possible. (only 1 module use count, lock issues.) */
537
538out:
539	free_page((unsigned long)buffer);
540	return err;
541}
542
543static const struct file_operations cciss_proc_fops = {
544	.owner	 = THIS_MODULE,
545	.open    = cciss_seq_open,
546	.read    = seq_read,
547	.llseek  = seq_lseek,
548	.release = seq_release,
549	.write	 = cciss_proc_write,
550};
551
552static void __devinit cciss_procinit(ctlr_info_t *h)
553{
554	struct proc_dir_entry *pde;
555
556	if (proc_cciss == NULL)
557		proc_cciss = proc_mkdir("driver/cciss", NULL);
558	if (!proc_cciss)
559		return;
560	pde = proc_create_data(h->devname, S_IWUSR | S_IRUSR | S_IRGRP |
561					S_IROTH, proc_cciss,
562					&cciss_proc_fops, h);
563}
564#endif				/* CONFIG_PROC_FS */
565
566#define MAX_PRODUCT_NAME_LEN 19
567
568#define to_hba(n) container_of(n, struct ctlr_info, dev)
569#define to_drv(n) container_of(n, drive_info_struct, dev)
570
571/* List of controllers which cannot be hard reset on kexec with reset_devices */
572static u32 unresettable_controller[] = {
573	0x324a103C, /* Smart Array P712m */
574	0x324b103C, /* SmartArray P711m */
575	0x3223103C, /* Smart Array P800 */
576	0x3234103C, /* Smart Array P400 */
577	0x3235103C, /* Smart Array P400i */
578	0x3211103C, /* Smart Array E200i */
579	0x3212103C, /* Smart Array E200 */
580	0x3213103C, /* Smart Array E200i */
581	0x3214103C, /* Smart Array E200i */
582	0x3215103C, /* Smart Array E200i */
583	0x3237103C, /* Smart Array E500 */
584	0x323D103C, /* Smart Array P700m */
585	0x409C0E11, /* Smart Array 6400 */
586	0x409D0E11, /* Smart Array 6400 EM */
587};
588
589/* List of controllers which cannot even be soft reset */
590static u32 soft_unresettable_controller[] = {
591	0x409C0E11, /* Smart Array 6400 */
592	0x409D0E11, /* Smart Array 6400 EM */
593};
594
595static int ctlr_is_hard_resettable(u32 board_id)
596{
597	int i;
598
599	for (i = 0; i < ARRAY_SIZE(unresettable_controller); i++)
600		if (unresettable_controller[i] == board_id)
601			return 0;
602	return 1;
603}
604
605static int ctlr_is_soft_resettable(u32 board_id)
606{
607	int i;
608
609	for (i = 0; i < ARRAY_SIZE(soft_unresettable_controller); i++)
610		if (soft_unresettable_controller[i] == board_id)
611			return 0;
612	return 1;
613}
614
615static int ctlr_is_resettable(u32 board_id)
616{
617	return ctlr_is_hard_resettable(board_id) ||
618		ctlr_is_soft_resettable(board_id);
619}
620
621static ssize_t host_show_resettable(struct device *dev,
622				    struct device_attribute *attr,
623				    char *buf)
624{
625	struct ctlr_info *h = to_hba(dev);
626
627	return snprintf(buf, 20, "%d\n", ctlr_is_resettable(h->board_id));
628}
629static DEVICE_ATTR(resettable, S_IRUGO, host_show_resettable, NULL);
630
631static ssize_t host_store_rescan(struct device *dev,
632				 struct device_attribute *attr,
633				 const char *buf, size_t count)
634{
635	struct ctlr_info *h = to_hba(dev);
636
637	add_to_scan_list(h);
638	wake_up_process(cciss_scan_thread);
639	wait_for_completion_interruptible(&h->scan_wait);
640
641	return count;
642}
643static DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
644
645static ssize_t host_show_transport_mode(struct device *dev,
646				 struct device_attribute *attr,
647				 char *buf)
648{
649	struct ctlr_info *h = to_hba(dev);
650
651	return snprintf(buf, 20, "%s\n",
652		h->transMethod & CFGTBL_Trans_Performant ?
653			"performant" : "simple");
654}
655static DEVICE_ATTR(transport_mode, S_IRUGO, host_show_transport_mode, NULL);
656
657static ssize_t dev_show_unique_id(struct device *dev,
658				 struct device_attribute *attr,
659				 char *buf)
660{
661	drive_info_struct *drv = to_drv(dev);
662	struct ctlr_info *h = to_hba(drv->dev.parent);
663	__u8 sn[16];
664	unsigned long flags;
665	int ret = 0;
666
667	spin_lock_irqsave(&h->lock, flags);
668	if (h->busy_configuring)
669		ret = -EBUSY;
670	else
671		memcpy(sn, drv->serial_no, sizeof(sn));
672	spin_unlock_irqrestore(&h->lock, flags);
673
674	if (ret)
675		return ret;
676	else
677		return snprintf(buf, 16 * 2 + 2,
678				"%02X%02X%02X%02X%02X%02X%02X%02X"
679				"%02X%02X%02X%02X%02X%02X%02X%02X\n",
680				sn[0], sn[1], sn[2], sn[3],
681				sn[4], sn[5], sn[6], sn[7],
682				sn[8], sn[9], sn[10], sn[11],
683				sn[12], sn[13], sn[14], sn[15]);
684}
685static DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
686
687static ssize_t dev_show_vendor(struct device *dev,
688			       struct device_attribute *attr,
689			       char *buf)
690{
691	drive_info_struct *drv = to_drv(dev);
692	struct ctlr_info *h = to_hba(drv->dev.parent);
693	char vendor[VENDOR_LEN + 1];
694	unsigned long flags;
695	int ret = 0;
696
697	spin_lock_irqsave(&h->lock, flags);
698	if (h->busy_configuring)
699		ret = -EBUSY;
700	else
701		memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
702	spin_unlock_irqrestore(&h->lock, flags);
703
704	if (ret)
705		return ret;
706	else
707		return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
708}
709static DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
710
711static ssize_t dev_show_model(struct device *dev,
712			      struct device_attribute *attr,
713			      char *buf)
714{
715	drive_info_struct *drv = to_drv(dev);
716	struct ctlr_info *h = to_hba(drv->dev.parent);
717	char model[MODEL_LEN + 1];
718	unsigned long flags;
719	int ret = 0;
720
721	spin_lock_irqsave(&h->lock, flags);
722	if (h->busy_configuring)
723		ret = -EBUSY;
724	else
725		memcpy(model, drv->model, MODEL_LEN + 1);
726	spin_unlock_irqrestore(&h->lock, flags);
727
728	if (ret)
729		return ret;
730	else
731		return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
732}
733static DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
734
735static ssize_t dev_show_rev(struct device *dev,
736			    struct device_attribute *attr,
737			    char *buf)
738{
739	drive_info_struct *drv = to_drv(dev);
740	struct ctlr_info *h = to_hba(drv->dev.parent);
741	char rev[REV_LEN + 1];
742	unsigned long flags;
743	int ret = 0;
744
745	spin_lock_irqsave(&h->lock, flags);
746	if (h->busy_configuring)
747		ret = -EBUSY;
748	else
749		memcpy(rev, drv->rev, REV_LEN + 1);
750	spin_unlock_irqrestore(&h->lock, flags);
751
752	if (ret)
753		return ret;
754	else
755		return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
756}
757static DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
758
759static ssize_t cciss_show_lunid(struct device *dev,
760				struct device_attribute *attr, char *buf)
761{
762	drive_info_struct *drv = to_drv(dev);
763	struct ctlr_info *h = to_hba(drv->dev.parent);
764	unsigned long flags;
765	unsigned char lunid[8];
766
767	spin_lock_irqsave(&h->lock, flags);
768	if (h->busy_configuring) {
769		spin_unlock_irqrestore(&h->lock, flags);
770		return -EBUSY;
771	}
772	if (!drv->heads) {
773		spin_unlock_irqrestore(&h->lock, flags);
774		return -ENOTTY;
775	}
776	memcpy(lunid, drv->LunID, sizeof(lunid));
777	spin_unlock_irqrestore(&h->lock, flags);
778	return snprintf(buf, 20, "0x%02x%02x%02x%02x%02x%02x%02x%02x\n",
779		lunid[0], lunid[1], lunid[2], lunid[3],
780		lunid[4], lunid[5], lunid[6], lunid[7]);
781}
782static DEVICE_ATTR(lunid, S_IRUGO, cciss_show_lunid, NULL);
783
784static ssize_t cciss_show_raid_level(struct device *dev,
785				     struct device_attribute *attr, char *buf)
786{
787	drive_info_struct *drv = to_drv(dev);
788	struct ctlr_info *h = to_hba(drv->dev.parent);
789	int raid;
790	unsigned long flags;
791
792	spin_lock_irqsave(&h->lock, flags);
793	if (h->busy_configuring) {
794		spin_unlock_irqrestore(&h->lock, flags);
795		return -EBUSY;
796	}
797	raid = drv->raid_level;
798	spin_unlock_irqrestore(&h->lock, flags);
799	if (raid < 0 || raid > RAID_UNKNOWN)
800		raid = RAID_UNKNOWN;
801
802	return snprintf(buf, strlen(raid_label[raid]) + 7, "RAID %s\n",
803			raid_label[raid]);
804}
805static DEVICE_ATTR(raid_level, S_IRUGO, cciss_show_raid_level, NULL);
806
807static ssize_t cciss_show_usage_count(struct device *dev,
808				      struct device_attribute *attr, char *buf)
809{
810	drive_info_struct *drv = to_drv(dev);
811	struct ctlr_info *h = to_hba(drv->dev.parent);
812	unsigned long flags;
813	int count;
814
815	spin_lock_irqsave(&h->lock, flags);
816	if (h->busy_configuring) {
817		spin_unlock_irqrestore(&h->lock, flags);
818		return -EBUSY;
819	}
820	count = drv->usage_count;
821	spin_unlock_irqrestore(&h->lock, flags);
822	return snprintf(buf, 20, "%d\n", count);
823}
824static DEVICE_ATTR(usage_count, S_IRUGO, cciss_show_usage_count, NULL);
825
826static struct attribute *cciss_host_attrs[] = {
827	&dev_attr_rescan.attr,
828	&dev_attr_resettable.attr,
829	&dev_attr_transport_mode.attr,
830	NULL
831};
832
833static struct attribute_group cciss_host_attr_group = {
834	.attrs = cciss_host_attrs,
835};
836
837static const struct attribute_group *cciss_host_attr_groups[] = {
838	&cciss_host_attr_group,
839	NULL
840};
841
842static struct device_type cciss_host_type = {
843	.name		= "cciss_host",
844	.groups		= cciss_host_attr_groups,
845	.release	= cciss_hba_release,
846};
847
848static struct attribute *cciss_dev_attrs[] = {
849	&dev_attr_unique_id.attr,
850	&dev_attr_model.attr,
851	&dev_attr_vendor.attr,
852	&dev_attr_rev.attr,
853	&dev_attr_lunid.attr,
854	&dev_attr_raid_level.attr,
855	&dev_attr_usage_count.attr,
856	NULL
857};
858
859static struct attribute_group cciss_dev_attr_group = {
860	.attrs = cciss_dev_attrs,
861};
862
863static const struct attribute_group *cciss_dev_attr_groups[] = {
864	&cciss_dev_attr_group,
865	NULL
866};
867
868static struct device_type cciss_dev_type = {
869	.name		= "cciss_device",
870	.groups		= cciss_dev_attr_groups,
871	.release	= cciss_device_release,
872};
873
874static struct bus_type cciss_bus_type = {
875	.name		= "cciss",
876};
877
878/*
879 * cciss_hba_release is called when the reference count
880 * of h->dev goes to zero.
881 */
882static void cciss_hba_release(struct device *dev)
883{
884	/*
885	 * nothing to do, but need this to avoid a warning
886	 * about not having a release handler from lib/kref.c.
887	 */
888}
889
890/*
891 * Initialize sysfs entry for each controller.  This sets up and registers
892 * the 'cciss#' directory for each individual controller under
893 * /sys/bus/pci/devices/<dev>/.
894 */
895static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
896{
897	device_initialize(&h->dev);
898	h->dev.type = &cciss_host_type;
899	h->dev.bus = &cciss_bus_type;
900	dev_set_name(&h->dev, "%s", h->devname);
901	h->dev.parent = &h->pdev->dev;
902
903	return device_add(&h->dev);
904}
905
906/*
907 * Remove sysfs entries for an hba.
908 */
909static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
910{
911	device_del(&h->dev);
912	put_device(&h->dev); /* final put. */
913}
914
915/* cciss_device_release is called when the reference count
916 * of h->drv[x]dev goes to zero.
917 */
918static void cciss_device_release(struct device *dev)
919{
920	drive_info_struct *drv = to_drv(dev);
921	kfree(drv);
922}
923
924/*
925 * Initialize sysfs for each logical drive.  This sets up and registers
926 * the 'c#d#' directory for each individual logical drive under
927 * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
928 * /sys/block/cciss!c#d# to this entry.
929 */
930static long cciss_create_ld_sysfs_entry(struct ctlr_info *h,
931				       int drv_index)
932{
933	struct device *dev;
934
935	if (h->drv[drv_index]->device_initialized)
936		return 0;
937
938	dev = &h->drv[drv_index]->dev;
939	device_initialize(dev);
940	dev->type = &cciss_dev_type;
941	dev->bus = &cciss_bus_type;
942	dev_set_name(dev, "c%dd%d", h->ctlr, drv_index);
943	dev->parent = &h->dev;
944	h->drv[drv_index]->device_initialized = 1;
945	return device_add(dev);
946}
947
948/*
949 * Remove sysfs entries for a logical drive.
950 */
951static void cciss_destroy_ld_sysfs_entry(struct ctlr_info *h, int drv_index,
952	int ctlr_exiting)
953{
954	struct device *dev = &h->drv[drv_index]->dev;
955
956	/* special case for c*d0, we only destroy it on controller exit */
957	if (drv_index == 0 && !ctlr_exiting)
958		return;
959
960	device_del(dev);
961	put_device(dev); /* the "final" put. */
962	h->drv[drv_index] = NULL;
963}
964
965/*
966 * For operations that cannot sleep, a command block is allocated at init,
967 * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
968 * which ones are free or in use.
969 */
970static CommandList_struct *cmd_alloc(ctlr_info_t *h)
971{
972	CommandList_struct *c;
973	int i;
974	u64bit temp64;
975	dma_addr_t cmd_dma_handle, err_dma_handle;
976
977	do {
978		i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
979		if (i == h->nr_cmds)
980			return NULL;
981	} while (test_and_set_bit(i & (BITS_PER_LONG - 1),
982		  h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
983	c = h->cmd_pool + i;
984	memset(c, 0, sizeof(CommandList_struct));
985	cmd_dma_handle = h->cmd_pool_dhandle + i * sizeof(CommandList_struct);
986	c->err_info = h->errinfo_pool + i;
987	memset(c->err_info, 0, sizeof(ErrorInfo_struct));
988	err_dma_handle = h->errinfo_pool_dhandle
989	    + i * sizeof(ErrorInfo_struct);
990	h->nr_allocs++;
991
992	c->cmdindex = i;
993
994	INIT_LIST_HEAD(&c->list);
995	c->busaddr = (__u32) cmd_dma_handle;
996	temp64.val = (__u64) err_dma_handle;
997	c->ErrDesc.Addr.lower = temp64.val32.lower;
998	c->ErrDesc.Addr.upper = temp64.val32.upper;
999	c->ErrDesc.Len = sizeof(ErrorInfo_struct);
1000
1001	c->ctlr = h->ctlr;
1002	return c;
1003}
1004
1005/* allocate a command using pci_alloc_consistent, used for ioctls,
1006 * etc., not for the main i/o path.
1007 */
1008static CommandList_struct *cmd_special_alloc(ctlr_info_t *h)
1009{
1010	CommandList_struct *c;
1011	u64bit temp64;
1012	dma_addr_t cmd_dma_handle, err_dma_handle;
1013
1014	c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
1015		sizeof(CommandList_struct), &cmd_dma_handle);
1016	if (c == NULL)
1017		return NULL;
1018	memset(c, 0, sizeof(CommandList_struct));
1019
1020	c->cmdindex = -1;
1021
1022	c->err_info = (ErrorInfo_struct *)
1023	    pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
1024		    &err_dma_handle);
1025
1026	if (c->err_info == NULL) {
1027		pci_free_consistent(h->pdev,
1028			sizeof(CommandList_struct), c, cmd_dma_handle);
1029		return NULL;
1030	}
1031	memset(c->err_info, 0, sizeof(ErrorInfo_struct));
1032
1033	INIT_LIST_HEAD(&c->list);
1034	c->busaddr = (__u32) cmd_dma_handle;
1035	temp64.val = (__u64) err_dma_handle;
1036	c->ErrDesc.Addr.lower = temp64.val32.lower;
1037	c->ErrDesc.Addr.upper = temp64.val32.upper;
1038	c->ErrDesc.Len = sizeof(ErrorInfo_struct);
1039
1040	c->ctlr = h->ctlr;
1041	return c;
1042}
1043
1044static void cmd_free(ctlr_info_t *h, CommandList_struct *c)
1045{
1046	int i;
1047
1048	i = c - h->cmd_pool;
1049	clear_bit(i & (BITS_PER_LONG - 1),
1050		  h->cmd_pool_bits + (i / BITS_PER_LONG));
1051	h->nr_frees++;
1052}
1053
1054static void cmd_special_free(ctlr_info_t *h, CommandList_struct *c)
1055{
1056	u64bit temp64;
1057
1058	temp64.val32.lower = c->ErrDesc.Addr.lower;
1059	temp64.val32.upper = c->ErrDesc.Addr.upper;
1060	pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
1061			    c->err_info, (dma_addr_t) temp64.val);
1062	pci_free_consistent(h->pdev, sizeof(CommandList_struct), c,
1063		(dma_addr_t) cciss_tag_discard_error_bits(h, (u32) c->busaddr));
1064}
1065
1066static inline ctlr_info_t *get_host(struct gendisk *disk)
1067{
1068	return disk->queue->queuedata;
1069}
1070
1071static inline drive_info_struct *get_drv(struct gendisk *disk)
1072{
1073	return disk->private_data;
1074}
1075
1076/*
1077 * Open.  Make sure the device is really there.
1078 */
1079static int cciss_open(struct block_device *bdev, fmode_t mode)
1080{
1081	ctlr_info_t *h = get_host(bdev->bd_disk);
1082	drive_info_struct *drv = get_drv(bdev->bd_disk);
1083
1084	dev_dbg(&h->pdev->dev, "cciss_open %s\n", bdev->bd_disk->disk_name);
1085	if (drv->busy_configuring)
1086		return -EBUSY;
1087	/*
1088	 * Root is allowed to open raw volume zero even if it's not configured
1089	 * so array config can still work. Root is also allowed to open any
1090	 * volume that has a LUN ID, so it can issue IOCTL to reread the
1091	 * disk information.  I don't think I really like this
1092	 * but I'm already using way to many device nodes to claim another one
1093	 * for "raw controller".
1094	 */
1095	if (drv->heads == 0) {
1096		if (MINOR(bdev->bd_dev) != 0) {	/* not node 0? */
1097			/* if not node 0 make sure it is a partition = 0 */
1098			if (MINOR(bdev->bd_dev) & 0x0f) {
1099				return -ENXIO;
1100				/* if it is, make sure we have a LUN ID */
1101			} else if (memcmp(drv->LunID, CTLR_LUNID,
1102				sizeof(drv->LunID))) {
1103				return -ENXIO;
1104			}
1105		}
1106		if (!capable(CAP_SYS_ADMIN))
1107			return -EPERM;
1108	}
1109	drv->usage_count++;
1110	h->usage_count++;
1111	return 0;
1112}
1113
1114static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode)
1115{
1116	int ret;
1117
1118	mutex_lock(&cciss_mutex);
1119	ret = cciss_open(bdev, mode);
1120	mutex_unlock(&cciss_mutex);
1121
1122	return ret;
1123}
1124
1125/*
1126 * Close.  Sync first.
1127 */
1128static int cciss_release(struct gendisk *disk, fmode_t mode)
1129{
1130	ctlr_info_t *h;
1131	drive_info_struct *drv;
1132
1133	mutex_lock(&cciss_mutex);
1134	h = get_host(disk);
1135	drv = get_drv(disk);
1136	dev_dbg(&h->pdev->dev, "cciss_release %s\n", disk->disk_name);
1137	drv->usage_count--;
1138	h->usage_count--;
1139	mutex_unlock(&cciss_mutex);
1140	return 0;
1141}
1142
1143static int do_ioctl(struct block_device *bdev, fmode_t mode,
1144		    unsigned cmd, unsigned long arg)
1145{
1146	int ret;
1147	mutex_lock(&cciss_mutex);
1148	ret = cciss_ioctl(bdev, mode, cmd, arg);
1149	mutex_unlock(&cciss_mutex);
1150	return ret;
1151}
1152
1153#ifdef CONFIG_COMPAT
1154
1155static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1156				  unsigned cmd, unsigned long arg);
1157static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1158				      unsigned cmd, unsigned long arg);
1159
1160static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
1161			      unsigned cmd, unsigned long arg)
1162{
1163	switch (cmd) {
1164	case CCISS_GETPCIINFO:
1165	case CCISS_GETINTINFO:
1166	case CCISS_SETINTINFO:
1167	case CCISS_GETNODENAME:
1168	case CCISS_SETNODENAME:
1169	case CCISS_GETHEARTBEAT:
1170	case CCISS_GETBUSTYPES:
1171	case CCISS_GETFIRMVER:
1172	case CCISS_GETDRIVVER:
1173	case CCISS_REVALIDVOLS:
1174	case CCISS_DEREGDISK:
1175	case CCISS_REGNEWDISK:
1176	case CCISS_REGNEWD:
1177	case CCISS_RESCANDISK:
1178	case CCISS_GETLUNINFO:
1179		return do_ioctl(bdev, mode, cmd, arg);
1180
1181	case CCISS_PASSTHRU32:
1182		return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
1183	case CCISS_BIG_PASSTHRU32:
1184		return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
1185
1186	default:
1187		return -ENOIOCTLCMD;
1188	}
1189}
1190
1191static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1192				  unsigned cmd, unsigned long arg)
1193{
1194	IOCTL32_Command_struct __user *arg32 =
1195	    (IOCTL32_Command_struct __user *) arg;
1196	IOCTL_Command_struct arg64;
1197	IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
1198	int err;
1199	u32 cp;
1200
1201	err = 0;
1202	err |=
1203	    copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1204			   sizeof(arg64.LUN_info));
1205	err |=
1206	    copy_from_user(&arg64.Request, &arg32->Request,
1207			   sizeof(arg64.Request));
1208	err |=
1209	    copy_from_user(&arg64.error_info, &arg32->error_info,
1210			   sizeof(arg64.error_info));
1211	err |= get_user(arg64.buf_size, &arg32->buf_size);
1212	err |= get_user(cp, &arg32->buf);
1213	arg64.buf = compat_ptr(cp);
1214	err |= copy_to_user(p, &arg64, sizeof(arg64));
1215
1216	if (err)
1217		return -EFAULT;
1218
1219	err = do_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
1220	if (err)
1221		return err;
1222	err |=
1223	    copy_in_user(&arg32->error_info, &p->error_info,
1224			 sizeof(arg32->error_info));
1225	if (err)
1226		return -EFAULT;
1227	return err;
1228}
1229
1230static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1231				      unsigned cmd, unsigned long arg)
1232{
1233	BIG_IOCTL32_Command_struct __user *arg32 =
1234	    (BIG_IOCTL32_Command_struct __user *) arg;
1235	BIG_IOCTL_Command_struct arg64;
1236	BIG_IOCTL_Command_struct __user *p =
1237	    compat_alloc_user_space(sizeof(arg64));
1238	int err;
1239	u32 cp;
1240
1241	memset(&arg64, 0, sizeof(arg64));
1242	err = 0;
1243	err |=
1244	    copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1245			   sizeof(arg64.LUN_info));
1246	err |=
1247	    copy_from_user(&arg64.Request, &arg32->Request,
1248			   sizeof(arg64.Request));
1249	err |=
1250	    copy_from_user(&arg64.error_info, &arg32->error_info,
1251			   sizeof(arg64.error_info));
1252	err |= get_user(arg64.buf_size, &arg32->buf_size);
1253	err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1254	err |= get_user(cp, &arg32->buf);
1255	arg64.buf = compat_ptr(cp);
1256	err |= copy_to_user(p, &arg64, sizeof(arg64));
1257
1258	if (err)
1259		return -EFAULT;
1260
1261	err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1262	if (err)
1263		return err;
1264	err |=
1265	    copy_in_user(&arg32->error_info, &p->error_info,
1266			 sizeof(arg32->error_info));
1267	if (err)
1268		return -EFAULT;
1269	return err;
1270}
1271#endif
1272
1273static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1274{
1275	drive_info_struct *drv = get_drv(bdev->bd_disk);
1276
1277	if (!drv->cylinders)
1278		return -ENXIO;
1279
1280	geo->heads = drv->heads;
1281	geo->sectors = drv->sectors;
1282	geo->cylinders = drv->cylinders;
1283	return 0;
1284}
1285
1286static void check_ioctl_unit_attention(ctlr_info_t *h, CommandList_struct *c)
1287{
1288	if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1289			c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1290		(void)check_for_unit_attention(h, c);
1291}
1292
1293static int cciss_getpciinfo(ctlr_info_t *h, void __user *argp)
1294{
1295	cciss_pci_info_struct pciinfo;
1296
1297	if (!argp)
1298		return -EINVAL;
1299	pciinfo.domain = pci_domain_nr(h->pdev->bus);
1300	pciinfo.bus = h->pdev->bus->number;
1301	pciinfo.dev_fn = h->pdev->devfn;
1302	pciinfo.board_id = h->board_id;
1303	if (copy_to_user(argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1304		return -EFAULT;
1305	return 0;
1306}
1307
1308static int cciss_getintinfo(ctlr_info_t *h, void __user *argp)
1309{
1310	cciss_coalint_struct intinfo;
1311
1312	if (!argp)
1313		return -EINVAL;
1314	intinfo.delay = readl(&h->cfgtable->HostWrite.CoalIntDelay);
1315	intinfo.count = readl(&h->cfgtable->HostWrite.CoalIntCount);
1316	if (copy_to_user
1317	    (argp, &intinfo, sizeof(cciss_coalint_struct)))
1318		return -EFAULT;
1319	return 0;
1320}
1321
1322static int cciss_setintinfo(ctlr_info_t *h, void __user *argp)
1323{
1324	cciss_coalint_struct intinfo;
1325	unsigned long flags;
1326	int i;
1327
1328	if (!argp)
1329		return -EINVAL;
1330	if (!capable(CAP_SYS_ADMIN))
1331		return -EPERM;
1332	if (copy_from_user(&intinfo, argp, sizeof(intinfo)))
1333		return -EFAULT;
1334	if ((intinfo.delay == 0) && (intinfo.count == 0))
1335		return -EINVAL;
1336	spin_lock_irqsave(&h->lock, flags);
1337	/* Update the field, and then ring the doorbell */
1338	writel(intinfo.delay, &(h->cfgtable->HostWrite.CoalIntDelay));
1339	writel(intinfo.count, &(h->cfgtable->HostWrite.CoalIntCount));
1340	writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1341
1342	for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1343		if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
1344			break;
1345		udelay(1000); /* delay and try again */
1346	}
1347	spin_unlock_irqrestore(&h->lock, flags);
1348	if (i >= MAX_IOCTL_CONFIG_WAIT)
1349		return -EAGAIN;
1350	return 0;
1351}
1352
1353static int cciss_getnodename(ctlr_info_t *h, void __user *argp)
1354{
1355	NodeName_type NodeName;
1356	int i;
1357
1358	if (!argp)
1359		return -EINVAL;
1360	for (i = 0; i < 16; i++)
1361		NodeName[i] = readb(&h->cfgtable->ServerName[i]);
1362	if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1363		return -EFAULT;
1364	return 0;
1365}
1366
1367static int cciss_setnodename(ctlr_info_t *h, void __user *argp)
1368{
1369	NodeName_type NodeName;
1370	unsigned long flags;
1371	int i;
1372
1373	if (!argp)
1374		return -EINVAL;
1375	if (!capable(CAP_SYS_ADMIN))
1376		return -EPERM;
1377	if (copy_from_user(NodeName, argp, sizeof(NodeName_type)))
1378		return -EFAULT;
1379	spin_lock_irqsave(&h->lock, flags);
1380	/* Update the field, and then ring the doorbell */
1381	for (i = 0; i < 16; i++)
1382		writeb(NodeName[i], &h->cfgtable->ServerName[i]);
1383	writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1384	for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1385		if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
1386			break;
1387		udelay(1000); /* delay and try again */
1388	}
1389	spin_unlock_irqrestore(&h->lock, flags);
1390	if (i >= MAX_IOCTL_CONFIG_WAIT)
1391		return -EAGAIN;
1392	return 0;
1393}
1394
1395static int cciss_getheartbeat(ctlr_info_t *h, void __user *argp)
1396{
1397	Heartbeat_type heartbeat;
1398
1399	if (!argp)
1400		return -EINVAL;
1401	heartbeat = readl(&h->cfgtable->HeartBeat);
1402	if (copy_to_user(argp, &heartbeat, sizeof(Heartbeat_type)))
1403		return -EFAULT;
1404	return 0;
1405}
1406
1407static int cciss_getbustypes(ctlr_info_t *h, void __user *argp)
1408{
1409	BusTypes_type BusTypes;
1410
1411	if (!argp)
1412		return -EINVAL;
1413	BusTypes = readl(&h->cfgtable->BusTypes);
1414	if (copy_to_user(argp, &BusTypes, sizeof(BusTypes_type)))
1415		return -EFAULT;
1416	return 0;
1417}
1418
1419static int cciss_getfirmver(ctlr_info_t *h, void __user *argp)
1420{
1421	FirmwareVer_type firmware;
1422
1423	if (!argp)
1424		return -EINVAL;
1425	memcpy(firmware, h->firm_ver, 4);
1426
1427	if (copy_to_user
1428	    (argp, firmware, sizeof(FirmwareVer_type)))
1429		return -EFAULT;
1430	return 0;
1431}
1432
1433static int cciss_getdrivver(ctlr_info_t *h, void __user *argp)
1434{
1435	DriverVer_type DriverVer = DRIVER_VERSION;
1436
1437	if (!argp)
1438		return -EINVAL;
1439	if (copy_to_user(argp, &DriverVer, sizeof(DriverVer_type)))
1440		return -EFAULT;
1441	return 0;
1442}
1443
1444static int cciss_getluninfo(ctlr_info_t *h,
1445	struct gendisk *disk, void __user *argp)
1446{
1447	LogvolInfo_struct luninfo;
1448	drive_info_struct *drv = get_drv(disk);
1449
1450	if (!argp)
1451		return -EINVAL;
1452	memcpy(&luninfo.LunID, drv->LunID, sizeof(luninfo.LunID));
1453	luninfo.num_opens = drv->usage_count;
1454	luninfo.num_parts = 0;
1455	if (copy_to_user(argp, &luninfo, sizeof(LogvolInfo_struct)))
1456		return -EFAULT;
1457	return 0;
1458}
1459
1460static int cciss_passthru(ctlr_info_t *h, void __user *argp)
1461{
1462	IOCTL_Command_struct iocommand;
1463	CommandList_struct *c;
1464	char *buff = NULL;
1465	u64bit temp64;
1466	DECLARE_COMPLETION_ONSTACK(wait);
1467
1468	if (!argp)
1469		return -EINVAL;
1470
1471	if (!capable(CAP_SYS_RAWIO))
1472		return -EPERM;
1473
1474	if (copy_from_user
1475	    (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1476		return -EFAULT;
1477	if ((iocommand.buf_size < 1) &&
1478	    (iocommand.Request.Type.Direction != XFER_NONE)) {
1479		return -EINVAL;
1480	}
1481	if (iocommand.buf_size > 0) {
1482		buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1483		if (buff == NULL)
1484			return -EFAULT;
1485	}
1486	if (iocommand.Request.Type.Direction == XFER_WRITE) {
1487		/* Copy the data into the buffer we created */
1488		if (copy_from_user(buff, iocommand.buf, iocommand.buf_size)) {
1489			kfree(buff);
1490			return -EFAULT;
1491		}
1492	} else {
1493		memset(buff, 0, iocommand.buf_size);
1494	}
1495	c = cmd_special_alloc(h);
1496	if (!c) {
1497		kfree(buff);
1498		return -ENOMEM;
1499	}
1500	/* Fill in the command type */
1501	c->cmd_type = CMD_IOCTL_PEND;
1502	/* Fill in Command Header */
1503	c->Header.ReplyQueue = 0;   /* unused in simple mode */
1504	if (iocommand.buf_size > 0) { /* buffer to fill */
1505		c->Header.SGList = 1;
1506		c->Header.SGTotal = 1;
1507	} else { /* no buffers to fill */
1508		c->Header.SGList = 0;
1509		c->Header.SGTotal = 0;
1510	}
1511	c->Header.LUN = iocommand.LUN_info;
1512	/* use the kernel address the cmd block for tag */
1513	c->Header.Tag.lower = c->busaddr;
1514
1515	/* Fill in Request block */
1516	c->Request = iocommand.Request;
1517
1518	/* Fill in the scatter gather information */
1519	if (iocommand.buf_size > 0) {
1520		temp64.val = pci_map_single(h->pdev, buff,
1521			iocommand.buf_size, PCI_DMA_BIDIRECTIONAL);
1522		c->SG[0].Addr.lower = temp64.val32.lower;
1523		c->SG[0].Addr.upper = temp64.val32.upper;
1524		c->SG[0].Len = iocommand.buf_size;
1525		c->SG[0].Ext = 0;  /* we are not chaining */
1526	}
1527	c->waiting = &wait;
1528
1529	enqueue_cmd_and_start_io(h, c);
1530	wait_for_completion(&wait);
1531
1532	/* unlock the buffers from DMA */
1533	temp64.val32.lower = c->SG[0].Addr.lower;
1534	temp64.val32.upper = c->SG[0].Addr.upper;
1535	pci_unmap_single(h->pdev, (dma_addr_t) temp64.val, iocommand.buf_size,
1536			 PCI_DMA_BIDIRECTIONAL);
1537	check_ioctl_unit_attention(h, c);
1538
1539	/* Copy the error information out */
1540	iocommand.error_info = *(c->err_info);
1541	if (copy_to_user(argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1542		kfree(buff);
1543		cmd_special_free(h, c);
1544		return -EFAULT;
1545	}
1546
1547	if (iocommand.Request.Type.Direction == XFER_READ) {
1548		/* Copy the data out of the buffer we created */
1549		if (copy_to_user(iocommand.buf, buff, iocommand.buf_size)) {
1550			kfree(buff);
1551			cmd_special_free(h, c);
1552			return -EFAULT;
1553		}
1554	}
1555	kfree(buff);
1556	cmd_special_free(h, c);
1557	return 0;
1558}
1559
1560static int cciss_bigpassthru(ctlr_info_t *h, void __user *argp)
1561{
1562	BIG_IOCTL_Command_struct *ioc;
1563	CommandList_struct *c;
1564	unsigned char **buff = NULL;
1565	int *buff_size = NULL;
1566	u64bit temp64;
1567	BYTE sg_used = 0;
1568	int status = 0;
1569	int i;
1570	DECLARE_COMPLETION_ONSTACK(wait);
1571	__u32 left;
1572	__u32 sz;
1573	BYTE __user *data_ptr;
1574
1575	if (!argp)
1576		return -EINVAL;
1577	if (!capable(CAP_SYS_RAWIO))
1578		return -EPERM;
1579	ioc = kmalloc(sizeof(*ioc), GFP_KERNEL);
1580	if (!ioc) {
1581		status = -ENOMEM;
1582		goto cleanup1;
1583	}
1584	if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1585		status = -EFAULT;
1586		goto cleanup1;
1587	}
1588	if ((ioc->buf_size < 1) &&
1589	    (ioc->Request.Type.Direction != XFER_NONE)) {
1590		status = -EINVAL;
1591		goto cleanup1;
1592	}
1593	/* Check kmalloc limits  using all SGs */
1594	if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1595		status = -EINVAL;
1596		goto cleanup1;
1597	}
1598	if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1599		status = -EINVAL;
1600		goto cleanup1;
1601	}
1602	buff = kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1603	if (!buff) {
1604		status = -ENOMEM;
1605		goto cleanup1;
1606	}
1607	buff_size = kmalloc(MAXSGENTRIES * sizeof(int), GFP_KERNEL);
1608	if (!buff_size) {
1609		status = -ENOMEM;
1610		goto cleanup1;
1611	}
1612	left = ioc->buf_size;
1613	data_ptr = ioc->buf;
1614	while (left) {
1615		sz = (left > ioc->malloc_size) ? ioc->malloc_size : left;
1616		buff_size[sg_used] = sz;
1617		buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1618		if (buff[sg_used] == NULL) {
1619			status = -ENOMEM;
1620			goto cleanup1;
1621		}
1622		if (ioc->Request.Type.Direction == XFER_WRITE) {
1623			if (copy_from_user(buff[sg_used], data_ptr, sz)) {
1624				status = -EFAULT;
1625				goto cleanup1;
1626			}
1627		} else {
1628			memset(buff[sg_used], 0, sz);
1629		}
1630		left -= sz;
1631		data_ptr += sz;
1632		sg_used++;
1633	}
1634	c = cmd_special_alloc(h);
1635	if (!c) {
1636		status = -ENOMEM;
1637		goto cleanup1;
1638	}
1639	c->cmd_type = CMD_IOCTL_PEND;
1640	c->Header.ReplyQueue = 0;
1641	c->Header.SGList = sg_used;
1642	c->Header.SGTotal = sg_used;
1643	c->Header.LUN = ioc->LUN_info;
1644	c->Header.Tag.lower = c->busaddr;
1645
1646	c->Request = ioc->Request;
1647	for (i = 0; i < sg_used; i++) {
1648		temp64.val = pci_map_single(h->pdev, buff[i], buff_size[i],
1649				    PCI_DMA_BIDIRECTIONAL);
1650		c->SG[i].Addr.lower = temp64.val32.lower;
1651		c->SG[i].Addr.upper = temp64.val32.upper;
1652		c->SG[i].Len = buff_size[i];
1653		c->SG[i].Ext = 0;	/* we are not chaining */
1654	}
1655	c->waiting = &wait;
1656	enqueue_cmd_and_start_io(h, c);
1657	wait_for_completion(&wait);
1658	/* unlock the buffers from DMA */
1659	for (i = 0; i < sg_used; i++) {
1660		temp64.val32.lower = c->SG[i].Addr.lower;
1661		temp64.val32.upper = c->SG[i].Addr.upper;
1662		pci_unmap_single(h->pdev,
1663			(dma_addr_t) temp64.val, buff_size[i],
1664			PCI_DMA_BIDIRECTIONAL);
1665	}
1666	check_ioctl_unit_attention(h, c);
1667	/* Copy the error information out */
1668	ioc->error_info = *(c->err_info);
1669	if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1670		cmd_special_free(h, c);
1671		status = -EFAULT;
1672		goto cleanup1;
1673	}
1674	if (ioc->Request.Type.Direction == XFER_READ) {
1675		/* Copy the data out of the buffer we created */
1676		BYTE __user *ptr = ioc->buf;
1677		for (i = 0; i < sg_used; i++) {
1678			if (copy_to_user(ptr, buff[i], buff_size[i])) {
1679				cmd_special_free(h, c);
1680				status = -EFAULT;
1681				goto cleanup1;
1682			}
1683			ptr += buff_size[i];
1684		}
1685	}
1686	cmd_special_free(h, c);
1687	status = 0;
1688cleanup1:
1689	if (buff) {
1690		for (i = 0; i < sg_used; i++)
1691			kfree(buff[i]);
1692		kfree(buff);
1693	}
1694	kfree(buff_size);
1695	kfree(ioc);
1696	return status;
1697}
1698
1699static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1700	unsigned int cmd, unsigned long arg)
1701{
1702	struct gendisk *disk = bdev->bd_disk;
1703	ctlr_info_t *h = get_host(disk);
1704	void __user *argp = (void __user *)arg;
1705
1706	dev_dbg(&h->pdev->dev, "cciss_ioctl: Called with cmd=%x %lx\n",
1707		cmd, arg);
1708	switch (cmd) {
1709	case CCISS_GETPCIINFO:
1710		return cciss_getpciinfo(h, argp);
1711	case CCISS_GETINTINFO:
1712		return cciss_getintinfo(h, argp);
1713	case CCISS_SETINTINFO:
1714		return cciss_setintinfo(h, argp);
1715	case CCISS_GETNODENAME:
1716		return cciss_getnodename(h, argp);
1717	case CCISS_SETNODENAME:
1718		return cciss_setnodename(h, argp);
1719	case CCISS_GETHEARTBEAT:
1720		return cciss_getheartbeat(h, argp);
1721	case CCISS_GETBUSTYPES:
1722		return cciss_getbustypes(h, argp);
1723	case CCISS_GETFIRMVER:
1724		return cciss_getfirmver(h, argp);
1725	case CCISS_GETDRIVVER:
1726		return cciss_getdrivver(h, argp);
1727	case CCISS_DEREGDISK:
1728	case CCISS_REGNEWD:
1729	case CCISS_REVALIDVOLS:
1730		return rebuild_lun_table(h, 0, 1);
1731	case CCISS_GETLUNINFO:
1732		return cciss_getluninfo(h, disk, argp);
1733	case CCISS_PASSTHRU:
1734		return cciss_passthru(h, argp);
1735	case CCISS_BIG_PASSTHRU:
1736		return cciss_bigpassthru(h, argp);
1737
1738	/* scsi_cmd_blk_ioctl handles these, below, though some are not */
1739	/* very meaningful for cciss.  SG_IO is the main one people want. */
1740
1741	case SG_GET_VERSION_NUM:
1742	case SG_SET_TIMEOUT:
1743	case SG_GET_TIMEOUT:
1744	case SG_GET_RESERVED_SIZE:
1745	case SG_SET_RESERVED_SIZE:
1746	case SG_EMULATED_HOST:
1747	case SG_IO:
1748	case SCSI_IOCTL_SEND_COMMAND:
1749		return scsi_cmd_blk_ioctl(bdev, mode, cmd, argp);
1750
1751	/* scsi_cmd_blk_ioctl would normally handle these, below, but */
1752	/* they aren't a good fit for cciss, as CD-ROMs are */
1753	/* not supported, and we don't have any bus/target/lun */
1754	/* which we present to the kernel. */
1755
1756	case CDROM_SEND_PACKET:
1757	case CDROMCLOSETRAY:
1758	case CDROMEJECT:
1759	case SCSI_IOCTL_GET_IDLUN:
1760	case SCSI_IOCTL_GET_BUS_NUMBER:
1761	default:
1762		return -ENOTTY;
1763	}
1764}
1765
1766static void cciss_check_queues(ctlr_info_t *h)
1767{
1768	int start_queue = h->next_to_run;
1769	int i;
1770
1771	/* check to see if we have maxed out the number of commands that can
1772	 * be placed on the queue.  If so then exit.  We do this check here
1773	 * in case the interrupt we serviced was from an ioctl and did not
1774	 * free any new commands.
1775	 */
1776	if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1777		return;
1778
1779	/* We have room on the queue for more commands.  Now we need to queue
1780	 * them up.  We will also keep track of the next queue to run so
1781	 * that every queue gets a chance to be started first.
1782	 */
1783	for (i = 0; i < h->highest_lun + 1; i++) {
1784		int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1785		/* make sure the disk has been added and the drive is real
1786		 * because this can be called from the middle of init_one.
1787		 */
1788		if (!h->drv[curr_queue])
1789			continue;
1790		if (!(h->drv[curr_queue]->queue) ||
1791			!(h->drv[curr_queue]->heads))
1792			continue;
1793		blk_start_queue(h->gendisk[curr_queue]->queue);
1794
1795		/* check to see if we have maxed out the number of commands
1796		 * that can be placed on the queue.
1797		 */
1798		if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1799			if (curr_queue == start_queue) {
1800				h->next_to_run =
1801				    (start_queue + 1) % (h->highest_lun + 1);
1802				break;
1803			} else {
1804				h->next_to_run = curr_queue;
1805				break;
1806			}
1807		}
1808	}
1809}
1810
1811static void cciss_softirq_done(struct request *rq)
1812{
1813	CommandList_struct *c = rq->completion_data;
1814	ctlr_info_t *h = hba[c->ctlr];
1815	SGDescriptor_struct *curr_sg = c->SG;
1816	u64bit temp64;
1817	unsigned long flags;
1818	int i, ddir;
1819	int sg_index = 0;
1820
1821	if (c->Request.Type.Direction == XFER_READ)
1822		ddir = PCI_DMA_FROMDEVICE;
1823	else
1824		ddir = PCI_DMA_TODEVICE;
1825
1826	/* command did not need to be retried */
1827	/* unmap the DMA mapping for all the scatter gather elements */
1828	for (i = 0; i < c->Header.SGList; i++) {
1829		if (curr_sg[sg_index].Ext == CCISS_SG_CHAIN) {
1830			cciss_unmap_sg_chain_block(h, c);
1831			/* Point to the next block */
1832			curr_sg = h->cmd_sg_list[c->cmdindex];
1833			sg_index = 0;
1834		}
1835		temp64.val32.lower = curr_sg[sg_index].Addr.lower;
1836		temp64.val32.upper = curr_sg[sg_index].Addr.upper;
1837		pci_unmap_page(h->pdev, temp64.val, curr_sg[sg_index].Len,
1838				ddir);
1839		++sg_index;
1840	}
1841
1842	dev_dbg(&h->pdev->dev, "Done with %p\n", rq);
1843
1844	/* set the residual count for pc requests */
1845	if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
1846		rq->resid_len = c->err_info->ResidualCnt;
1847
1848	blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1849
1850	spin_lock_irqsave(&h->lock, flags);
1851	cmd_free(h, c);
1852	cciss_check_queues(h);
1853	spin_unlock_irqrestore(&h->lock, flags);
1854}
1855
1856static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1857	unsigned char scsi3addr[], uint32_t log_unit)
1858{
1859	memcpy(scsi3addr, h->drv[log_unit]->LunID,
1860		sizeof(h->drv[log_unit]->LunID));
1861}
1862
1863/* This function gets the SCSI vendor, model, and revision of a logical drive
1864 * via the inquiry page 0.  Model, vendor, and rev are set to empty strings if
1865 * they cannot be read.
1866 */
1867static void cciss_get_device_descr(ctlr_info_t *h, int logvol,
1868				   char *vendor, char *model, char *rev)
1869{
1870	int rc;
1871	InquiryData_struct *inq_buf;
1872	unsigned char scsi3addr[8];
1873
1874	*vendor = '\0';
1875	*model = '\0';
1876	*rev = '\0';
1877
1878	inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1879	if (!inq_buf)
1880		return;
1881
1882	log_unit_to_scsi3addr(h, scsi3addr, logvol);
1883	rc = sendcmd_withirq(h, CISS_INQUIRY, inq_buf, sizeof(*inq_buf), 0,
1884			scsi3addr, TYPE_CMD);
1885	if (rc == IO_OK) {
1886		memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1887		vendor[VENDOR_LEN] = '\0';
1888		memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1889		model[MODEL_LEN] = '\0';
1890		memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1891		rev[REV_LEN] = '\0';
1892	}
1893
1894	kfree(inq_buf);
1895	return;
1896}
1897
1898/* This function gets the serial number of a logical drive via
1899 * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1900 * number cannot be had, for whatever reason, 16 bytes of 0xff
1901 * are returned instead.
1902 */
1903static void cciss_get_serial_no(ctlr_info_t *h, int logvol,
1904				unsigned char *serial_no, int buflen)
1905{
1906#define PAGE_83_INQ_BYTES 64
1907	int rc;
1908	unsigned char *buf;
1909	unsigned char scsi3addr[8];
1910
1911	if (buflen > 16)
1912		buflen = 16;
1913	memset(serial_no, 0xff, buflen);
1914	buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1915	if (!buf)
1916		return;
1917	memset(serial_no, 0, buflen);
1918	log_unit_to_scsi3addr(h, scsi3addr, logvol);
1919	rc = sendcmd_withirq(h, CISS_INQUIRY, buf,
1920		PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1921	if (rc == IO_OK)
1922		memcpy(serial_no, &buf[8], buflen);
1923	kfree(buf);
1924	return;
1925}
1926
1927/*
1928 * cciss_add_disk sets up the block device queue for a logical drive
1929 */
1930static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1931				int drv_index)
1932{
1933	disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1934	if (!disk->queue)
1935		goto init_queue_failure;
1936	sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1937	disk->major = h->major;
1938	disk->first_minor = drv_index << NWD_SHIFT;
1939	disk->fops = &cciss_fops;
1940	if (cciss_create_ld_sysfs_entry(h, drv_index))
1941		goto cleanup_queue;
1942	disk->private_data = h->drv[drv_index];
1943	disk->driverfs_dev = &h->drv[drv_index]->dev;
1944
1945	/* Set up queue information */
1946	blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1947
1948	/* This is a hardware imposed limit. */
1949	blk_queue_max_segments(disk->queue, h->maxsgentries);
1950
1951	blk_queue_max_hw_sectors(disk->queue, h->cciss_max_sectors);
1952
1953	blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1954
1955	disk->queue->queuedata = h;
1956
1957	blk_queue_logical_block_size(disk->queue,
1958				     h->drv[drv_index]->block_size);
1959
1960	/* Make sure all queue data is written out before */
1961	/* setting h->drv[drv_index]->queue, as setting this */
1962	/* allows the interrupt handler to start the queue */
1963	wmb();
1964	h->drv[drv_index]->queue = disk->queue;
1965	add_disk(disk);
1966	return 0;
1967
1968cleanup_queue:
1969	blk_cleanup_queue(disk->queue);
1970	disk->queue = NULL;
1971init_queue_failure:
1972	return -1;
1973}
1974
1975/* This function will check the usage_count of the drive to be updated/added.
1976 * If the usage_count is zero and it is a heretofore unknown drive, or,
1977 * the drive's capacity, geometry, or serial number has changed,
1978 * then the drive information will be updated and the disk will be
1979 * re-registered with the kernel.  If these conditions don't hold,
1980 * then it will be left alone for the next reboot.  The exception to this
1981 * is disk 0 which will always be left registered with the kernel since it
1982 * is also the controller node.  Any changes to disk 0 will show up on
1983 * the next reboot.
1984 */
1985static void cciss_update_drive_info(ctlr_info_t *h, int drv_index,
1986	int first_time, int via_ioctl)
1987{
1988	struct gendisk *disk;
1989	InquiryData_struct *inq_buff = NULL;
1990	unsigned int block_size;
1991	sector_t total_size;
1992	unsigned long flags = 0;
1993	int ret = 0;
1994	drive_info_struct *drvinfo;
1995
1996	/* Get information about the disk and modify the driver structure */
1997	inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1998	drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
1999	if (inq_buff == NULL || drvinfo == NULL)
2000		goto mem_msg;
2001
2002	/* testing to see if 16-byte CDBs are already being used */
2003	if (h->cciss_read == CCISS_READ_16) {
2004		cciss_read_capacity_16(h, drv_index,
2005			&total_size, &block_size);
2006
2007	} else {
2008		cciss_read_capacity(h, drv_index, &total_size, &block_size);
2009		/* if read_capacity returns all F's this volume is >2TB */
2010		/* in size so we switch to 16-byte CDB's for all */
2011		/* read/write ops */
2012		if (total_size == 0xFFFFFFFFULL) {
2013			cciss_read_capacity_16(h, drv_index,
2014			&total_size, &block_size);
2015			h->cciss_read = CCISS_READ_16;
2016			h->cciss_write = CCISS_WRITE_16;
2017		} else {
2018			h->cciss_read = CCISS_READ_10;
2019			h->cciss_write = CCISS_WRITE_10;
2020		}
2021	}
2022
2023	cciss_geometry_inquiry(h, drv_index, total_size, block_size,
2024			       inq_buff, drvinfo);
2025	drvinfo->block_size = block_size;
2026	drvinfo->nr_blocks = total_size + 1;
2027
2028	cciss_get_device_descr(h, drv_index, drvinfo->vendor,
2029				drvinfo->model, drvinfo->rev);
2030	cciss_get_serial_no(h, drv_index, drvinfo->serial_no,
2031			sizeof(drvinfo->serial_no));
2032	/* Save the lunid in case we deregister the disk, below. */
2033	memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
2034		sizeof(drvinfo->LunID));
2035
2036	/* Is it the same disk we already know, and nothing's changed? */
2037	if (h->drv[drv_index]->raid_level != -1 &&
2038		((memcmp(drvinfo->serial_no,
2039				h->drv[drv_index]->serial_no, 16) == 0) &&
2040		drvinfo->block_size == h->drv[drv_index]->block_size &&
2041		drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
2042		drvinfo->heads == h->drv[drv_index]->heads &&
2043		drvinfo->sectors == h->drv[drv_index]->sectors &&
2044		drvinfo->cylinders == h->drv[drv_index]->cylinders))
2045			/* The disk is unchanged, nothing to update */
2046			goto freeret;
2047
2048	/* If we get here it's not the same disk, or something's changed,
2049	 * so we need to * deregister it, and re-register it, if it's not
2050	 * in use.
2051	 * If the disk already exists then deregister it before proceeding
2052	 * (unless it's the first disk (for the controller node).
2053	 */
2054	if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
2055		dev_warn(&h->pdev->dev, "disk %d has changed.\n", drv_index);
2056		spin_lock_irqsave(&h->lock, flags);
2057		h->drv[drv_index]->busy_configuring = 1;
2058		spin_unlock_irqrestore(&h->lock, flags);
2059
2060		/* deregister_disk sets h->drv[drv_index]->queue = NULL
2061		 * which keeps the interrupt handler from starting
2062		 * the queue.
2063		 */
2064		ret = deregister_disk(h, drv_index, 0, via_ioctl);
2065	}
2066
2067	/* If the disk is in use return */
2068	if (ret)
2069		goto freeret;
2070
2071	/* Save the new information from cciss_geometry_inquiry
2072	 * and serial number inquiry.  If the disk was deregistered
2073	 * above, then h->drv[drv_index] will be NULL.
2074	 */
2075	if (h->drv[drv_index] == NULL) {
2076		drvinfo->device_initialized = 0;
2077		h->drv[drv_index] = drvinfo;
2078		drvinfo = NULL; /* so it won't be freed below. */
2079	} else {
2080		/* special case for cxd0 */
2081		h->drv[drv_index]->block_size = drvinfo->block_size;
2082		h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
2083		h->drv[drv_index]->heads = drvinfo->heads;
2084		h->drv[drv_index]->sectors = drvinfo->sectors;
2085		h->drv[drv_index]->cylinders = drvinfo->cylinders;
2086		h->drv[drv_index]->raid_level = drvinfo->raid_level;
2087		memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
2088		memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
2089			VENDOR_LEN + 1);
2090		memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
2091		memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
2092	}
2093
2094	++h->num_luns;
2095	disk = h->gendisk[drv_index];
2096	set_capacity(disk, h->drv[drv_index]->nr_blocks);
2097
2098	/* If it's not disk 0 (drv_index != 0)
2099	 * or if it was disk 0, but there was previously
2100	 * no actual corresponding configured logical drive
2101	 * (raid_leve == -1) then we want to update the
2102	 * logical drive's information.
2103	 */
2104	if (drv_index || first_time) {
2105		if (cciss_add_disk(h, disk, drv_index) != 0) {
2106			cciss_free_gendisk(h, drv_index);
2107			cciss_free_drive_info(h, drv_index);
2108			dev_warn(&h->pdev->dev, "could not update disk %d\n",
2109				drv_index);
2110			--h->num_luns;
2111		}
2112	}
2113
2114freeret:
2115	kfree(inq_buff);
2116	kfree(drvinfo);
2117	return;
2118mem_msg:
2119	dev_err(&h->pdev->dev, "out of memory\n");
2120	goto freeret;
2121}
2122
2123/* This function will find the first index of the controllers drive array
2124 * that has a null drv pointer and allocate the drive info struct and
2125 * will return that index   This is where new drives will be added.
2126 * If the index to be returned is greater than the highest_lun index for
2127 * the controller then highest_lun is set * to this new index.
2128 * If there are no available indexes or if tha allocation fails, then -1
2129 * is returned.  * "controller_node" is used to know if this is a real
2130 * logical drive, or just the controller node, which determines if this
2131 * counts towards highest_lun.
2132 */
2133static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
2134{
2135	int i;
2136	drive_info_struct *drv;
2137
2138	/* Search for an empty slot for our drive info */
2139	for (i = 0; i < CISS_MAX_LUN; i++) {
2140
2141		/* if not cxd0 case, and it's occupied, skip it. */
2142		if (h->drv[i] && i != 0)
2143			continue;
2144		/*
2145		 * If it's cxd0 case, and drv is alloc'ed already, and a
2146		 * disk is configured there, skip it.
2147		 */
2148		if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2149			continue;
2150
2151		/*
2152		 * We've found an empty slot.  Update highest_lun
2153		 * provided this isn't just the fake cxd0 controller node.
2154		 */
2155		if (i > h->highest_lun && !controller_node)
2156			h->highest_lun = i;
2157
2158		/* If adding a real disk at cxd0, and it's already alloc'ed */
2159		if (i == 0 && h->drv[i] != NULL)
2160			return i;
2161
2162		/*
2163		 * Found an empty slot, not already alloc'ed.  Allocate it.
2164		 * Mark it with raid_level == -1, so we know it's new later on.
2165		 */
2166		drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2167		if (!drv)
2168			return -1;
2169		drv->raid_level = -1; /* so we know it's new */
2170		h->drv[i] = drv;
2171		return i;
2172	}
2173	return -1;
2174}
2175
2176static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2177{
2178	kfree(h->drv[drv_index]);
2179	h->drv[drv_index] = NULL;
2180}
2181
2182static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2183{
2184	put_disk(h->gendisk[drv_index]);
2185	h->gendisk[drv_index] = NULL;
2186}
2187
2188/* cciss_add_gendisk finds a free hba[]->drv structure
2189 * and allocates a gendisk if needed, and sets the lunid
2190 * in the drvinfo structure.   It returns the index into
2191 * the ->drv[] array, or -1 if none are free.
2192 * is_controller_node indicates whether highest_lun should
2193 * count this disk, or if it's only being added to provide
2194 * a means to talk to the controller in case no logical
2195 * drives have yet been configured.
2196 */
2197static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2198	int controller_node)
2199{
2200	int drv_index;
2201
2202	drv_index = cciss_alloc_drive_info(h, controller_node);
2203	if (drv_index == -1)
2204		return -1;
2205
2206	/*Check if the gendisk needs to be allocated */
2207	if (!h->gendisk[drv_index]) {
2208		h->gendisk[drv_index] =
2209			alloc_disk(1 << NWD_SHIFT);
2210		if (!h->gendisk[drv_index]) {
2211			dev_err(&h->pdev->dev,
2212				"could not allocate a new disk %d\n",
2213				drv_index);
2214			goto err_free_drive_info;
2215		}
2216	}
2217	memcpy(h->drv[drv_index]->LunID, lunid,
2218		sizeof(h->drv[drv_index]->LunID));
2219	if (cciss_create_ld_sysfs_entry(h, drv_index))
2220		goto err_free_disk;
2221	/* Don't need to mark this busy because nobody */
2222	/* else knows about this disk yet to contend */
2223	/* for access to it. */
2224	h->drv[drv_index]->busy_configuring = 0;
2225	wmb();
2226	return drv_index;
2227
2228err_free_disk:
2229	cciss_free_gendisk(h, drv_index);
2230err_free_drive_info:
2231	cciss_free_drive_info(h, drv_index);
2232	return -1;
2233}
2234
2235/* This is for the special case of a controller which
2236 * has no logical drives.  In this case, we still need
2237 * to register a disk so the controller can be accessed
2238 * by the Array Config Utility.
2239 */
2240static void cciss_add_controller_node(ctlr_info_t *h)
2241{
2242	struct gendisk *disk;
2243	int drv_index;
2244
2245	if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2246		return;
2247
2248	drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2249	if (drv_index == -1)
2250		goto error;
2251	h->drv[drv_index]->block_size = 512;
2252	h->drv[drv_index]->nr_blocks = 0;
2253	h->drv[drv_index]->heads = 0;
2254	h->drv[drv_index]->sectors = 0;
2255	h->drv[drv_index]->cylinders = 0;
2256	h->drv[drv_index]->raid_level = -1;
2257	memset(h->drv[drv_index]->serial_no, 0, 16);
2258	disk = h->gendisk[drv_index];
2259	if (cciss_add_disk(h, disk, drv_index) == 0)
2260		return;
2261	cciss_free_gendisk(h, drv_index);
2262	cciss_free_drive_info(h, drv_index);
2263error:
2264	dev_warn(&h->pdev->dev, "could not add disk 0.\n");
2265	return;
2266}
2267
2268/* This function will add and remove logical drives from the Logical
2269 * drive array of the controller and maintain persistency of ordering
2270 * so that mount points are preserved until the next reboot.  This allows
2271 * for the removal of logical drives in the middle of the drive array
2272 * without a re-ordering of those drives.
2273 * INPUT
2274 * h		= The controller to perform the operations on
2275 */
2276static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2277	int via_ioctl)
2278{
2279	int num_luns;
2280	ReportLunData_struct *ld_buff = NULL;
2281	int return_code;
2282	int listlength = 0;
2283	int i;
2284	int drv_found;
2285	int drv_index = 0;
2286	unsigned char lunid[8] = CTLR_LUNID;
2287	unsigned long flags;
2288
2289	if (!capable(CAP_SYS_RAWIO))
2290		return -EPERM;
2291
2292	/* Set busy_configuring flag for this operation */
2293	spin_lock_irqsave(&h->lock, flags);
2294	if (h->busy_configuring) {
2295		spin_unlock_irqrestore(&h->lock, flags);
2296		return -EBUSY;
2297	}
2298	h->busy_configuring = 1;
2299	spin_unlock_irqrestore(&h->lock, flags);
2300
2301	ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2302	if (ld_buff == NULL)
2303		goto mem_msg;
2304
2305	return_code = sendcmd_withirq(h, CISS_REPORT_LOG, ld_buff,
2306				      sizeof(ReportLunData_struct),
2307				      0, CTLR_LUNID, TYPE_CMD);
2308
2309	if (return_code == IO_OK)
2310		listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2311	else {	/* reading number of logical volumes failed */
2312		dev_warn(&h->pdev->dev,
2313			"report logical volume command failed\n");
2314		listlength = 0;
2315		goto freeret;
2316	}
2317
2318	num_luns = listlength / 8;	/* 8 bytes per entry */
2319	if (num_luns > CISS_MAX_LUN) {
2320		num_luns = CISS_MAX_LUN;
2321		dev_warn(&h->pdev->dev, "more luns configured"
2322		       " on controller than can be handled by"
2323		       " this driver.\n");
2324	}
2325
2326	if (num_luns == 0)
2327		cciss_add_controller_node(h);
2328
2329	/* Compare controller drive array to driver's drive array
2330	 * to see if any drives are missing on the controller due
2331	 * to action of Array Config Utility (user deletes drive)
2332	 * and deregister logical drives which have disappeared.
2333	 */
2334	for (i = 0; i <= h->highest_lun; i++) {
2335		int j;
2336		drv_found = 0;
2337
2338		/* skip holes in the array from already deleted drives */
2339		if (h->drv[i] == NULL)
2340			continue;
2341
2342		for (j = 0; j < num_luns; j++) {
2343			memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2344			if (memcmp(h->drv[i]->LunID, lunid,
2345				sizeof(lunid)) == 0) {
2346				drv_found = 1;
2347				break;
2348			}
2349		}
2350		if (!drv_found) {
2351			/* Deregister it from the OS, it's gone. */
2352			spin_lock_irqsave(&h->lock, flags);
2353			h->drv[i]->busy_configuring = 1;
2354			spin_unlock_irqrestore(&h->lock, flags);
2355			return_code = deregister_disk(h, i, 1, via_ioctl);
2356			if (h->drv[i] != NULL)
2357				h->drv[i]->busy_configuring = 0;
2358		}
2359	}
2360
2361	/* Compare controller drive array to driver's drive array.
2362	 * Check for updates in the drive information and any new drives
2363	 * on the controller due to ACU adding logical drives, or changing
2364	 * a logical drive's size, etc.  Reregister any new/changed drives
2365	 */
2366	for (i = 0; i < num_luns; i++) {
2367		int j;
2368
2369		drv_found = 0;
2370
2371		memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2372		/* Find if the LUN is already in the drive array
2373		 * of the driver.  If so then update its info
2374		 * if not in use.  If it does not exist then find
2375		 * the first free index and add it.
2376		 */
2377		for (j = 0; j <= h->highest_lun; j++) {
2378			if (h->drv[j] != NULL &&
2379				memcmp(h->drv[j]->LunID, lunid,
2380					sizeof(h->drv[j]->LunID)) == 0) {
2381				drv_index = j;
2382				drv_found = 1;
2383				break;
2384			}
2385		}
2386
2387		/* check if the drive was found already in the array */
2388		if (!drv_found) {
2389			drv_index = cciss_add_gendisk(h, lunid, 0);
2390			if (drv_index == -1)
2391				goto freeret;
2392		}
2393		cciss_update_drive_info(h, drv_index, first_time, via_ioctl);
2394	}		/* end for */
2395
2396freeret:
2397	kfree(ld_buff);
2398	h->busy_configuring = 0;
2399	/* We return -1 here to tell the ACU that we have registered/updated
2400	 * all of the drives that we can and to keep it from calling us
2401	 * additional times.
2402	 */
2403	return -1;
2404mem_msg:
2405	dev_err(&h->pdev->dev, "out of memory\n");
2406	h->busy_configuring = 0;
2407	goto freeret;
2408}
2409
2410static void cciss_clear_drive_info(drive_info_struct *drive_info)
2411{
2412	/* zero out the disk size info */
2413	drive_info->nr_blocks = 0;
2414	drive_info->block_size = 0;
2415	drive_info->heads = 0;
2416	drive_info->sectors = 0;
2417	drive_info->cylinders = 0;
2418	drive_info->raid_level = -1;
2419	memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2420	memset(drive_info->model, 0, sizeof(drive_info->model));
2421	memset(drive_info->rev, 0, sizeof(drive_info->rev));
2422	memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2423	/*
2424	 * don't clear the LUNID though, we need to remember which
2425	 * one this one is.
2426	 */
2427}
2428
2429/* This function will deregister the disk and it's queue from the
2430 * kernel.  It must be called with the controller lock held and the
2431 * drv structures busy_configuring flag set.  It's parameters are:
2432 *
2433 * disk = This is the disk to be deregistered
2434 * drv  = This is the drive_info_struct associated with the disk to be
2435 *        deregistered.  It contains information about the disk used
2436 *        by the driver.
2437 * clear_all = This flag determines whether or not the disk information
2438 *             is going to be completely cleared out and the highest_lun
2439 *             reset.  Sometimes we want to clear out information about
2440 *             the disk in preparation for re-adding it.  In this case
2441 *             the highest_lun should be left unchanged and the LunID
2442 *             should not be cleared.
2443 * via_ioctl
2444 *    This indicates whether we've reached this path via ioctl.
2445 *    This affects the maximum usage count allowed for c0d0 to be messed with.
2446 *    If this path is reached via ioctl(), then the max_usage_count will
2447 *    be 1, as the process calling ioctl() has got to have the device open.
2448 *    If we get here via sysfs, then the max usage count will be zero.
2449*/
2450static int deregister_disk(ctlr_info_t *h, int drv_index,
2451			   int clear_all, int via_ioctl)
2452{
2453	int i;
2454	struct gendisk *disk;
2455	drive_info_struct *drv;
2456	int recalculate_highest_lun;
2457
2458	if (!capable(CAP_SYS_RAWIO))
2459		return -EPERM;
2460
2461	drv = h->drv[drv_index];
2462	disk = h->gendisk[drv_index];
2463
2464	/* make sure logical volume is NOT is use */
2465	if (clear_all || (h->gendisk[0] == disk)) {
2466		if (drv->usage_count > via_ioctl)
2467			return -EBUSY;
2468	} else if (drv->usage_count > 0)
2469		return -EBUSY;
2470
2471	recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2472
2473	/* invalidate the devices and deregister the disk.  If it is disk
2474	 * zero do not deregister it but just zero out it's values.  This
2475	 * allows us to delete disk zero but keep the controller registered.
2476	 */
2477	if (h->gendisk[0] != disk) {
2478		struct request_queue *q = disk->queue;
2479		if (disk->flags & GENHD_FL_UP) {
2480			cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2481			del_gendisk(disk);
2482		}
2483		if (q)
2484			blk_cleanup_queue(q);
2485		/* If clear_all is set then we are deleting the logical
2486		 * drive, not just refreshing its info.  For drives
2487		 * other than disk 0 we will call put_disk.  We do not
2488		 * do this for disk 0 as we need it to be able to
2489		 * configure the controller.
2490		 */
2491		if (clear_all){
2492			/* This isn't pretty, but we need to find the
2493			 * disk in our array and NULL our the pointer.
2494			 * This is so that we will call alloc_disk if
2495			 * this index is used again later.
2496			 */
2497			for (i=0; i < CISS_MAX_LUN; i++){
2498				if (h->gendisk[i] == disk) {
2499					h->gendisk[i] = NULL;
2500					break;
2501				}
2502			}
2503			put_disk(disk);
2504		}
2505	} else {
2506		set_capacity(disk, 0);
2507		cciss_clear_drive_info(drv);
2508	}
2509
2510	--h->num_luns;
2511
2512	/* if it was the last disk, find the new hightest lun */
2513	if (clear_all && recalculate_highest_lun) {
2514		int newhighest = -1;
2515		for (i = 0; i <= h->highest_lun; i++) {
2516			/* if the disk has size > 0, it is available */
2517			if (h->drv[i] && h->drv[i]->heads)
2518				newhighest = i;
2519		}
2520		h->highest_lun = newhighest;
2521	}
2522	return 0;
2523}
2524
2525static int fill_cmd(ctlr_info_t *h, CommandList_struct *c, __u8 cmd, void *buff,
2526		size_t size, __u8 page_code, unsigned char *scsi3addr,
2527		int cmd_type)
2528{
2529	u64bit buff_dma_handle;
2530	int status = IO_OK;
2531
2532	c->cmd_type = CMD_IOCTL_PEND;
2533	c->Header.ReplyQueue = 0;
2534	if (buff != NULL) {
2535		c->Header.SGList = 1;
2536		c->Header.SGTotal = 1;
2537	} else {
2538		c->Header.SGList = 0;
2539		c->Header.SGTotal = 0;
2540	}
2541	c->Header.Tag.lower = c->busaddr;
2542	memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2543
2544	c->Request.Type.Type = cmd_type;
2545	if (cmd_type == TYPE_CMD) {
2546		switch (cmd) {
2547		case CISS_INQUIRY:
2548			/* are we trying to read a vital product page */
2549			if (page_code != 0) {
2550				c->Request.CDB[1] = 0x01;
2551				c->Request.CDB[2] = page_code;
2552			}
2553			c->Request.CDBLen = 6;
2554			c->Request.Type.Attribute = ATTR_SIMPLE;
2555			c->Request.Type.Direction = XFER_READ;
2556			c->Request.Timeout = 0;
2557			c->Request.CDB[0] = CISS_INQUIRY;
2558			c->Request.CDB[4] = size & 0xFF;
2559			break;
2560		case CISS_REPORT_LOG:
2561		case CISS_REPORT_PHYS:
2562			/* Talking to controller so It's a physical command
2563			   mode = 00 target = 0.  Nothing to write.
2564			 */
2565			c->Request.CDBLen = 12;
2566			c->Request.Type.Attribute = ATTR_SIMPLE;
2567			c->Request.Type.Direction = XFER_READ;
2568			c->Request.Timeout = 0;
2569			c->Request.CDB[0] = cmd;
2570			c->Request.CDB[6] = (size >> 24) & 0xFF; /* MSB */
2571			c->Request.CDB[7] = (size >> 16) & 0xFF;
2572			c->Request.CDB[8] = (size >> 8) & 0xFF;
2573			c->Request.CDB[9] = size & 0xFF;
2574			break;
2575
2576		case CCISS_READ_CAPACITY:
2577			c->Request.CDBLen = 10;
2578			c->Request.Type.Attribute = ATTR_SIMPLE;
2579			c->Request.Type.Direction = XFER_READ;
2580			c->Request.Timeout = 0;
2581			c->Request.CDB[0] = cmd;
2582			break;
2583		case CCISS_READ_CAPACITY_16:
2584			c->Request.CDBLen = 16;
2585			c->Request.Type.Attribute = ATTR_SIMPLE;
2586			c->Request.Type.Direction = XFER_READ;
2587			c->Request.Timeout = 0;
2588			c->Request.CDB[0] = cmd;
2589			c->Request.CDB[1] = 0x10;
2590			c->Request.CDB[10] = (size >> 24) & 0xFF;
2591			c->Request.CDB[11] = (size >> 16) & 0xFF;
2592			c->Request.CDB[12] = (size >> 8) & 0xFF;
2593			c->Request.CDB[13] = size & 0xFF;
2594			c->Request.Timeout = 0;
2595			c->Request.CDB[0] = cmd;
2596			break;
2597		case CCISS_CACHE_FLUSH:
2598			c->Request.CDBLen = 12;
2599			c->Request.Type.Attribute = ATTR_SIMPLE;
2600			c->Request.Type.Direction = XFER_WRITE;
2601			c->Request.Timeout = 0;
2602			c->Request.CDB[0] = BMIC_WRITE;
2603			c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2604			c->Request.CDB[7] = (size >> 8) & 0xFF;
2605			c->Request.CDB[8] = size & 0xFF;
2606			break;
2607		case TEST_UNIT_READY:
2608			c->Request.CDBLen = 6;
2609			c->Request.Type.Attribute = ATTR_SIMPLE;
2610			c->Request.Type.Direction = XFER_NONE;
2611			c->Request.Timeout = 0;
2612			break;
2613		default:
2614			dev_warn(&h->pdev->dev, "Unknown Command 0x%c\n", cmd);
2615			return IO_ERROR;
2616		}
2617	} else if (cmd_type == TYPE_MSG) {
2618		switch (cmd) {
2619		case CCISS_ABORT_MSG:
2620			c->Request.CDBLen = 12;
2621			c->Request.Type.Attribute = ATTR_SIMPLE;
2622			c->Request.Type.Direction = XFER_WRITE;
2623			c->Request.Timeout = 0;
2624			c->Request.CDB[0] = cmd;	/* abort */
2625			c->Request.CDB[1] = 0;	/* abort a command */
2626			/* buff contains the tag of the command to abort */
2627			memcpy(&c->Request.CDB[4], buff, 8);
2628			break;
2629		case CCISS_RESET_MSG:
2630			c->Request.CDBLen = 16;
2631			c->Request.Type.Attribute = ATTR_SIMPLE;
2632			c->Request.Type.Direction = XFER_NONE;
2633			c->Request.Timeout = 0;
2634			memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2635			c->Request.CDB[0] = cmd;	/* reset */
2636			c->Request.CDB[1] = CCISS_RESET_TYPE_TARGET;
2637			break;
2638		case CCISS_NOOP_MSG:
2639			c->Request.CDBLen = 1;
2640			c->Request.Type.Attribute = ATTR_SIMPLE;
2641			c->Request.Type.Direction = XFER_WRITE;
2642			c->Request.Timeout = 0;
2643			c->Request.CDB[0] = cmd;
2644			break;
2645		default:
2646			dev_warn(&h->pdev->dev,
2647				"unknown message type %d\n", cmd);
2648			return IO_ERROR;
2649		}
2650	} else {
2651		dev_warn(&h->pdev->dev, "unknown command type %d\n", cmd_type);
2652		return IO_ERROR;
2653	}
2654	/* Fill in the scatter gather information */
2655	if (size > 0) {
2656		buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2657							     buff, size,
2658							     PCI_DMA_BIDIRECTIONAL);
2659		c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2660		c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2661		c->SG[0].Len = size;
2662		c->SG[0].Ext = 0;	/* we are not chaining */
2663	}
2664	return status;
2665}
2666
2667static int __devinit cciss_send_reset(ctlr_info_t *h, unsigned char *scsi3addr,
2668	u8 reset_type)
2669{
2670	CommandList_struct *c;
2671	int return_status;
2672
2673	c = cmd_alloc(h);
2674	if (!c)
2675		return -ENOMEM;
2676	return_status = fill_cmd(h, c, CCISS_RESET_MSG, NULL, 0, 0,
2677		CTLR_LUNID, TYPE_MSG);
2678	c->Request.CDB[1] = reset_type; /* fill_cmd defaults to target reset */
2679	if (return_status != IO_OK) {
2680		cmd_special_free(h, c);
2681		return return_status;
2682	}
2683	c->waiting = NULL;
2684	enqueue_cmd_and_start_io(h, c);
2685	/* Don't wait for completion, the reset won't complete.  Don't free
2686	 * the command either.  This is the last command we will send before
2687	 * re-initializing everything, so it doesn't matter and won't leak.
2688	 */
2689	return 0;
2690}
2691
2692static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2693{
2694	switch (c->err_info->ScsiStatus) {
2695	case SAM_STAT_GOOD:
2696		return IO_OK;
2697	case SAM_STAT_CHECK_CONDITION:
2698		switch (0xf & c->err_info->SenseInfo[2]) {
2699		case 0: return IO_OK; /* no sense */
2700		case 1: return IO_OK; /* recovered error */
2701		default:
2702			if (check_for_unit_attention(h, c))
2703				return IO_NEEDS_RETRY;
2704			dev_warn(&h->pdev->dev, "cmd 0x%02x "
2705				"check condition, sense key = 0x%02x\n",
2706				c->Request.CDB[0], c->err_info->SenseInfo[2]);
2707		}
2708		break;
2709	default:
2710		dev_warn(&h->pdev->dev, "cmd 0x%02x"
2711			"scsi status = 0x%02x\n",
2712			c->Request.CDB[0], c->err_info->ScsiStatus);
2713		break;
2714	}
2715	return IO_ERROR;
2716}
2717
2718static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2719{
2720	int return_status = IO_OK;
2721
2722	if (c->err_info->CommandStatus == CMD_SUCCESS)
2723		return IO_OK;
2724
2725	switch (c->err_info->CommandStatus) {
2726	case CMD_TARGET_STATUS:
2727		return_status = check_target_status(h, c);
2728		break;
2729	case CMD_DATA_UNDERRUN:
2730	case CMD_DATA_OVERRUN:
2731		/* expected for inquiry and report lun commands */
2732		break;
2733	case CMD_INVALID:
2734		dev_warn(&h->pdev->dev, "cmd 0x%02x is "
2735		       "reported invalid\n", c->Request.CDB[0]);
2736		return_status = IO_ERROR;
2737		break;
2738	case CMD_PROTOCOL_ERR:
2739		dev_warn(&h->pdev->dev, "cmd 0x%02x has "
2740		       "protocol error\n", c->Request.CDB[0]);
2741		return_status = IO_ERROR;
2742		break;
2743	case CMD_HARDWARE_ERR:
2744		dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2745		       " hardware error\n", c->Request.CDB[0]);
2746		return_status = IO_ERROR;
2747		break;
2748	case CMD_CONNECTION_LOST:
2749		dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2750		       "connection lost\n", c->Request.CDB[0]);
2751		return_status = IO_ERROR;
2752		break;
2753	case CMD_ABORTED:
2754		dev_warn(&h->pdev->dev, "cmd 0x%02x was "
2755		       "aborted\n", c->Request.CDB[0]);
2756		return_status = IO_ERROR;
2757		break;
2758	case CMD_ABORT_FAILED:
2759		dev_warn(&h->pdev->dev, "cmd 0x%02x reports "
2760		       "abort failed\n", c->Request.CDB[0]);
2761		return_status = IO_ERROR;
2762		break;
2763	case CMD_UNSOLICITED_ABORT:
2764		dev_warn(&h->pdev->dev, "unsolicited abort 0x%02x\n",
2765			c->Request.CDB[0]);
2766		return_status = IO_NEEDS_RETRY;
2767		break;
2768	case CMD_UNABORTABLE:
2769		dev_warn(&h->pdev->dev, "cmd unabortable\n");
2770		return_status = IO_ERROR;
2771		break;
2772	default:
2773		dev_warn(&h->pdev->dev, "cmd 0x%02x returned "
2774		       "unknown status %x\n", c->Request.CDB[0],
2775		       c->err_info->CommandStatus);
2776		return_status = IO_ERROR;
2777	}
2778	return return_status;
2779}
2780
2781static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2782	int attempt_retry)
2783{
2784	DECLARE_COMPLETION_ONSTACK(wait);
2785	u64bit buff_dma_handle;
2786	int return_status = IO_OK;
2787
2788resend_cmd2:
2789	c->waiting = &wait;
2790	enqueue_cmd_and_start_io(h, c);
2791
2792	wait_for_completion(&wait);
2793
2794	if (c->err_info->CommandStatus == 0 || !attempt_retry)
2795		goto command_done;
2796
2797	return_status = process_sendcmd_error(h, c);
2798
2799	if (return_status == IO_NEEDS_RETRY &&
2800		c->retry_count < MAX_CMD_RETRIES) {
2801		dev_warn(&h->pdev->dev, "retrying 0x%02x\n",
2802			c->Request.CDB[0]);
2803		c->retry_count++;
2804		/* erase the old error information */
2805		memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2806		return_status = IO_OK;
2807		INIT_COMPLETION(wait);
2808		goto resend_cmd2;
2809	}
2810
2811command_done:
2812	/* unlock the buffers from DMA */
2813	buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2814	buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2815	pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2816			 c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2817	return return_status;
2818}
2819
2820static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
2821			   __u8 page_code, unsigned char scsi3addr[],
2822			int cmd_type)
2823{
2824	CommandList_struct *c;
2825	int return_status;
2826
2827	c = cmd_special_alloc(h);
2828	if (!c)
2829		return -ENOMEM;
2830	return_status = fill_cmd(h, c, cmd, buff, size, page_code,
2831		scsi3addr, cmd_type);
2832	if (return_status == IO_OK)
2833		return_status = sendcmd_withirq_core(h, c, 1);
2834
2835	cmd_special_free(h, c);
2836	return return_status;
2837}
2838
2839static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
2840				   sector_t total_size,
2841				   unsigned int block_size,
2842				   InquiryData_struct *inq_buff,
2843				   drive_info_struct *drv)
2844{
2845	int return_code;
2846	unsigned long t;
2847	unsigned char scsi3addr[8];
2848
2849	memset(inq_buff, 0, sizeof(InquiryData_struct));
2850	log_unit_to_scsi3addr(h, scsi3addr, logvol);
2851	return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
2852			sizeof(*inq_buff), 0xC1, scsi3addr, TYPE_CMD);
2853	if (return_code == IO_OK) {
2854		if (inq_buff->data_byte[8] == 0xFF) {
2855			dev_warn(&h->pdev->dev,
2856			       "reading geometry failed, volume "
2857			       "does not support reading geometry\n");
2858			drv->heads = 255;
2859			drv->sectors = 32;	/* Sectors per track */
2860			drv->cylinders = total_size + 1;
2861			drv->raid_level = RAID_UNKNOWN;
2862		} else {
2863			drv->heads = inq_buff->data_byte[6];
2864			drv->sectors = inq_buff->data_byte[7];
2865			drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2866			drv->cylinders += inq_buff->data_byte[5];
2867			drv->raid_level = inq_buff->data_byte[8];
2868		}
2869		drv->block_size = block_size;
2870		drv->nr_blocks = total_size + 1;
2871		t = drv->heads * drv->sectors;
2872		if (t > 1) {
2873			sector_t real_size = total_size + 1;
2874			unsigned long rem = sector_div(real_size, t);
2875			if (rem)
2876				real_size++;
2877			drv->cylinders = real_size;
2878		}
2879	} else {		/* Get geometry failed */
2880		dev_warn(&h->pdev->dev, "reading geometry failed\n");
2881	}
2882}
2883
2884static void
2885cciss_read_capacity(ctlr_info_t *h, int logvol, sector_t *total_size,
2886		    unsigned int *block_size)
2887{
2888	ReadCapdata_struct *buf;
2889	int return_code;
2890	unsigned char scsi3addr[8];
2891
2892	buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2893	if (!buf) {
2894		dev_warn(&h->pdev->dev, "out of memory\n");
2895		return;
2896	}
2897
2898	log_unit_to_scsi3addr(h, scsi3addr, logvol);
2899	return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY, buf,
2900		sizeof(ReadCapdata_struct), 0, scsi3addr, TYPE_CMD);
2901	if (return_code == IO_OK) {
2902		*total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2903		*block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2904	} else {		/* read capacity command failed */
2905		dev_warn(&h->pdev->dev, "read capacity failed\n");
2906		*total_size = 0;
2907		*block_size = BLOCK_SIZE;
2908	}
2909	kfree(buf);
2910}
2911
2912static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
2913	sector_t *total_size, unsigned int *block_size)
2914{
2915	ReadCapdata_struct_16 *buf;
2916	int return_code;
2917	unsigned char scsi3addr[8];
2918
2919	buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2920	if (!buf) {
2921		dev_warn(&h->pdev->dev, "out of memory\n");
2922		return;
2923	}
2924
2925	log_unit_to_scsi3addr(h, scsi3addr, logvol);
2926	return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY_16,
2927		buf, sizeof(ReadCapdata_struct_16),
2928			0, scsi3addr, TYPE_CMD);
2929	if (return_code == IO_OK) {
2930		*total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2931		*block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2932	} else {		/* read capacity command failed */
2933		dev_warn(&h->pdev->dev, "read capacity failed\n");
2934		*total_size = 0;
2935		*block_size = BLOCK_SIZE;
2936	}
2937	dev_info(&h->pdev->dev, "      blocks= %llu block_size= %d\n",
2938	       (unsigned long long)*total_size+1, *block_size);
2939	kfree(buf);
2940}
2941
2942static int cciss_revalidate(struct gendisk *disk)
2943{
2944	ctlr_info_t *h = get_host(disk);
2945	drive_info_struct *drv = get_drv(disk);
2946	int logvol;
2947	int FOUND = 0;
2948	unsigned int block_size;
2949	sector_t total_size;
2950	InquiryData_struct *inq_buff = NULL;
2951
2952	for (logvol = 0; logvol <= h->highest_lun; logvol++) {
2953		if (!h->drv[logvol])
2954			continue;
2955		if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2956			sizeof(drv->LunID)) == 0) {
2957			FOUND = 1;
2958			break;
2959		}
2960	}
2961
2962	if (!FOUND)
2963		return 1;
2964
2965	inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2966	if (inq_buff == NULL) {
2967		dev_warn(&h->pdev->dev, "out of memory\n");
2968		return 1;
2969	}
2970	if (h->cciss_read == CCISS_READ_10) {
2971		cciss_read_capacity(h, logvol,
2972					&total_size, &block_size);
2973	} else {
2974		cciss_read_capacity_16(h, logvol,
2975					&total_size, &block_size);
2976	}
2977	cciss_geometry_inquiry(h, logvol, total_size, block_size,
2978			       inq_buff, drv);
2979
2980	blk_queue_logical_block_size(drv->queue, drv->block_size);
2981	set_capacity(disk, drv->nr_blocks);
2982
2983	kfree(inq_buff);
2984	return 0;
2985}
2986
2987/*
2988 * Map (physical) PCI mem into (virtual) kernel space
2989 */
2990static void __iomem *remap_pci_mem(ulong base, ulong size)
2991{
2992	ulong page_base = ((ulong) base) & PAGE_MASK;
2993	ulong page_offs = ((ulong) base) - page_base;
2994	void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2995
2996	return page_remapped ? (page_remapped + page_offs) : NULL;
2997}
2998
2999/*
3000 * Takes jobs of the Q and sends them to the hardware, then puts it on
3001 * the Q to wait for completion.
3002 */
3003static void start_io(ctlr_info_t *h)
3004{
3005	CommandList_struct *c;
3006
3007	while (!list_empty(&h->reqQ)) {
3008		c = list_entry(h->reqQ.next, CommandList_struct, list);
3009		/* can't do anything if fifo is full */
3010		if ((h->access.fifo_full(h))) {
3011			dev_warn(&h->pdev->dev, "fifo full\n");
3012			break;
3013		}
3014
3015		/* Get the first entry from the Request Q */
3016		removeQ(c);
3017		h->Qdepth--;
3018
3019		/* Tell the controller execute command */
3020		h->access.submit_command(h, c);
3021
3022		/* Put job onto the completed Q */
3023		addQ(&h->cmpQ, c);
3024	}
3025}
3026
3027/* Assumes that h->lock is held. */
3028/* Zeros out the error record and then resends the command back */
3029/* to the controller */
3030static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
3031{
3032	/* erase the old error information */
3033	memset(c->err_info, 0, sizeof(ErrorInfo_struct));
3034
3035	/* add it to software queue and then send it to the controller */
3036	addQ(&h->reqQ, c);
3037	h->Qdepth++;
3038	if (h->Qdepth > h->maxQsinceinit)
3039		h->maxQsinceinit = h->Qdepth;
3040
3041	start_io(h);
3042}
3043
3044static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
3045	unsigned int msg_byte, unsigned int host_byte,
3046	unsigned int driver_byte)
3047{
3048	/* inverse of macros in scsi.h */
3049	return (scsi_status_byte & 0xff) |
3050		((msg_byte & 0xff) << 8) |
3051		((host_byte & 0xff) << 16) |
3052		((driver_byte & 0xff) << 24);
3053}
3054
3055static inline int evaluate_target_status(ctlr_info_t *h,
3056			CommandList_struct *cmd, int *retry_cmd)
3057{
3058	unsigned char sense_key;
3059	unsigned char status_byte, msg_byte, host_byte, driver_byte;
3060	int error_value;
3061
3062	*retry_cmd = 0;
3063	/* If we get in here, it means we got "target status", that is, scsi status */
3064	status_byte = cmd->err_info->ScsiStatus;
3065	driver_byte = DRIVER_OK;
3066	msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
3067
3068	if (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC)
3069		host_byte = DID_PASSTHROUGH;
3070	else
3071		host_byte = DID_OK;
3072
3073	error_value = make_status_bytes(status_byte, msg_byte,
3074		host_byte, driver_byte);
3075
3076	if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
3077		if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC)
3078			dev_warn(&h->pdev->dev, "cmd %p "
3079			       "has SCSI Status 0x%x\n",
3080			       cmd, cmd->err_info->ScsiStatus);
3081		return error_value;
3082	}
3083
3084	/* check the sense key */
3085	sense_key = 0xf & cmd->err_info->SenseInfo[2];
3086	/* no status or recovered error */
3087	if (((sense_key == 0x0) || (sense_key == 0x1)) &&
3088	    (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC))
3089		error_value = 0;
3090
3091	if (check_for_unit_attention(h, cmd)) {
3092		*retry_cmd = !(cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC);
3093		return 0;
3094	}
3095
3096	/* Not SG_IO or similar? */
3097	if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC) {
3098		if (error_value != 0)
3099			dev_warn(&h->pdev->dev, "cmd %p has CHECK CONDITION"
3100			       " sense key = 0x%x\n", cmd, sense_key);
3101		return error_value;
3102	}
3103
3104	/* SG_IO or similar, copy sense data back */
3105	if (cmd->rq->sense) {
3106		if (cmd->rq->sense_len > cmd->err_info->SenseLen)
3107			cmd->rq->sense_len = cmd->err_info->SenseLen;
3108		memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
3109			cmd->rq->sense_len);
3110	} else
3111		cmd->rq->sense_len = 0;
3112
3113	return error_value;
3114}
3115
3116/* checks the status of the job and calls complete buffers to mark all
3117 * buffers for the completed job. Note that this function does not need
3118 * to hold the hba/queue lock.
3119 */
3120static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
3121				    int timeout)
3122{
3123	int retry_cmd = 0;
3124	struct request *rq = cmd->rq;
3125
3126	rq->errors = 0;
3127
3128	if (timeout)
3129		rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
3130
3131	if (cmd->err_info->CommandStatus == 0)	/* no error has occurred */
3132		goto after_error_processing;
3133
3134	switch (cmd->err_info->CommandStatus) {
3135	case CMD_TARGET_STATUS:
3136		rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
3137		break;
3138	case CMD_DATA_UNDERRUN:
3139		if (cmd->rq->cmd_type == REQ_TYPE_FS) {
3140			dev_warn(&h->pdev->dev, "cmd %p has"
3141			       " completed with data underrun "
3142			       "reported\n", cmd);
3143			cmd->rq->resid_len = cmd->err_info->ResidualCnt;
3144		}
3145		break;
3146	case CMD_DATA_OVERRUN:
3147		if (cmd->rq->cmd_type == REQ_TYPE_FS)
3148			dev_warn(&h->pdev->dev, "cciss: cmd %p has"
3149			       " completed with data overrun "
3150			       "reported\n", cmd);
3151		break;
3152	case CMD_INVALID:
3153		dev_warn(&h->pdev->dev, "cciss: cmd %p is "
3154		       "reported invalid\n", cmd);
3155		rq->errors = make_status_bytes(SAM_STAT_GOOD,
3156			cmd->err_info->CommandStatus, DRIVER_OK,
3157			(cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3158				DID_PASSTHROUGH : DID_ERROR);
3159		break;
3160	case CMD_PROTOCOL_ERR:
3161		dev_warn(&h->pdev->dev, "cciss: cmd %p has "
3162		       "protocol error\n", cmd);
3163		rq->errors = make_status_bytes(SAM_STAT_GOOD,
3164			cmd->err_info->CommandStatus, DRIVER_OK,
3165			(cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3166				DID_PASSTHROUGH : DID_ERROR);
3167		break;
3168	case CMD_HARDWARE_ERR:
3169		dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3170		       " hardware error\n", cmd);
3171		rq->errors = make_status_bytes(SAM_STAT_GOOD,
3172			cmd->err_info->CommandStatus, DRIVER_OK,
3173			(cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3174				DID_PASSTHROUGH : DID_ERROR);
3175		break;
3176	case CMD_CONNECTION_LOST:
3177		dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3178		       "connection lost\n", cmd);
3179		rq->errors = make_status_bytes(SAM_STAT_GOOD,
3180			cmd->err_info->CommandStatus, DRIVER_OK,
3181			(cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3182				DID_PASSTHROUGH : DID_ERROR);
3183		break;
3184	case CMD_ABORTED:
3185		dev_warn(&h->pdev->dev, "cciss: cmd %p was "
3186		       "aborted\n", cmd);
3187		rq->errors = make_status_bytes(SAM_STAT_GOOD,
3188			cmd->err_info->CommandStatus, DRIVER_OK,
3189			(cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3190				DID_PASSTHROUGH : DID_ABORT);
3191		break;
3192	case CMD_ABORT_FAILED:
3193		dev_warn(&h->pdev->dev, "cciss: cmd %p reports "
3194		       "abort failed\n", cmd);
3195		rq->errors = make_status_bytes(SAM_STAT_GOOD,
3196			cmd->err_info->CommandStatus, DRIVER_OK,
3197			(cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3198				DID_PASSTHROUGH : DID_ERROR);
3199		break;
3200	case CMD_UNSOLICITED_ABORT:
3201		dev_warn(&h->pdev->dev, "cciss%d: unsolicited "
3202		       "abort %p\n", h->ctlr, cmd);
3203		if (cmd->retry_count < MAX_CMD_RETRIES) {
3204			retry_cmd = 1;
3205			dev_warn(&h->pdev->dev, "retrying %p\n", cmd);
3206			cmd->retry_count++;
3207		} else
3208			dev_warn(&h->pdev->dev,
3209				"%p retried too many times\n", cmd);
3210		rq->errors = make_status_bytes(SAM_STAT_GOOD,
3211			cmd->err_info->CommandStatus, DRIVER_OK,
3212			(cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3213				DID_PASSTHROUGH : DID_ABORT);
3214		break;
3215	case CMD_TIMEOUT:
3216		dev_warn(&h->pdev->dev, "cmd %p timedout\n", cmd);
3217		rq->errors = make_status_bytes(SAM_STAT_GOOD,
3218			cmd->err_info->CommandStatus, DRIVER_OK,
3219			(cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3220				DID_PASSTHROUGH : DID_ERROR);
3221		break;
3222	case CMD_UNABORTABLE:
3223		dev_warn(&h->pdev->dev, "cmd %p unabortable\n", cmd);
3224		rq->errors = make_status_bytes(SAM_STAT_GOOD,
3225			cmd->err_info->CommandStatus, DRIVER_OK,
3226			cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC ?
3227				DID_PASSTHROUGH : DID_ERROR);
3228		break;
3229	default:
3230		dev_warn(&h->pdev->dev, "cmd %p returned "
3231		       "unknown status %x\n", cmd,
3232		       cmd->err_info->CommandStatus);
3233		rq->errors = make_status_bytes(SAM_STAT_GOOD,
3234			cmd->err_info->CommandStatus, DRIVER_OK,
3235			(cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3236				DID_PASSTHROUGH : DID_ERROR);
3237	}
3238
3239after_error_processing:
3240
3241	/* We need to return this command */
3242	if (retry_cmd) {
3243		resend_cciss_cmd(h, cmd);
3244		return;
3245	}
3246	cmd->rq->completion_data = cmd;
3247	blk_complete_request(cmd->rq);
3248}
3249
3250static inline u32 cciss_tag_contains_index(u32 tag)
3251{
3252#define DIRECT_LOOKUP_BIT 0x10
3253	return tag & DIRECT_LOOKUP_BIT;
3254}
3255
3256static inline u32 cciss_tag_to_index(u32 tag)
3257{
3258#define DIRECT_LOOKUP_SHIFT 5
3259	return tag >> DIRECT_LOOKUP_SHIFT;
3260}
3261
3262static inline u32 cciss_tag_discard_error_bits(ctlr_info_t *h, u32 tag)
3263{
3264#define CCISS_PERF_ERROR_BITS ((1 << DIRECT_LOOKUP_SHIFT) - 1)
3265#define CCISS_SIMPLE_ERROR_BITS 0x03
3266	if (likely(h->transMethod & CFGTBL_Trans_Performant))
3267		return tag & ~CCISS_PERF_ERROR_BITS;
3268	return tag & ~CCISS_SIMPLE_ERROR_BITS;
3269}
3270
3271static inline void cciss_mark_tag_indexed(u32 *tag)
3272{
3273	*tag |= DIRECT_LOOKUP_BIT;
3274}
3275
3276static inline void cciss_set_tag_index(u32 *tag, u32 index)
3277{
3278	*tag |= (index << DIRECT_LOOKUP_SHIFT);
3279}
3280
3281/*
3282 * Get a request and submit it to the controller.
3283 */
3284static void do_cciss_request(struct request_queue *q)
3285{
3286	ctlr_info_t *h = q->queuedata;
3287	CommandList_struct *c;
3288	sector_t start_blk;
3289	int seg;
3290	struct request *creq;
3291	u64bit temp64;
3292	struct scatterlist *tmp_sg;
3293	SGDescriptor_struct *curr_sg;
3294	drive_info_struct *drv;
3295	int i, dir;
3296	int sg_index = 0;
3297	int chained = 0;
3298
3299      queue:
3300	creq = blk_peek_request(q);
3301	if (!creq)
3302		goto startio;
3303
3304	BUG_ON(creq->nr_phys_segments > h->maxsgentries);
3305
3306	c = cmd_alloc(h);
3307	if (!c)
3308		goto full;
3309
3310	blk_start_request(creq);
3311
3312	tmp_sg = h->scatter_list[c->cmdindex];
3313	spin_unlock_irq(q->queue_lock);
3314
3315	c->cmd_type = CMD_RWREQ;
3316	c->rq = creq;
3317
3318	/* fill in the request */
3319	drv = creq->rq_disk->private_data;
3320	c->Header.ReplyQueue = 0;	/* unused in simple mode */
3321	/* got command from pool, so use the command block index instead */
3322	/* for direct lookups. */
3323	/* The first 2 bits are reserved for controller error reporting. */
3324	cciss_set_tag_index(&c->Header.Tag.lower, c->cmdindex);
3325	cciss_mark_tag_indexed(&c->Header.Tag.lower);
3326	memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3327	c->Request.CDBLen = 10;	/* 12 byte commands not in FW yet; */
3328	c->Request.Type.Type = TYPE_CMD;	/* It is a command. */
3329	c->Request.Type.Attribute = ATTR_SIMPLE;
3330	c->Request.Type.Direction =
3331	    (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3332	c->Request.Timeout = 0;	/* Don't time out */
3333	c->Request.CDB[0] =
3334	    (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3335	start_blk = blk_rq_pos(creq);
3336	dev_dbg(&h->pdev->dev, "sector =%d nr_sectors=%d\n",
3337	       (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3338	sg_init_table(tmp_sg, h->maxsgentries);
3339	seg = blk_rq_map_sg(q, creq, tmp_sg);
3340
3341	/* get the DMA records for the setup */
3342	if (c->Request.Type.Direction == XFER_READ)
3343		dir = PCI_DMA_FROMDEVICE;
3344	else
3345		dir = PCI_DMA_TODEVICE;
3346
3347	curr_sg = c->SG;
3348	sg_index = 0;
3349	chained = 0;
3350
3351	for (i = 0; i < seg; i++) {
3352		if (((sg_index+1) == (h->max_cmd_sgentries)) &&
3353			!chained && ((seg - i) > 1)) {
3354			/* Point to next chain block. */
3355			curr_sg = h->cmd_sg_list[c->cmdindex];
3356			sg_index = 0;
3357			chained = 1;
3358		}
3359		curr_sg[sg_index].Len = tmp_sg[i].length;
3360		temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3361						tmp_sg[i].offset,
3362						tmp_sg[i].length, dir);
3363		curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3364		curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3365		curr_sg[sg_index].Ext = 0;  /* we are not chaining */
3366		++sg_index;
3367	}
3368	if (chained)
3369		cciss_map_sg_chain_block(h, c, h->cmd_sg_list[c->cmdindex],
3370			(seg - (h->max_cmd_sgentries - 1)) *
3371				sizeof(SGDescriptor_struct));
3372
3373	/* track how many SG entries we are using */
3374	if (seg > h->maxSG)
3375		h->maxSG = seg;
3376
3377	dev_dbg(&h->pdev->dev, "Submitting %u sectors in %d segments "
3378			"chained[%d]\n",
3379			blk_rq_sectors(creq), seg, chained);
3380
3381	c->Header.SGTotal = seg + chained;
3382	if (seg <= h->max_cmd_sgentries)
3383		c->Header.SGList = c->Header.SGTotal;
3384	else
3385		c->Header.SGList = h->max_cmd_sgentries;
3386	set_performant_mode(h, c);
3387
3388	if (likely(creq->cmd_type == REQ_TYPE_FS)) {
3389		if(h->cciss_read == CCISS_READ_10) {
3390			c->Request.CDB[1] = 0;
3391			c->Request.CDB[2] = (start_blk >> 24) & 0xff; /* MSB */
3392			c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3393			c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3394			c->Request.CDB[5] = start_blk & 0xff;
3395			c->Request.CDB[6] = 0; /* (sect >> 24) & 0xff; MSB */
3396			c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3397			c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3398			c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3399		} else {
3400			u32 upper32 = upper_32_bits(start_blk);
3401
3402			c->Request.CDBLen = 16;
3403			c->Request.CDB[1]= 0;
3404			c->Request.CDB[2]= (upper32 >> 24) & 0xff; /* MSB */
3405			c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3406			c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3407			c->Request.CDB[5]= upper32 & 0xff;
3408			c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3409			c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3410			c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3411			c->Request.CDB[9]= start_blk & 0xff;
3412			c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3413			c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3414			c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3415			c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3416			c->Request.CDB[14] = c->Request.CDB[15] = 0;
3417		}
3418	} else if (creq->cmd_type == REQ_TYPE_BLOCK_PC) {
3419		c->Request.CDBLen = creq->cmd_len;
3420		memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3421	} else {
3422		dev_warn(&h->pdev->dev, "bad request type %d\n",
3423			creq->cmd_type);
3424		BUG();
3425	}
3426
3427	spin_lock_irq(q->queue_lock);
3428
3429	addQ(&h->reqQ, c);
3430	h->Qdepth++;
3431	if (h->Qdepth > h->maxQsinceinit)
3432		h->maxQsinceinit = h->Qdepth;
3433
3434	goto queue;
3435full:
3436	blk_stop_queue(q);
3437startio:
3438	/* We will already have the driver lock here so not need
3439	 * to lock it.
3440	 */
3441	start_io(h);
3442}
3443
3444static inline unsigned long get_next_completion(ctlr_info_t *h)
3445{
3446	return h->access.command_completed(h);
3447}
3448
3449static inline int interrupt_pending(ctlr_info_t *h)
3450{
3451	return h->access.intr_pending(h);
3452}
3453
3454static inline long interrupt_not_for_us(ctlr_info_t *h)
3455{
3456	return ((h->access.intr_pending(h) == 0) ||
3457		(h->interrupts_enabled == 0));
3458}
3459
3460static inline int bad_tag(ctlr_info_t *h, u32 tag_index,
3461			u32 raw_tag)
3462{
3463	if (unlikely(tag_index >= h->nr_cmds)) {
3464		dev_warn(&h->pdev->dev, "bad tag 0x%08x ignored.\n", raw_tag);
3465		return 1;
3466	}
3467	return 0;
3468}
3469
3470static inline void finish_cmd(ctlr_info_t *h, CommandList_struct *c,
3471				u32 raw_tag)
3472{
3473	removeQ(c);
3474	if (likely(c->cmd_type == CMD_RWREQ))
3475		complete_command(h, c, 0);
3476	else if (c->cmd_type == CMD_IOCTL_PEND)
3477		complete(c->waiting);
3478#ifdef CONFIG_CISS_SCSI_TAPE
3479	else if (c->cmd_type == CMD_SCSI)
3480		complete_scsi_command(c, 0, raw_tag);
3481#endif
3482}
3483
3484static inline u32 next_command(ctlr_info_t *h)
3485{
3486	u32 a;
3487
3488	if (unlikely(!(h->transMethod & CFGTBL_Trans_Performant)))
3489		return h->access.command_completed(h);
3490
3491	if ((*(h->reply_pool_head) & 1) == (h->reply_pool_wraparound)) {
3492		a = *(h->reply_pool_head); /* Next cmd in ring buffer */
3493		(h->reply_pool_head)++;
3494		h->commands_outstanding--;
3495	} else {
3496		a = FIFO_EMPTY;
3497	}
3498	/* Check for wraparound */
3499	if (h->reply_pool_head == (h->reply_pool + h->max_commands)) {
3500		h->reply_pool_head = h->reply_pool;
3501		h->reply_pool_wraparound ^= 1;
3502	}
3503	return a;
3504}
3505
3506/* process completion of an indexed ("direct lookup") command */
3507static inline u32 process_indexed_cmd(ctlr_info_t *h, u32 raw_tag)
3508{
3509	u32 tag_index;
3510	CommandList_struct *c;
3511
3512	tag_index = cciss_tag_to_index(raw_tag);
3513	if (bad_tag(h, tag_index, raw_tag))
3514		return next_command(h);
3515	c = h->cmd_pool + tag_index;
3516	finish_cmd(h, c, raw_tag);
3517	return next_command(h);
3518}
3519
3520/* process completion of a non-indexed command */
3521static inline u32 process_nonindexed_cmd(ctlr_info_t *h, u32 raw_tag)
3522{
3523	CommandList_struct *c = NULL;
3524	__u32 busaddr_masked, tag_masked;
3525
3526	tag_masked = cciss_tag_discard_error_bits(h, raw_tag);
3527	list_for_each_entry(c, &h->cmpQ, list) {
3528		busaddr_masked = cciss_tag_discard_error_bits(h, c->busaddr);
3529		if (busaddr_masked == tag_masked) {
3530			finish_cmd(h, c, raw_tag);
3531			return next_command(h);
3532		}
3533	}
3534	bad_tag(h, h->nr_cmds + 1, raw_tag);
3535	return next_command(h);
3536}
3537
3538/* Some controllers, like p400, will give us one interrupt
3539 * after a soft reset, even if we turned interrupts off.
3540 * Only need to check for this in the cciss_xxx_discard_completions
3541 * functions.
3542 */
3543static int ignore_bogus_interrupt(ctlr_info_t *h)
3544{
3545	if (likely(!reset_devices))
3546		return 0;
3547
3548	if (likely(h->interrupts_enabled))
3549		return 0;
3550
3551	dev_info(&h->pdev->dev, "Received interrupt while interrupts disabled "
3552		"(known firmware bug.)  Ignoring.\n");
3553
3554	return 1;
3555}
3556
3557static irqreturn_t cciss_intx_discard_completions(int irq, void *dev_id)
3558{
3559	ctlr_info_t *h = dev_id;
3560	unsigned long flags;
3561	u32 raw_tag;
3562
3563	if (ignore_bogus_interrupt(h))
3564		return IRQ_NONE;
3565
3566	if (interrupt_not_for_us(h))
3567		return IRQ_NONE;
3568	spin_lock_irqsave(&h->lock, flags);
3569	while (interrupt_pending(h)) {
3570		raw_tag = get_next_completion(h);
3571		while (raw_tag != FIFO_EMPTY)
3572			raw_tag = next_command(h);
3573	}
3574	spin_unlock_irqrestore(&h->lock, flags);
3575	return IRQ_HANDLED;
3576}
3577
3578static irqreturn_t cciss_msix_discard_completions(int irq, void *dev_id)
3579{
3580	ctlr_info_t *h = dev_id;
3581	unsigned long flags;
3582	u32 raw_tag;
3583
3584	if (ignore_bogus_interrupt(h))
3585		return IRQ_NONE;
3586
3587	spin_lock_irqsave(&h->lock, flags);
3588	raw_tag = get_next_completion(h);
3589	while (raw_tag != FIFO_EMPTY)
3590		raw_tag = next_command(h);
3591	spin_unlock_irqrestore(&h->lock, flags);
3592	return IRQ_HANDLED;
3593}
3594
3595static irqreturn_t do_cciss_intx(int irq, void *dev_id)
3596{
3597	ctlr_info_t *h = dev_id;
3598	unsigned long flags;
3599	u32 raw_tag;
3600
3601	if (interrupt_not_for_us(h))
3602		return IRQ_NONE;
3603	spin_lock_irqsave(&h->lock, flags);
3604	while (interrupt_pending(h)) {
3605		raw_tag = get_next_completion(h);
3606		while (raw_tag != FIFO_EMPTY) {
3607			if (cciss_tag_contains_index(raw_tag))
3608				raw_tag = process_indexed_cmd(h, raw_tag);
3609			else
3610				raw_tag = process_nonindexed_cmd(h, raw_tag);
3611		}
3612	}
3613	spin_unlock_irqrestore(&h->lock, flags);
3614	return IRQ_HANDLED;
3615}
3616
3617/* Add a second interrupt handler for MSI/MSI-X mode. In this mode we never
3618 * check the interrupt pending register because it is not set.
3619 */
3620static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id)
3621{
3622	ctlr_info_t *h = dev_id;
3623	unsigned long flags;
3624	u32 raw_tag;
3625
3626	spin_lock_irqsave(&h->lock, flags);
3627	raw_tag = get_next_completion(h);
3628	while (raw_tag != FIFO_EMPTY) {
3629		if (cciss_tag_contains_index(raw_tag))
3630			raw_tag = process_indexed_cmd(h, raw_tag);
3631		else
3632			raw_tag = process_nonindexed_cmd(h, raw_tag);
3633	}
3634	spin_unlock_irqrestore(&h->lock, flags);
3635	return IRQ_HANDLED;
3636}
3637
3638/**
3639 * add_to_scan_list() - add controller to rescan queue
3640 * @h:		      Pointer to the controller.
3641 *
3642 * Adds the controller to the rescan queue if not already on the queue.
3643 *
3644 * returns 1 if added to the queue, 0 if skipped (could be on the
3645 * queue already, or the controller could be initializing or shutting
3646 * down).
3647 **/
3648static int add_to_scan_list(struct ctlr_info *h)
3649{
3650	struct ctlr_info *test_h;
3651	int found = 0;
3652	int ret = 0;
3653
3654	if (h->busy_initializing)
3655		return 0;
3656
3657	if (!mutex_trylock(&h->busy_shutting_down))
3658		return 0;
3659
3660	mutex_lock(&scan_mutex);
3661	list_for_each_entry(test_h, &scan_q, scan_list) {
3662		if (test_h == h) {
3663			found = 1;
3664			break;
3665		}
3666	}
3667	if (!found && !h->busy_scanning) {
3668		INIT_COMPLETION(h->scan_wait);
3669		list_add_tail(&h->scan_list, &scan_q);
3670		ret = 1;
3671	}
3672	mutex_unlock(&scan_mutex);
3673	mutex_unlock(&h->busy_shutting_down);
3674
3675	return ret;
3676}
3677
3678/**
3679 * remove_from_scan_list() - remove controller from rescan queue
3680 * @h:			   Pointer to the controller.
3681 *
3682 * Removes the controller from the rescan queue if present. Blocks if
3683 * the controller is currently conducting a rescan.  The controller
3684 * can be in one of three states:
3685 * 1. Doesn't need a scan
3686 * 2. On the scan list, but not scanning yet (we remove it)
3687 * 3. Busy scanning (and not on the list). In this case we want to wait for
3688 *    the scan to complete to make sure the scanning thread for this
3689 *    controller is completely idle.
3690 **/
3691static void remove_from_scan_list(struct ctlr_info *h)
3692{
3693	struct ctlr_info *test_h, *tmp_h;
3694
3695	mutex_lock(&scan_mutex);
3696	list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3697		if (test_h == h) { /* state 2. */
3698			list_del(&h->scan_list);
3699			complete_all(&h->scan_wait);
3700			mutex_unlock(&scan_mutex);
3701			return;
3702		}
3703	}
3704	if (h->busy_scanning) { /* state 3. */
3705		mutex_unlock(&scan_mutex);
3706		wait_for_completion(&h->scan_wait);
3707	} else { /* state 1, nothing to do. */
3708		mutex_unlock(&scan_mutex);
3709	}
3710}
3711
3712/**
3713 * scan_thread() - kernel thread used to rescan controllers
3714 * @data:	 Ignored.
3715 *
3716 * A kernel thread used scan for drive topology changes on
3717 * controllers. The thread processes only one controller at a time
3718 * using a queue.  Controllers are added to the queue using
3719 * add_to_scan_list() and removed from the queue either after done
3720 * processing or using remove_from_scan_list().
3721 *
3722 * returns 0.
3723 **/
3724static int scan_thread(void *data)
3725{
3726	struct ctlr_info *h;
3727
3728	while (1) {
3729		set_current_state(TASK_INTERRUPTIBLE);
3730		schedule();
3731		if (kthread_should_stop())
3732			break;
3733
3734		while (1) {
3735			mutex_lock(&scan_mutex);
3736			if (list_empty(&scan_q)) {
3737				mutex_unlock(&scan_mutex);
3738				break;
3739			}
3740
3741			h = list_entry(scan_q.next,
3742				       struct ctlr_info,
3743				       scan_list);
3744			list_del(&h->scan_list);
3745			h->busy_scanning = 1;
3746			mutex_unlock(&scan_mutex);
3747
3748			rebuild_lun_table(h, 0, 0);
3749			complete_all(&h->scan_wait);
3750			mutex_lock(&scan_mutex);
3751			h->busy_scanning = 0;
3752			mutex_unlock(&scan_mutex);
3753		}
3754	}
3755
3756	return 0;
3757}
3758
3759static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3760{
3761	if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3762		return 0;
3763
3764	switch (c->err_info->SenseInfo[12]) {
3765	case STATE_CHANGED:
3766		dev_warn(&h->pdev->dev, "a state change "
3767			"detected, command retried\n");
3768		return 1;
3769	break;
3770	case LUN_FAILED:
3771		dev_warn(&h->pdev->dev, "LUN failure "
3772			"detected, action required\n");
3773		return 1;
3774	break;
3775	case REPORT_LUNS_CHANGED:
3776		dev_warn(&h->pdev->dev, "report LUN data changed\n");
3777	/*
3778	 * Here, we could call add_to_scan_list and wake up the scan thread,
3779	 * except that it's quite likely that we will get more than one
3780	 * REPORT_LUNS_CHANGED condition in quick succession, which means
3781	 * that those which occur after the first one will likely happen
3782	 * *during* the scan_thread's rescan.  And the rescan code is not
3783	 * robust enough to restart in the middle, undoing what it has already
3784	 * done, and it's not clear that it's even possible to do this, since
3785	 * part of what it does is notify the block layer, which starts
3786	 * doing it's own i/o to read partition tables and so on, and the
3787	 * driver doesn't have visibility to know what might need undoing.
3788	 * In any event, if possible, it is horribly complicated to get right
3789	 * so we just don't do it for now.
3790	 *
3791	 * Note: this REPORT_LUNS_CHANGED condition only occurs on the MSA2012.
3792	 */
3793		return 1;
3794	break;
3795	case POWER_OR_RESET:
3796		dev_warn(&h->pdev->dev,
3797			"a power on or device reset detected\n");
3798		return 1;
3799	break;
3800	case UNIT_ATTENTION_CLEARED:
3801		dev_warn(&h->pdev->dev,
3802			"unit attention cleared by another initiator\n");
3803		return 1;
3804	break;
3805	default:
3806		dev_warn(&h->pdev->dev, "unknown unit attention detected\n");
3807		return 1;
3808	}
3809}
3810
3811/*
3812 *  We cannot read the structure directly, for portability we must use
3813 *   the io functions.
3814 *   This is for debug only.
3815 */
3816static void print_cfg_table(ctlr_info_t *h)
3817{
3818	int i;
3819	char temp_name[17];
3820	CfgTable_struct *tb = h->cfgtable;
3821
3822	dev_dbg(&h->pdev->dev, "Controller Configuration information\n");
3823	dev_dbg(&h->pdev->dev, "------------------------------------\n");
3824	for (i = 0; i < 4; i++)
3825		temp_name[i] = readb(&(tb->Signature[i]));
3826	temp_name[4] = '\0';
3827	dev_dbg(&h->pdev->dev, "   Signature = %s\n", temp_name);
3828	dev_dbg(&h->pdev->dev, "   Spec Number = %d\n",
3829		readl(&(tb->SpecValence)));
3830	dev_dbg(&h->pdev->dev, "   Transport methods supported = 0x%x\n",
3831	       readl(&(tb->TransportSupport)));
3832	dev_dbg(&h->pdev->dev, "   Transport methods active = 0x%x\n",
3833	       readl(&(tb->TransportActive)));
3834	dev_dbg(&h->pdev->dev, "   Requested transport Method = 0x%x\n",
3835	       readl(&(tb->HostWrite.TransportRequest)));
3836	dev_dbg(&h->pdev->dev, "   Coalesce Interrupt Delay = 0x%x\n",
3837	       readl(&(tb->HostWrite.CoalIntDelay)));
3838	dev_dbg(&h->pdev->dev, "   Coalesce Interrupt Count = 0x%x\n",
3839	       readl(&(tb->HostWrite.CoalIntCount)));
3840	dev_dbg(&h->pdev->dev, "   Max outstanding commands = 0x%d\n",
3841	       readl(&(tb->CmdsOutMax)));
3842	dev_dbg(&h->pdev->dev, "   Bus Types = 0x%x\n",
3843		readl(&(tb->BusTypes)));
3844	for (i = 0; i < 16; i++)
3845		temp_name[i] = readb(&(tb->ServerName[i]));
3846	temp_name[16] = '\0';
3847	dev_dbg(&h->pdev->dev, "   Server Name = %s\n", temp_name);
3848	dev_dbg(&h->pdev->dev, "   Heartbeat Counter = 0x%x\n\n\n",
3849		readl(&(tb->HeartBeat)));
3850}
3851
3852static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3853{
3854	int i, offset, mem_type, bar_type;
3855	if (pci_bar_addr == PCI_BASE_ADDRESS_0)	/* looking for BAR zero? */
3856		return 0;
3857	offset = 0;
3858	for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3859		bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3860		if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3861			offset += 4;
3862		else {
3863			mem_type = pci_resource_flags(pdev, i) &
3864			    PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3865			switch (mem_type) {
3866			case PCI_BASE_ADDRESS_MEM_TYPE_32:
3867			case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3868				offset += 4;	/* 32 bit */
3869				break;
3870			case PCI_BASE_ADDRESS_MEM_TYPE_64:
3871				offset += 8;
3872				break;
3873			default:	/* reserved in PCI 2.2 */
3874				dev_warn(&pdev->dev,
3875				       "Base address is invalid\n");
3876				return -1;
3877				break;
3878			}
3879		}
3880		if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3881			return i + 1;
3882	}
3883	return -1;
3884}
3885
3886/* Fill in bucket_map[], given nsgs (the max number of
3887 * scatter gather elements supported) and bucket[],
3888 * which is an array of 8 integers.  The bucket[] array
3889 * contains 8 different DMA transfer sizes (in 16
3890 * byte increments) which the controller uses to fetch
3891 * commands.  This function fills in bucket_map[], which
3892 * maps a given number of scatter gather elements to one of
3893 * the 8 DMA transfer sizes.  The point of it is to allow the
3894 * controller to only do as much DMA as needed to fetch the
3895 * command, with the DMA transfer size encoded in the lower
3896 * bits of the command address.
3897 */
3898static void  calc_bucket_map(int bucket[], int num_buckets,
3899	int nsgs, int *bucket_map)
3900{
3901	int i, j, b, size;
3902
3903	/* even a command with 0 SGs requires 4 blocks */
3904#define MINIMUM_TRANSFER_BLOCKS 4
3905#define NUM_BUCKETS 8
3906	/* Note, bucket_map must have nsgs+1 entries. */
3907	for (i = 0; i <= nsgs; i++) {
3908		/* Compute size of a command with i SG entries */
3909		size = i + MINIMUM_TRANSFER_BLOCKS;
3910		b = num_buckets; /* Assume the biggest bucket */
3911		/* Find the bucket that is just big enough */
3912		for (j = 0; j < 8; j++) {
3913			if (bucket[j] >= size) {
3914				b = j;
3915				break;
3916			}
3917		}
3918		/* for a command with i SG entries, use bucket b. */
3919		bucket_map[i] = b;
3920	}
3921}
3922
3923static void __devinit cciss_wait_for_mode_change_ack(ctlr_info_t *h)
3924{
3925	int i;
3926
3927	/* under certain very rare conditions, this can take awhile.
3928	 * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3929	 * as we enter this code.) */
3930	for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3931		if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3932			break;
3933		usleep_range(10000, 20000);
3934	}
3935}
3936
3937static __devinit void cciss_enter_performant_mode(ctlr_info_t *h,
3938	u32 use_short_tags)
3939{
3940	/* This is a bit complicated.  There are 8 registers on
3941	 * the controller which we write to to tell it 8 different
3942	 * sizes of commands which there may be.  It's a way of
3943	 * reducing the DMA done to fetch each command.  Encoded into
3944	 * each command's tag are 3 bits which communicate to the controller
3945	 * which of the eight sizes that command fits within.  The size of
3946	 * each command depends on how many scatter gather entries there are.
3947	 * Each SG entry requires 16 bytes.  The eight registers are programmed
3948	 * with the number of 16-byte blocks a command of that size requires.
3949	 * The smallest command possible requires 5 such 16 byte blocks.
3950	 * the largest command possible requires MAXSGENTRIES + 4 16-byte
3951	 * blocks.  Note, this only extends to the SG entries contained
3952	 * within the command block, and does not extend to chained blocks
3953	 * of SG elements.   bft[] contains the eight values we write to
3954	 * the registers.  They are not evenly distributed, but have more
3955	 * sizes for small commands, and fewer sizes for larger commands.
3956	 */
3957	__u32 trans_offset;
3958	int bft[8] = { 5, 6, 8, 10, 12, 20, 28, MAXSGENTRIES + 4};
3959			/*
3960			 *  5 = 1 s/g entry or 4k
3961			 *  6 = 2 s/g entry or 8k
3962			 *  8 = 4 s/g entry or 16k
3963			 * 10 = 6 s/g entry or 24k
3964			 */
3965	unsigned long register_value;
3966	BUILD_BUG_ON(28 > MAXSGENTRIES + 4);
3967
3968	h->reply_pool_wraparound = 1; /* spec: init to 1 */
3969
3970	/* Controller spec: zero out this buffer. */
3971	memset(h->reply_pool, 0, h->max_commands * sizeof(__u64));
3972	h->reply_pool_head = h->reply_pool;
3973
3974	trans_offset = readl(&(h->cfgtable->TransMethodOffset));
3975	calc_bucket_map(bft, ARRAY_SIZE(bft), h->maxsgentries,
3976				h->blockFetchTable);
3977	writel(bft[0], &h->transtable->BlockFetch0);
3978	writel(bft[1], &h->transtable->BlockFetch1);
3979	writel(bft[2], &h->transtable->BlockFetch2);
3980	writel(bft[3], &h->transtable->BlockFetch3);
3981	writel(bft[4], &h->transtable->BlockFetch4);
3982	writel(bft[5], &h->transtable->BlockFetch5);
3983	writel(bft[6], &h->transtable->BlockFetch6);
3984	writel(bft[7], &h->transtable->BlockFetch7);
3985
3986	/* size of controller ring buffer */
3987	writel(h->max_commands, &h->transtable->RepQSize);
3988	writel(1, &h->transtable->RepQCount);
3989	writel(0, &h->transtable->RepQCtrAddrLow32);
3990	writel(0, &h->transtable->RepQCtrAddrHigh32);
3991	writel(h->reply_pool_dhandle, &h->transtable->RepQAddr0Low32);
3992	writel(0, &h->transtable->RepQAddr0High32);
3993	writel(CFGTBL_Trans_Performant | use_short_tags,
3994			&(h->cfgtable->HostWrite.TransportRequest));
3995
3996	writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
3997	cciss_wait_for_mode_change_ack(h);
3998	register_value = readl(&(h->cfgtable->TransportActive));
3999	if (!(register_value & CFGTBL_Trans_Performant))
4000		dev_warn(&h->pdev->dev, "cciss: unable to get board into"
4001					" performant mode\n");
4002}
4003
4004static void __devinit cciss_put_controller_into_performant_mode(ctlr_info_t *h)
4005{
4006	__u32 trans_support;
4007
4008	if (cciss_simple_mode)
4009		return;
4010
4011	dev_dbg(&h->pdev->dev, "Trying to put board into Performant mode\n");
4012	/* Attempt to put controller into performant mode if supported */
4013	/* Does board support performant mode? */
4014	trans_support = readl(&(h->cfgtable->TransportSupport));
4015	if (!(trans_support & PERFORMANT_MODE))
4016		return;
4017
4018	dev_dbg(&h->pdev->dev, "Placing controller into performant mode\n");
4019	/* Performant mode demands commands on a 32 byte boundary
4020	 * pci_alloc_consistent aligns on page boundarys already.
4021	 * Just need to check if divisible by 32
4022	 */
4023	if ((sizeof(CommandList_struct) % 32) != 0) {
4024		dev_warn(&h->pdev->dev, "%s %d %s\n",
4025			"cciss info: command size[",
4026			(int)sizeof(CommandList_struct),
4027			"] not divisible by 32, no performant mode..\n");
4028		return;
4029	}
4030
4031	/* Performant mode ring buffer and supporting data structures */
4032	h->reply_pool = (__u64 *)pci_alloc_consistent(
4033		h->pdev, h->max_commands * sizeof(__u64),
4034		&(h->reply_pool_dhandle));
4035
4036	/* Need a block fetch table for performant mode */
4037	h->blockFetchTable = kmalloc(((h->maxsgentries+1) *
4038		sizeof(__u32)), GFP_KERNEL);
4039
4040	if ((h->reply_pool == NULL) || (h->blockFetchTable == NULL))
4041		goto clean_up;
4042
4043	cciss_enter_performant_mode(h,
4044		trans_support & CFGTBL_Trans_use_short_tags);
4045
4046	/* Change the access methods to the performant access methods */
4047	h->access = SA5_performant_access;
4048	h->transMethod = CFGTBL_Trans_Performant;
4049
4050	return;
4051clean_up:
4052	kfree(h->blockFetchTable);
4053	if (h->reply_pool)
4054		pci_free_consistent(h->pdev,
4055				h->max_commands * sizeof(__u64),
4056				h->reply_pool,
4057				h->reply_pool_dhandle);
4058	return;
4059
4060} /* cciss_put_controller_into_performant_mode */
4061
4062/* If MSI/MSI-X is supported by the kernel we will try to enable it on
4063 * controllers that are capable. If not, we use IO-APIC mode.
4064 */
4065
4066static void __devinit cciss_interrupt_mode(ctlr_info_t *h)
4067{
4068#ifdef CONFIG_PCI_MSI
4069	int err;
4070	struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
4071	{0, 2}, {0, 3}
4072	};
4073
4074	/* Some boards advertise MSI but don't really support it */
4075	if ((h->board_id == 0x40700E11) || (h->board_id == 0x40800E11) ||
4076	    (h->board_id == 0x40820E11) || (h->board_id == 0x40830E11))
4077		goto default_int_mode;
4078
4079	if (pci_find_capability(h->pdev, PCI_CAP_ID_MSIX)) {
4080		err = pci_enable_msix(h->pdev, cciss_msix_entries, 4);
4081		if (!err) {
4082			h->intr[0] = cciss_msix_entries[0].vector;
4083			h->intr[1] = cciss_msix_entries[1].vector;
4084			h->intr[2] = cciss_msix_entries[2].vector;
4085			h->intr[3] = cciss_msix_entries[3].vector;
4086			h->msix_vector = 1;
4087			return;
4088		}
4089		if (err > 0) {
4090			dev_warn(&h->pdev->dev,
4091				"only %d MSI-X vectors available\n", err);
4092			goto default_int_mode;
4093		} else {
4094			dev_warn(&h->pdev->dev,
4095				"MSI-X init failed %d\n", err);
4096			goto default_int_mode;
4097		}
4098	}
4099	if (pci_find_capability(h->pdev, PCI_CAP_ID_MSI)) {
4100		if (!pci_enable_msi(h->pdev))
4101			h->msi_vector = 1;
4102		else
4103			dev_warn(&h->pdev->dev, "MSI init failed\n");
4104	}
4105default_int_mode:
4106#endif				/* CONFIG_PCI_MSI */
4107	/* if we get here we're going to use the default interrupt mode */
4108	h->intr[h->intr_mode] = h->pdev->irq;
4109	return;
4110}
4111
4112static int __devinit cciss_lookup_board_id(struct pci_dev *pdev, u32 *board_id)
4113{
4114	int i;
4115	u32 subsystem_vendor_id, subsystem_device_id;
4116
4117	subsystem_vendor_id = pdev->subsystem_vendor;
4118	subsystem_device_id = pdev->subsystem_device;
4119	*board_id = ((subsystem_device_id << 16) & 0xffff0000) |
4120			subsystem_vendor_id;
4121
4122	for (i = 0; i < ARRAY_SIZE(products); i++)
4123		if (*board_id == products[i].board_id)
4124			return i;
4125	dev_warn(&pdev->dev, "unrecognized board ID: 0x%08x, ignoring.\n",
4126		*board_id);
4127	return -ENODEV;
4128}
4129
4130static inline bool cciss_board_disabled(ctlr_info_t *h)
4131{
4132	u16 command;
4133
4134	(void) pci_read_config_word(h->pdev, PCI_COMMAND, &command);
4135	return ((command & PCI_COMMAND_MEMORY) == 0);
4136}
4137
4138static int __devinit cciss_pci_find_memory_BAR(struct pci_dev *pdev,
4139	unsigned long *memory_bar)
4140{
4141	int i;
4142
4143	for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
4144		if (pci_resource_flags(pdev, i) & IORESOURCE_MEM) {
4145			/* addressing mode bits already removed */
4146			*memory_bar = pci_resource_start(pdev, i);
4147			dev_dbg(&pdev->dev, "memory BAR = %lx\n",
4148				*memory_bar);
4149			return 0;
4150		}
4151	dev_warn(&pdev->dev, "no memory BAR found\n");
4152	return -ENODEV;
4153}
4154
4155static int __devinit cciss_wait_for_board_state(struct pci_dev *pdev,
4156	void __iomem *vaddr, int wait_for_ready)
4157#define BOARD_READY 1
4158#define BOARD_NOT_READY 0
4159{
4160	int i, iterations;
4161	u32 scratchpad;
4162
4163	if (wait_for_ready)
4164		iterations = CCISS_BOARD_READY_ITERATIONS;
4165	else
4166		iterations = CCISS_BOARD_NOT_READY_ITERATIONS;
4167
4168	for (i = 0; i < iterations; i++) {
4169		scratchpad = readl(vaddr + SA5_SCRATCHPAD_OFFSET);
4170		if (wait_for_ready) {
4171			if (scratchpad == CCISS_FIRMWARE_READY)
4172				return 0;
4173		} else {
4174			if (scratchpad != CCISS_FIRMWARE_READY)
4175				return 0;
4176		}
4177		msleep(CCISS_BOARD_READY_POLL_INTERVAL_MSECS);
4178	}
4179	dev_warn(&pdev->dev, "board not ready, timed out.\n");
4180	return -ENODEV;
4181}
4182
4183static int __devinit cciss_find_cfg_addrs(struct pci_dev *pdev,
4184	void __iomem *vaddr, u32 *cfg_base_addr, u64 *cfg_base_addr_index,
4185	u64 *cfg_offset)
4186{
4187	*cfg_base_addr = readl(vaddr + SA5_CTCFG_OFFSET);
4188	*cfg_offset = readl(vaddr + SA5_CTMEM_OFFSET);
4189	*cfg_base_addr &= (u32) 0x0000ffff;
4190	*cfg_base_addr_index = find_PCI_BAR_index(pdev, *cfg_base_addr);
4191	if (*cfg_base_addr_index == -1) {
4192		dev_warn(&pdev->dev, "cannot find cfg_base_addr_index, "
4193			"*cfg_base_addr = 0x%08x\n", *cfg_base_addr);
4194		return -ENODEV;
4195	}
4196	return 0;
4197}
4198
4199static int __devinit cciss_find_cfgtables(ctlr_info_t *h)
4200{
4201	u64 cfg_offset;
4202	u32 cfg_base_addr;
4203	u64 cfg_base_addr_index;
4204	u32 trans_offset;
4205	int rc;
4206
4207	rc = cciss_find_cfg_addrs(h->pdev, h->vaddr, &cfg_base_addr,
4208		&cfg_base_addr_index, &cfg_offset);
4209	if (rc)
4210		return rc;
4211	h->cfgtable = remap_pci_mem(pci_resource_start(h->pdev,
4212		cfg_base_addr_index) + cfg_offset, sizeof(h->cfgtable));
4213	if (!h->cfgtable)
4214		return -ENOMEM;
4215	rc = write_driver_ver_to_cfgtable(h->cfgtable);
4216	if (rc)
4217		return rc;
4218	/* Find performant mode table. */
4219	trans_offset = readl(&h->cfgtable->TransMethodOffset);
4220	h->transtable = remap_pci_mem(pci_resource_start(h->pdev,
4221				cfg_base_addr_index)+cfg_offset+trans_offset,
4222				sizeof(*h->transtable));
4223	if (!h->transtable)
4224		return -ENOMEM;
4225	return 0;
4226}
4227
4228static void __devinit cciss_get_max_perf_mode_cmds(struct ctlr_info *h)
4229{
4230	h->max_commands = readl(&(h->cfgtable->MaxPerformantModeCommands));
4231
4232	/* Limit commands in memory limited kdump scenario. */
4233	if (reset_devices && h->max_commands > 32)
4234		h->max_commands = 32;
4235
4236	if (h->max_commands < 16) {
4237		dev_warn(&h->pdev->dev, "Controller reports "
4238			"max supported commands of %d, an obvious lie. "
4239			"Using 16.  Ensure that firmware is up to date.\n",
4240			h->max_commands);
4241		h->max_commands = 16;
4242	}
4243}
4244
4245/* Interrogate the hardware for some limits:
4246 * max commands, max SG elements without chaining, and with chaining,
4247 * SG chain block size, etc.
4248 */
4249static void __devinit cciss_find_board_params(ctlr_info_t *h)
4250{
4251	cciss_get_max_perf_mode_cmds(h);
4252	h->nr_cmds = h->max_commands - 4 - cciss_tape_cmds;
4253	h->maxsgentries = readl(&(h->cfgtable->MaxSGElements));
4254	/*
4255	 * Limit in-command s/g elements to 32 save dma'able memory.
4256	 * Howvever spec says if 0, use 31
4257	 */
4258	h->max_cmd_sgentries = 31;
4259	if (h->maxsgentries > 512) {
4260		h->max_cmd_sgentries = 32;
4261		h->chainsize = h->maxsgentries - h->max_cmd_sgentries + 1;
4262		h->maxsgentries--; /* save one for chain pointer */
4263	} else {
4264		h->maxsgentries = 31; /* default to traditional values */
4265		h->chainsize = 0;
4266	}
4267}
4268
4269static inline bool CISS_signature_present(ctlr_info_t *h)
4270{
4271	if ((readb(&h->cfgtable->Signature[0]) != 'C') ||
4272	    (readb(&h->cfgtable->Signature[1]) != 'I') ||
4273	    (readb(&h->cfgtable->Signature[2]) != 'S') ||
4274	    (readb(&h->cfgtable->Signature[3]) != 'S')) {
4275		dev_warn(&h->pdev->dev, "not a valid CISS config table\n");
4276		return false;
4277	}
4278	return true;
4279}
4280
4281/* Need to enable prefetch in the SCSI core for 6400 in x86 */
4282static inline void cciss_enable_scsi_prefetch(ctlr_info_t *h)
4283{
4284#ifdef CONFIG_X86
4285	u32 prefetch;
4286
4287	prefetch = readl(&(h->cfgtable->SCSI_Prefetch));
4288	prefetch |= 0x100;
4289	writel(prefetch, &(h->cfgtable->SCSI_Prefetch));
4290#endif
4291}
4292
4293/* Disable DMA prefetch for the P600.  Otherwise an ASIC bug may result
4294 * in a prefetch beyond physical memory.
4295 */
4296static inline void cciss_p600_dma_prefetch_quirk(ctlr_info_t *h)
4297{
4298	u32 dma_prefetch;
4299	__u32 dma_refetch;
4300
4301	if (h->board_id != 0x3225103C)
4302		return;
4303	dma_prefetch = readl(h->vaddr + I2O_DMA1_CFG);
4304	dma_prefetch |= 0x8000;
4305	writel(dma_prefetch, h->vaddr + I2O_DMA1_CFG);
4306	pci_read_config_dword(h->pdev, PCI_COMMAND_PARITY, &dma_refetch);
4307	dma_refetch |= 0x1;
4308	pci_write_config_dword(h->pdev, PCI_COMMAND_PARITY, dma_refetch);
4309}
4310
4311static int __devinit cciss_pci_init(ctlr_info_t *h)
4312{
4313	int prod_index, err;
4314
4315	prod_index = cciss_lookup_board_id(h->pdev, &h->board_id);
4316	if (prod_index < 0)
4317		return -ENODEV;
4318	h->product_name = products[prod_index].product_name;
4319	h->access = *(products[prod_index].access);
4320
4321	if (cciss_board_disabled(h)) {
4322		dev_warn(&h->pdev->dev, "controller appears to be disabled\n");
4323		return -ENODEV;
4324	}
4325
4326	pci_disable_link_state(h->pdev, PCIE_LINK_STATE_L0S |
4327				PCIE_LINK_STATE_L1 | PCIE_LINK_STATE_CLKPM);
4328
4329	err = pci_enable_device(h->pdev);
4330	if (err) {
4331		dev_warn(&h->pdev->dev, "Unable to Enable PCI device\n");
4332		return err;
4333	}
4334
4335	err = pci_request_regions(h->pdev, "cciss");
4336	if (err) {
4337		dev_warn(&h->pdev->dev,
4338			"Cannot obtain PCI resources, aborting\n");
4339		return err;
4340	}
4341
4342	dev_dbg(&h->pdev->dev, "irq = %x\n", h->pdev->irq);
4343	dev_dbg(&h->pdev->dev, "board_id = %x\n", h->board_id);
4344
4345/* If the kernel supports MSI/MSI-X we will try to enable that functionality,
4346 * else we use the IO-APIC interrupt assigned to us by system ROM.
4347 */
4348	cciss_interrupt_mode(h);
4349	err = cciss_pci_find_memory_BAR(h->pdev, &h->paddr);
4350	if (err)
4351		goto err_out_free_res;
4352	h->vaddr = remap_pci_mem(h->paddr, 0x250);
4353	if (!h->vaddr) {
4354		err = -ENOMEM;
4355		goto err_out_free_res;
4356	}
4357	err = cciss_wait_for_board_state(h->pdev, h->vaddr, BOARD_READY);
4358	if (err)
4359		goto err_out_free_res;
4360	err = cciss_find_cfgtables(h);
4361	if (err)
4362		goto err_out_free_res;
4363	print_cfg_table(h);
4364	cciss_find_board_params(h);
4365
4366	if (!CISS_signature_present(h)) {
4367		err = -ENODEV;
4368		goto err_out_free_res;
4369	}
4370	cciss_enable_scsi_prefetch(h);
4371	cciss_p600_dma_prefetch_quirk(h);
4372	err = cciss_enter_simple_mode(h);
4373	if (err)
4374		goto err_out_free_res;
4375	cciss_put_controller_into_performant_mode(h);
4376	return 0;
4377
4378err_out_free_res:
4379	/*
4380	 * Deliberately omit pci_disable_device(): it does something nasty to
4381	 * Smart Array controllers that pci_enable_device does not undo
4382	 */
4383	if (h->transtable)
4384		iounmap(h->transtable);
4385	if (h->cfgtable)
4386		iounmap(h->cfgtable);
4387	if (h->vaddr)
4388		iounmap(h->vaddr);
4389	pci_release_regions(h->pdev);
4390	return err;
4391}
4392
4393/* Function to find the first free pointer into our hba[] array
4394 * Returns -1 if no free entries are left.
4395 */
4396static int alloc_cciss_hba(struct pci_dev *pdev)
4397{
4398	int i;
4399
4400	for (i = 0; i < MAX_CTLR; i++) {
4401		if (!hba[i]) {
4402			ctlr_info_t *h;
4403
4404			h = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
4405			if (!h)
4406				goto Enomem;
4407			hba[i] = h;
4408			return i;
4409		}
4410	}
4411	dev_warn(&pdev->dev, "This driver supports a maximum"
4412	       " of %d controllers.\n", MAX_CTLR);
4413	return -1;
4414Enomem:
4415	dev_warn(&pdev->dev, "out of memory.\n");
4416	return -1;
4417}
4418
4419static void free_hba(ctlr_info_t *h)
4420{
4421	int i;
4422
4423	hba[h->ctlr] = NULL;
4424	for (i = 0; i < h->highest_lun + 1; i++)
4425		if (h->gendisk[i] != NULL)
4426			put_disk(h->gendisk[i]);
4427	kfree(h);
4428}
4429
4430/* Send a message CDB to the firmware. */
4431static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
4432{
4433	typedef struct {
4434		CommandListHeader_struct CommandHeader;
4435		RequestBlock_struct Request;
4436		ErrDescriptor_struct ErrorDescriptor;
4437	} Command;
4438	static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
4439	Command *cmd;
4440	dma_addr_t paddr64;
4441	uint32_t paddr32, tag;
4442	void __iomem *vaddr;
4443	int i, err;
4444
4445	vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
4446	if (vaddr == NULL)
4447		return -ENOMEM;
4448
4449	/* The Inbound Post Queue only accepts 32-bit physical addresses for the
4450	   CCISS commands, so they must be allocated from the lower 4GiB of
4451	   memory. */
4452	err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
4453	if (err) {
4454		iounmap(vaddr);
4455		return -ENOMEM;
4456	}
4457
4458	cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
4459	if (cmd == NULL) {
4460		iounmap(vaddr);
4461		return -ENOMEM;
4462	}
4463
4464	/* This must fit, because of the 32-bit consistent DMA mask.  Also,
4465	   although there's no guarantee, we assume that the address is at
4466	   least 4-byte aligned (most likely, it's page-aligned). */
4467	paddr32 = paddr64;
4468
4469	cmd->CommandHeader.ReplyQueue = 0;
4470	cmd->CommandHeader.SGList = 0;
4471	cmd->CommandHeader.SGTotal = 0;
4472	cmd->CommandHeader.Tag.lower = paddr32;
4473	cmd->CommandHeader.Tag.upper = 0;
4474	memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
4475
4476	cmd->Request.CDBLen = 16;
4477	cmd->Request.Type.Type = TYPE_MSG;
4478	cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
4479	cmd->Request.Type.Direction = XFER_NONE;
4480	cmd->Request.Timeout = 0; /* Don't time out */
4481	cmd->Request.CDB[0] = opcode;
4482	cmd->Request.CDB[1] = type;
4483	memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
4484
4485	cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
4486	cmd->ErrorDescriptor.Addr.upper = 0;
4487	cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
4488
4489	writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
4490
4491	for (i = 0; i < 10; i++) {
4492		tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
4493		if ((tag & ~3) == paddr32)
4494			break;
4495		msleep(CCISS_POST_RESET_NOOP_TIMEOUT_MSECS);
4496	}
4497
4498	iounmap(vaddr);
4499
4500	/* we leak the DMA buffer here ... no choice since the controller could
4501	   still complete the command. */
4502	if (i == 10) {
4503		dev_err(&pdev->dev,
4504			"controller message %02x:%02x timed out\n",
4505			opcode, type);
4506		return -ETIMEDOUT;
4507	}
4508
4509	pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
4510
4511	if (tag & 2) {
4512		dev_err(&pdev->dev, "controller message %02x:%02x failed\n",
4513			opcode, type);
4514		return -EIO;
4515	}
4516
4517	dev_info(&pdev->dev, "controller message %02x:%02x succeeded\n",
4518		opcode, type);
4519	return 0;
4520}
4521
4522#define cciss_noop(p) cciss_message(p, 3, 0)
4523
4524static int cciss_controller_hard_reset(struct pci_dev *pdev,
4525	void * __iomem vaddr, u32 use_doorbell)
4526{
4527	u16 pmcsr;
4528	int pos;
4529
4530	if (use_doorbell) {
4531		/* For everything after the P600, the PCI power state method
4532		 * of resetting the controller doesn't work, so we have this
4533		 * other way using the doorbell register.
4534		 */
4535		dev_info(&pdev->dev, "using doorbell to reset controller\n");
4536		writel(use_doorbell, vaddr + SA5_DOORBELL);
4537	} else { /* Try to do it the PCI power state way */
4538
4539		/* Quoting from the Open CISS Specification: "The Power
4540		 * Management Control/Status Register (CSR) controls the power
4541		 * state of the device.  The normal operating state is D0,
4542		 * CSR=00h.  The software off state is D3, CSR=03h.  To reset
4543		 * the controller, place the interface device in D3 then to D0,
4544		 * this causes a secondary PCI reset which will reset the
4545		 * controller." */
4546
4547		pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4548		if (pos == 0) {
4549			dev_err(&pdev->dev,
4550				"cciss_controller_hard_reset: "
4551				"PCI PM not supported\n");
4552			return -ENODEV;
4553		}
4554		dev_info(&pdev->dev, "using PCI PM to reset controller\n");
4555		/* enter the D3hot power management state */
4556		pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4557		pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4558		pmcsr |= PCI_D3hot;
4559		pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4560
4561		msleep(500);
4562
4563		/* enter the D0 power management state */
4564		pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4565		pmcsr |= PCI_D0;
4566		pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4567
4568		/*
4569		 * The P600 requires a small delay when changing states.
4570		 * Otherwise we may think the board did not reset and we bail.
4571		 * This for kdump only and is particular to the P600.
4572		 */
4573		msleep(500);
4574	}
4575	return 0;
4576}
4577
4578static __devinit void init_driver_version(char *driver_version, int len)
4579{
4580	memset(driver_version, 0, len);
4581	strncpy(driver_version, "cciss " DRIVER_NAME, len - 1);
4582}
4583
4584static __devinit int write_driver_ver_to_cfgtable(
4585	CfgTable_struct __iomem *cfgtable)
4586{
4587	char *driver_version;
4588	int i, size = sizeof(cfgtable->driver_version);
4589
4590	driver_version = kmalloc(size, GFP_KERNEL);
4591	if (!driver_version)
4592		return -ENOMEM;
4593
4594	init_driver_version(driver_version, size);
4595	for (i = 0; i < size; i++)
4596		writeb(driver_version[i], &cfgtable->driver_version[i]);
4597	kfree(driver_version);
4598	return 0;
4599}
4600
4601static __devinit void read_driver_ver_from_cfgtable(
4602	CfgTable_struct __iomem *cfgtable, unsigned char *driver_ver)
4603{
4604	int i;
4605
4606	for (i = 0; i < sizeof(cfgtable->driver_version); i++)
4607		driver_ver[i] = readb(&cfgtable->driver_version[i]);
4608}
4609
4610static __devinit int controller_reset_failed(
4611	CfgTable_struct __iomem *cfgtable)
4612{
4613
4614	char *driver_ver, *old_driver_ver;
4615	int rc, size = sizeof(cfgtable->driver_version);
4616
4617	old_driver_ver = kmalloc(2 * size, GFP_KERNEL);
4618	if (!old_driver_ver)
4619		return -ENOMEM;
4620	driver_ver = old_driver_ver + size;
4621
4622	/* After a reset, the 32 bytes of "driver version" in the cfgtable
4623	 * should have been changed, otherwise we know the reset failed.
4624	 */
4625	init_driver_version(old_driver_ver, size);
4626	read_driver_ver_from_cfgtable(cfgtable, driver_ver);
4627	rc = !memcmp(driver_ver, old_driver_ver, size);
4628	kfree(old_driver_ver);
4629	return rc;
4630}
4631
4632/* This does a hard reset of the controller using PCI power management
4633 * states or using the doorbell register. */
4634static __devinit int cciss_kdump_hard_reset_controller(struct pci_dev *pdev)
4635{
4636	u64 cfg_offset;
4637	u32 cfg_base_addr;
4638	u64 cfg_base_addr_index;
4639	void __iomem *vaddr;
4640	unsigned long paddr;
4641	u32 misc_fw_support;
4642	int rc;
4643	CfgTable_struct __iomem *cfgtable;
4644	u32 use_doorbell;
4645	u32 board_id;
4646	u16 command_register;
4647
4648	/* For controllers as old a the p600, this is very nearly
4649	 * the same thing as
4650	 *
4651	 * pci_save_state(pci_dev);
4652	 * pci_set_power_state(pci_dev, PCI_D3hot);
4653	 * pci_set_power_state(pci_dev, PCI_D0);
4654	 * pci_restore_state(pci_dev);
4655	 *
4656	 * For controllers newer than the P600, the pci power state
4657	 * method of resetting doesn't work so we have another way
4658	 * using the doorbell register.
4659	 */
4660
4661	/* Exclude 640x boards.  These are two pci devices in one slot
4662	 * which share a battery backed cache module.  One controls the
4663	 * cache, the other accesses the cache through the one that controls
4664	 * it.  If we reset the one controlling the cache, the other will
4665	 * likely not be happy.  Just forbid resetting this conjoined mess.
4666	 */
4667	cciss_lookup_board_id(pdev, &board_id);
4668	if (!ctlr_is_resettable(board_id)) {
4669		dev_warn(&pdev->dev, "Cannot reset Smart Array 640x "
4670				"due to shared cache module.");
4671		return -ENODEV;
4672	}
4673
4674	/* if controller is soft- but not hard resettable... */
4675	if (!ctlr_is_hard_resettable(board_id))
4676		return -ENOTSUPP; /* try soft reset later. */
4677
4678	/* Save the PCI command register */
4679	pci_read_config_word(pdev, 4, &command_register);
4680	/* Turn the board off.  This is so that later pci_restore_state()
4681	 * won't turn the board on before the rest of config space is ready.
4682	 */
4683	pci_disable_device(pdev);
4684	pci_save_state(pdev);
4685
4686	/* find the first memory BAR, so we can find the cfg table */
4687	rc = cciss_pci_find_memory_BAR(pdev, &paddr);
4688	if (rc)
4689		return rc;
4690	vaddr = remap_pci_mem(paddr, 0x250);
4691	if (!vaddr)
4692		return -ENOMEM;
4693
4694	/* find cfgtable in order to check if reset via doorbell is supported */
4695	rc = cciss_find_cfg_addrs(pdev, vaddr, &cfg_base_addr,
4696					&cfg_base_addr_index, &cfg_offset);
4697	if (rc)
4698		goto unmap_vaddr;
4699	cfgtable = remap_pci_mem(pci_resource_start(pdev,
4700		       cfg_base_addr_index) + cfg_offset, sizeof(*cfgtable));
4701	if (!cfgtable) {
4702		rc = -ENOMEM;
4703		goto unmap_vaddr;
4704	}
4705	rc = write_driver_ver_to_cfgtable(cfgtable);
4706	if (rc)
4707		goto unmap_vaddr;
4708
4709	/* If reset via doorbell register is supported, use that.
4710	 * There are two such methods.  Favor the newest method.
4711	 */
4712	misc_fw_support = readl(&cfgtable->misc_fw_support);
4713	use_doorbell = misc_fw_support & MISC_FW_DOORBELL_RESET2;
4714	if (use_doorbell) {
4715		use_doorbell = DOORBELL_CTLR_RESET2;
4716	} else {
4717		use_doorbell = misc_fw_support & MISC_FW_DOORBELL_RESET;
4718		if (use_doorbell) {
4719			dev_warn(&pdev->dev, "Controller claims that "
4720				"'Bit 2 doorbell reset' is "
4721				"supported, but not 'bit 5 doorbell reset'.  "
4722				"Firmware update is recommended.\n");
4723			rc = -ENOTSUPP; /* use the soft reset */
4724			goto unmap_cfgtable;
4725		}
4726	}
4727
4728	rc = cciss_controller_hard_reset(pdev, vaddr, use_doorbell);
4729	if (rc)
4730		goto unmap_cfgtable;
4731	pci_restore_state(pdev);
4732	rc = pci_enable_device(pdev);
4733	if (rc) {
4734		dev_warn(&pdev->dev, "failed to enable device.\n");
4735		goto unmap_cfgtable;
4736	}
4737	pci_write_config_word(pdev, 4, command_register);
4738
4739	/* Some devices (notably the HP Smart Array 5i Controller)
4740	   need a little pause here */
4741	msleep(CCISS_POST_RESET_PAUSE_MSECS);
4742
4743	/* Wait for board to become not ready, then ready. */
4744	dev_info(&pdev->dev, "Waiting for board to reset.\n");
4745	rc = cciss_wait_for_board_state(pdev, vaddr, BOARD_NOT_READY);
4746	if (rc) {
4747		dev_warn(&pdev->dev, "Failed waiting for board to hard reset."
4748				"  Will try soft reset.\n");
4749		rc = -ENOTSUPP; /* Not expected, but try soft reset later */
4750		goto unmap_cfgtable;
4751	}
4752	rc = cciss_wait_for_board_state(pdev, vaddr, BOARD_READY);
4753	if (rc) {
4754		dev_warn(&pdev->dev,
4755			"failed waiting for board to become ready "
4756			"after hard reset\n");
4757		goto unmap_cfgtable;
4758	}
4759
4760	rc = controller_reset_failed(vaddr);
4761	if (rc < 0)
4762		goto unmap_cfgtable;
4763	if (rc) {
4764		dev_warn(&pdev->dev, "Unable to successfully hard reset "
4765			"controller. Will try soft reset.\n");
4766		rc = -ENOTSUPP; /* Not expected, but try soft reset later */
4767	} else {
4768		dev_info(&pdev->dev, "Board ready after hard reset.\n");
4769	}
4770
4771unmap_cfgtable:
4772	iounmap(cfgtable);
4773
4774unmap_vaddr:
4775	iounmap(vaddr);
4776	return rc;
4777}
4778
4779static __devinit int cciss_init_reset_devices(struct pci_dev *pdev)
4780{
4781	int rc, i;
4782
4783	if (!reset_devices)
4784		return 0;
4785
4786	/* Reset the controller with a PCI power-cycle or via doorbell */
4787	rc = cciss_kdump_hard_reset_controller(pdev);
4788
4789	/* -ENOTSUPP here means we cannot reset the controller
4790	 * but it's already (and still) up and running in
4791	 * "performant mode".  Or, it might be 640x, which can't reset
4792	 * due to concerns about shared bbwc between 6402/6404 pair.
4793	 */
4794	if (rc == -ENOTSUPP)
4795		return rc; /* just try to do the kdump anyhow. */
4796	if (rc)
4797		return -ENODEV;
4798
4799	/* Now try to get the controller to respond to a no-op */
4800	dev_warn(&pdev->dev, "Waiting for controller to respond to no-op\n");
4801	for (i = 0; i < CCISS_POST_RESET_NOOP_RETRIES; i++) {
4802		if (cciss_noop(pdev) == 0)
4803			break;
4804		else
4805			dev_warn(&pdev->dev, "no-op failed%s\n",
4806				(i < CCISS_POST_RESET_NOOP_RETRIES - 1 ?
4807					"; re-trying" : ""));
4808		msleep(CCISS_POST_RESET_NOOP_INTERVAL_MSECS);
4809	}
4810	return 0;
4811}
4812
4813static __devinit int cciss_allocate_cmd_pool(ctlr_info_t *h)
4814{
4815	h->cmd_pool_bits = kmalloc(
4816		DIV_ROUND_UP(h->nr_cmds, BITS_PER_LONG) *
4817		sizeof(unsigned long), GFP_KERNEL);
4818	h->cmd_pool = pci_alloc_consistent(h->pdev,
4819		h->nr_cmds * sizeof(CommandList_struct),
4820		&(h->cmd_pool_dhandle));
4821	h->errinfo_pool = pci_alloc_consistent(h->pdev,
4822		h->nr_cmds * sizeof(ErrorInfo_struct),
4823		&(h->errinfo_pool_dhandle));
4824	if ((h->cmd_pool_bits == NULL)
4825		|| (h->cmd_pool == NULL)
4826		|| (h->errinfo_pool == NULL)) {
4827		dev_err(&h->pdev->dev, "out of memory");
4828		return -ENOMEM;
4829	}
4830	return 0;
4831}
4832
4833static __devinit int cciss_allocate_scatterlists(ctlr_info_t *h)
4834{
4835	int i;
4836
4837	/* zero it, so that on free we need not know how many were alloc'ed */
4838	h->scatter_list = kzalloc(h->max_commands *
4839				sizeof(struct scatterlist *), GFP_KERNEL);
4840	if (!h->scatter_list)
4841		return -ENOMEM;
4842
4843	for (i = 0; i < h->nr_cmds; i++) {
4844		h->scatter_list[i] = kmalloc(sizeof(struct scatterlist) *
4845						h->maxsgentries, GFP_KERNEL);
4846		if (h->scatter_list[i] == NULL) {
4847			dev_err(&h->pdev->dev, "could not allocate "
4848				"s/g lists\n");
4849			return -ENOMEM;
4850		}
4851	}
4852	return 0;
4853}
4854
4855static void cciss_free_scatterlists(ctlr_info_t *h)
4856{
4857	int i;
4858
4859	if (h->scatter_list) {
4860		for (i = 0; i < h->nr_cmds; i++)
4861			kfree(h->scatter_list[i]);
4862		kfree(h->scatter_list);
4863	}
4864}
4865
4866static void cciss_free_cmd_pool(ctlr_info_t *h)
4867{
4868	kfree(h->cmd_pool_bits);
4869	if (h->cmd_pool)
4870		pci_free_consistent(h->pdev,
4871			h->nr_cmds * sizeof(CommandList_struct),
4872			h->cmd_pool, h->cmd_pool_dhandle);
4873	if (h->errinfo_pool)
4874		pci_free_consistent(h->pdev,
4875			h->nr_cmds * sizeof(ErrorInfo_struct),
4876			h->errinfo_pool, h->errinfo_pool_dhandle);
4877}
4878
4879static int cciss_request_irq(ctlr_info_t *h,
4880	irqreturn_t (*msixhandler)(int, void *),
4881	irqreturn_t (*intxhandler)(int, void *))
4882{
4883	if (h->msix_vector || h->msi_vector) {
4884		if (!request_irq(h->intr[h->intr_mode], msixhandler,
4885				0, h->devname, h))
4886			return 0;
4887		dev_err(&h->pdev->dev, "Unable to get msi irq %d"
4888			" for %s\n", h->intr[h->intr_mode],
4889			h->devname);
4890		return -1;
4891	}
4892
4893	if (!request_irq(h->intr[h->intr_mode], intxhandler,
4894			IRQF_SHARED, h->devname, h))
4895		return 0;
4896	dev_err(&h->pdev->dev, "Unable to get irq %d for %s\n",
4897		h->intr[h->intr_mode], h->devname);
4898	return -1;
4899}
4900
4901static int __devinit cciss_kdump_soft_reset(ctlr_info_t *h)
4902{
4903	if (cciss_send_reset(h, CTLR_LUNID, CCISS_RESET_TYPE_CONTROLLER)) {
4904		dev_warn(&h->pdev->dev, "Resetting array controller failed.\n");
4905		return -EIO;
4906	}
4907
4908	dev_info(&h->pdev->dev, "Waiting for board to soft reset.\n");
4909	if (cciss_wait_for_board_state(h->pdev, h->vaddr, BOARD_NOT_READY)) {
4910		dev_warn(&h->pdev->dev, "Soft reset had no effect.\n");
4911		return -1;
4912	}
4913
4914	dev_info(&h->pdev->dev, "Board reset, awaiting READY status.\n");
4915	if (cciss_wait_for_board_state(h->pdev, h->vaddr, BOARD_READY)) {
4916		dev_warn(&h->pdev->dev, "Board failed to become ready "
4917			"after soft reset.\n");
4918		return -1;
4919	}
4920
4921	return 0;
4922}
4923
4924static void cciss_undo_allocations_after_kdump_soft_reset(ctlr_info_t *h)
4925{
4926	int ctlr = h->ctlr;
4927
4928	free_irq(h->intr[h->intr_mode], h);
4929#ifdef CONFIG_PCI_MSI
4930	if (h->msix_vector)
4931		pci_disable_msix(h->pdev);
4932	else if (h->msi_vector)
4933		pci_disable_msi(h->pdev);
4934#endif /* CONFIG_PCI_MSI */
4935	cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
4936	cciss_free_scatterlists(h);
4937	cciss_free_cmd_pool(h);
4938	kfree(h->blockFetchTable);
4939	if (h->reply_pool)
4940		pci_free_consistent(h->pdev, h->max_commands * sizeof(__u64),
4941				h->reply_pool, h->reply_pool_dhandle);
4942	if (h->transtable)
4943		iounmap(h->transtable);
4944	if (h->cfgtable)
4945		iounmap(h->cfgtable);
4946	if (h->vaddr)
4947		iounmap(h->vaddr);
4948	unregister_blkdev(h->major, h->devname);
4949	cciss_destroy_hba_sysfs_entry(h);
4950	pci_release_regions(h->pdev);
4951	kfree(h);
4952	hba[ctlr] = NULL;
4953}
4954
4955/*
4956 *  This is it.  Find all the controllers and register them.  I really hate
4957 *  stealing all these major device numbers.
4958 *  returns the number of block devices registered.
4959 */
4960static int __devinit cciss_init_one(struct pci_dev *pdev,
4961				    const struct pci_device_id *ent)
4962{
4963	int i;
4964	int j = 0;
4965	int rc;
4966	int try_soft_reset = 0;
4967	int dac, return_code;
4968	InquiryData_struct *inq_buff;
4969	ctlr_info_t *h;
4970	unsigned long flags;
4971
4972	rc = cciss_init_reset_devices(pdev);
4973	if (rc) {
4974		if (rc != -ENOTSUPP)
4975			return rc;
4976		/* If the reset fails in a particular way (it has no way to do
4977		 * a proper hard reset, so returns -ENOTSUPP) we can try to do
4978		 * a soft reset once we get the controller configured up to the
4979		 * point that it can accept a command.
4980		 */
4981		try_soft_reset = 1;
4982		rc = 0;
4983	}
4984
4985reinit_after_soft_reset:
4986
4987	i = alloc_cciss_hba(pdev);
4988	if (i < 0)
4989		return -1;
4990
4991	h = hba[i];
4992	h->pdev = pdev;
4993	h->busy_initializing = 1;
4994	h->intr_mode = cciss_simple_mode ? SIMPLE_MODE_INT : PERF_MODE_INT;
4995	INIT_LIST_HEAD(&h->cmpQ);
4996	INIT_LIST_HEAD(&h->reqQ);
4997	mutex_init(&h->busy_shutting_down);
4998
4999	if (cciss_pci_init(h) != 0)
5000		goto clean_no_release_regions;
5001
5002	sprintf(h->devname, "cciss%d", i);
5003	h->ctlr = i;
5004
5005	if (cciss_tape_cmds < 2)
5006		cciss_tape_cmds = 2;
5007	if (cciss_tape_cmds > 16)
5008		cciss_tape_cmds = 16;
5009
5010	init_completion(&h->scan_wait);
5011
5012	if (cciss_create_hba_sysfs_entry(h))
5013		goto clean0;
5014
5015	/* configure PCI DMA stuff */
5016	if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
5017		dac = 1;
5018	else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
5019		dac = 0;
5020	else {
5021		dev_err(&h->pdev->dev, "no suitable DMA available\n");
5022		goto clean1;
5023	}
5024
5025	/*
5026	 * register with the major number, or get a dynamic major number
5027	 * by passing 0 as argument.  This is done for greater than
5028	 * 8 controller support.
5029	 */
5030	if (i < MAX_CTLR_ORIG)
5031		h->major = COMPAQ_CISS_MAJOR + i;
5032	rc = register_blkdev(h->major, h->devname);
5033	if (rc == -EBUSY || rc == -EINVAL) {
5034		dev_err(&h->pdev->dev,
5035		       "Unable to get major number %d for %s "
5036		       "on hba %d\n", h->major, h->devname, i);
5037		goto clean1;
5038	} else {
5039		if (i >= MAX_CTLR_ORIG)
5040			h->major = rc;
5041	}
5042
5043	/* make sure the board interrupts are off */
5044	h->access.set_intr_mask(h, CCISS_INTR_OFF);
5045	rc = cciss_request_irq(h, do_cciss_msix_intr, do_cciss_intx);
5046	if (rc)
5047		goto clean2;
5048
5049	dev_info(&h->pdev->dev, "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
5050	       h->devname, pdev->device, pci_name(pdev),
5051	       h->intr[h->intr_mode], dac ? "" : " not");
5052
5053	if (cciss_allocate_cmd_pool(h))
5054		goto clean4;
5055
5056	if (cciss_allocate_scatterlists(h))
5057		goto clean4;
5058
5059	h->cmd_sg_list = cciss_allocate_sg_chain_blocks(h,
5060		h->chainsize, h->nr_cmds);
5061	if (!h->cmd_sg_list && h->chainsize > 0)
5062		goto clean4;
5063
5064	spin_lock_init(&h->lock);
5065
5066	/* Initialize the pdev driver private data.
5067	   have it point to h.  */
5068	pci_set_drvdata(pdev, h);
5069	/* command and error info recs zeroed out before
5070	   they are used */
5071	memset(h->cmd_pool_bits, 0,
5072	       DIV_ROUND_UP(h->nr_cmds, BITS_PER_LONG)
5073			* sizeof(unsigned long));
5074
5075	h->num_luns = 0;
5076	h->highest_lun = -1;
5077	for (j = 0; j < CISS_MAX_LUN; j++) {
5078		h->drv[j] = NULL;
5079		h->gendisk[j] = NULL;
5080	}
5081
5082	/* At this point, the controller is ready to take commands.
5083	 * Now, if reset_devices and the hard reset didn't work, try
5084	 * the soft reset and see if that works.
5085	 */
5086	if (try_soft_reset) {
5087
5088		/* This is kind of gross.  We may or may not get a completion
5089		 * from the soft reset command, and if we do, then the value
5090		 * from the fifo may or may not be valid.  So, we wait 10 secs
5091		 * after the reset throwing away any completions we get during
5092		 * that time.  Unregister the interrupt handler and register
5093		 * fake ones to scoop up any residual completions.
5094		 */
5095		spin_lock_irqsave(&h->lock, flags);
5096		h->access.set_intr_mask(h, CCISS_INTR_OFF);
5097		spin_unlock_irqrestore(&h->lock, flags);
5098		free_irq(h->intr[h->intr_mode], h);
5099		rc = cciss_request_irq(h, cciss_msix_discard_completions,
5100					cciss_intx_discard_completions);
5101		if (rc) {
5102			dev_warn(&h->pdev->dev, "Failed to request_irq after "
5103				"soft reset.\n");
5104			goto clean4;
5105		}
5106
5107		rc = cciss_kdump_soft_reset(h);
5108		if (rc) {
5109			dev_warn(&h->pdev->dev, "Soft reset failed.\n");
5110			goto clean4;
5111		}
5112
5113		dev_info(&h->pdev->dev, "Board READY.\n");
5114		dev_info(&h->pdev->dev,
5115			"Waiting for stale completions to drain.\n");
5116		h->access.set_intr_mask(h, CCISS_INTR_ON);
5117		msleep(10000);
5118		h->access.set_intr_mask(h, CCISS_INTR_OFF);
5119
5120		rc = controller_reset_failed(h->cfgtable);
5121		if (rc)
5122			dev_info(&h->pdev->dev,
5123				"Soft reset appears to have failed.\n");
5124
5125		/* since the controller's reset, we have to go back and re-init
5126		 * everything.  Easiest to just forget what we've done and do it
5127		 * all over again.
5128		 */
5129		cciss_undo_allocations_after_kdump_soft_reset(h);
5130		try_soft_reset = 0;
5131		if (rc)
5132			/* don't go to clean4, we already unallocated */
5133			return -ENODEV;
5134
5135		goto reinit_after_soft_reset;
5136	}
5137
5138	cciss_scsi_setup(h);
5139
5140	/* Turn the interrupts on so we can service requests */
5141	h->access.set_intr_mask(h, CCISS_INTR_ON);
5142
5143	/* Get the firmware version */
5144	inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
5145	if (inq_buff == NULL) {
5146		dev_err(&h->pdev->dev, "out of memory\n");
5147		goto clean4;
5148	}
5149
5150	return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
5151		sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
5152	if (return_code == IO_OK) {
5153		h->firm_ver[0] = inq_buff->data_byte[32];
5154		h->firm_ver[1] = inq_buff->data_byte[33];
5155		h->firm_ver[2] = inq_buff->data_byte[34];
5156		h->firm_ver[3] = inq_buff->data_byte[35];
5157	} else {	 /* send command failed */
5158		dev_warn(&h->pdev->dev, "unable to determine firmware"
5159			" version of controller\n");
5160	}
5161	kfree(inq_buff);
5162
5163	cciss_procinit(h);
5164
5165	h->cciss_max_sectors = 8192;
5166
5167	rebuild_lun_table(h, 1, 0);
5168	cciss_engage_scsi(h);
5169	h->busy_initializing = 0;
5170	return 1;
5171
5172clean4:
5173	cciss_free_cmd_pool(h);
5174	cciss_free_scatterlists(h);
5175	cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
5176	free_irq(h->intr[h->intr_mode], h);
5177clean2:
5178	unregister_blkdev(h->major, h->devname);
5179clean1:
5180	cciss_destroy_hba_sysfs_entry(h);
5181clean0:
5182	pci_release_regions(pdev);
5183clean_no_release_regions:
5184	h->busy_initializing = 0;
5185
5186	/*
5187	 * Deliberately omit pci_disable_device(): it does something nasty to
5188	 * Smart Array controllers that pci_enable_device does not undo
5189	 */
5190	pci_set_drvdata(pdev, NULL);
5191	free_hba(h);
5192	return -1;
5193}
5194
5195static void cciss_shutdown(struct pci_dev *pdev)
5196{
5197	ctlr_info_t *h;
5198	char *flush_buf;
5199	int return_code;
5200
5201	h = pci_get_drvdata(pdev);
5202	flush_buf = kzalloc(4, GFP_KERNEL);
5203	if (!flush_buf) {
5204		dev_warn(&h->pdev->dev, "cache not flushed, out of memory.\n");
5205		return;
5206	}
5207	/* write all data in the battery backed cache to disk */
5208	memset(flush_buf, 0, 4);
5209	return_code = sendcmd_withirq(h, CCISS_CACHE_FLUSH, flush_buf,
5210		4, 0, CTLR_LUNID, TYPE_CMD);
5211	kfree(flush_buf);
5212	if (return_code != IO_OK)
5213		dev_warn(&h->pdev->dev, "Error flushing cache\n");
5214	h->access.set_intr_mask(h, CCISS_INTR_OFF);
5215	free_irq(h->intr[h->intr_mode], h);
5216}
5217
5218static int __devinit cciss_enter_simple_mode(struct ctlr_info *h)
5219{
5220	u32 trans_support;
5221
5222	trans_support = readl(&(h->cfgtable->TransportSupport));
5223	if (!(trans_support & SIMPLE_MODE))
5224		return -ENOTSUPP;
5225
5226	h->max_commands = readl(&(h->cfgtable->CmdsOutMax));
5227	writel(CFGTBL_Trans_Simple, &(h->cfgtable->HostWrite.TransportRequest));
5228	writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
5229	cciss_wait_for_mode_change_ack(h);
5230	print_cfg_table(h);
5231	if (!(readl(&(h->cfgtable->TransportActive)) & CFGTBL_Trans_Simple)) {
5232		dev_warn(&h->pdev->dev, "unable to get board into simple mode\n");
5233		return -ENODEV;
5234	}
5235	h->transMethod = CFGTBL_Trans_Simple;
5236	return 0;
5237}
5238
5239
5240static void __devexit cciss_remove_one(struct pci_dev *pdev)
5241{
5242	ctlr_info_t *h;
5243	int i, j;
5244
5245	if (pci_get_drvdata(pdev) == NULL) {
5246		dev_err(&pdev->dev, "Unable to remove device\n");
5247		return;
5248	}
5249
5250	h = pci_get_drvdata(pdev);
5251	i = h->ctlr;
5252	if (hba[i] == NULL) {
5253		dev_err(&pdev->dev, "device appears to already be removed\n");
5254		return;
5255	}
5256
5257	mutex_lock(&h->busy_shutting_down);
5258
5259	remove_from_scan_list(h);
5260	remove_proc_entry(h->devname, proc_cciss);
5261	unregister_blkdev(h->major, h->devname);
5262
5263	/* remove it from the disk list */
5264	for (j = 0; j < CISS_MAX_LUN; j++) {
5265		struct gendisk *disk = h->gendisk[j];
5266		if (disk) {
5267			struct request_queue *q = disk->queue;
5268
5269			if (disk->flags & GENHD_FL_UP) {
5270				cciss_destroy_ld_sysfs_entry(h, j, 1);
5271				del_gendisk(disk);
5272			}
5273			if (q)
5274				blk_cleanup_queue(q);
5275		}
5276	}
5277
5278#ifdef CONFIG_CISS_SCSI_TAPE
5279	cciss_unregister_scsi(h);	/* unhook from SCSI subsystem */
5280#endif
5281
5282	cciss_shutdown(pdev);
5283
5284#ifdef CONFIG_PCI_MSI
5285	if (h->msix_vector)
5286		pci_disable_msix(h->pdev);
5287	else if (h->msi_vector)
5288		pci_disable_msi(h->pdev);
5289#endif				/* CONFIG_PCI_MSI */
5290
5291	iounmap(h->transtable);
5292	iounmap(h->cfgtable);
5293	iounmap(h->vaddr);
5294
5295	cciss_free_cmd_pool(h);
5296	/* Free up sg elements */
5297	for (j = 0; j < h->nr_cmds; j++)
5298		kfree(h->scatter_list[j]);
5299	kfree(h->scatter_list);
5300	cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
5301	kfree(h->blockFetchTable);
5302	if (h->reply_pool)
5303		pci_free_consistent(h->pdev, h->max_commands * sizeof(__u64),
5304				h->reply_pool, h->reply_pool_dhandle);
5305	/*
5306	 * Deliberately omit pci_disable_device(): it does something nasty to
5307	 * Smart Array controllers that pci_enable_device does not undo
5308	 */
5309	pci_release_regions(pdev);
5310	pci_set_drvdata(pdev, NULL);
5311	cciss_destroy_hba_sysfs_entry(h);
5312	mutex_unlock(&h->busy_shutting_down);
5313	free_hba(h);
5314}
5315
5316static struct pci_driver cciss_pci_driver = {
5317	.name = "cciss",
5318	.probe = cciss_init_one,
5319	.remove = __devexit_p(cciss_remove_one),
5320	.id_table = cciss_pci_device_id,	/* id_table */
5321	.shutdown = cciss_shutdown,
5322};
5323
5324/*
5325 *  This is it.  Register the PCI driver information for the cards we control
5326 *  the OS will call our registered routines when it finds one of our cards.
5327 */
5328static int __init cciss_init(void)
5329{
5330	int err;
5331
5332	/*
5333	 * The hardware requires that commands are aligned on a 64-bit
5334	 * boundary. Given that we use pci_alloc_consistent() to allocate an
5335	 * array of them, the size must be a multiple of 8 bytes.
5336	 */
5337	BUILD_BUG_ON(sizeof(CommandList_struct) % COMMANDLIST_ALIGNMENT);
5338	printk(KERN_INFO DRIVER_NAME "\n");
5339
5340	err = bus_register(&cciss_bus_type);
5341	if (err)
5342		return err;
5343
5344	/* Start the scan thread */
5345	cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
5346	if (IS_ERR(cciss_scan_thread)) {
5347		err = PTR_ERR(cciss_scan_thread);
5348		goto err_bus_unregister;
5349	}
5350
5351	/* Register for our PCI devices */
5352	err = pci_register_driver(&cciss_pci_driver);
5353	if (err)
5354		goto err_thread_stop;
5355
5356	return err;
5357
5358err_thread_stop:
5359	kthread_stop(cciss_scan_thread);
5360err_bus_unregister:
5361	bus_unregister(&cciss_bus_type);
5362
5363	return err;
5364}
5365
5366static void __exit cciss_cleanup(void)
5367{
5368	int i;
5369
5370	pci_unregister_driver(&cciss_pci_driver);
5371	/* double check that all controller entrys have been removed */
5372	for (i = 0; i < MAX_CTLR; i++) {
5373		if (hba[i] != NULL) {
5374			dev_warn(&hba[i]->pdev->dev,
5375				"had to remove controller\n");
5376			cciss_remove_one(hba[i]->pdev);
5377		}
5378	}
5379	kthread_stop(cciss_scan_thread);
5380	if (proc_cciss)
5381		remove_proc_entry("driver/cciss", NULL);
5382	bus_unregister(&cciss_bus_type);
5383}
5384
5385module_init(cciss_init);
5386module_exit(cciss_cleanup);
5387