NFS: switch NFSROOT to use new rpcbind client
[linux-2.6/kvm.git] / drivers / block / cciss.c
blob65a725cd3422ec565d9508de378d32cb9c52ffd5
1 /*
2 * Disk Array driver for HP SA 5xxx and 6xxx Controllers
3 * Copyright 2000, 2006 Hewlett-Packard Development Company, L.P.
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
13 * NON INFRINGEMENT. See the GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
19 * Questions/Comments/Bugfixes to iss_storagedev@hp.com
23 #include <linux/module.h>
24 #include <linux/interrupt.h>
25 #include <linux/types.h>
26 #include <linux/pci.h>
27 #include <linux/kernel.h>
28 #include <linux/slab.h>
29 #include <linux/delay.h>
30 #include <linux/major.h>
31 #include <linux/fs.h>
32 #include <linux/bio.h>
33 #include <linux/blkpg.h>
34 #include <linux/timer.h>
35 #include <linux/proc_fs.h>
36 #include <linux/init.h>
37 #include <linux/hdreg.h>
38 #include <linux/spinlock.h>
39 #include <linux/compat.h>
40 #include <linux/blktrace_api.h>
41 #include <asm/uaccess.h>
42 #include <asm/io.h>
44 #include <linux/dma-mapping.h>
45 #include <linux/blkdev.h>
46 #include <linux/genhd.h>
47 #include <linux/completion.h>
49 #define CCISS_DRIVER_VERSION(maj,min,submin) ((maj<<16)|(min<<8)|(submin))
50 #define DRIVER_NAME "HP CISS Driver (v 3.6.14)"
51 #define DRIVER_VERSION CCISS_DRIVER_VERSION(3,6,14)
53 /* Embedded module documentation macros - see modules.h */
54 MODULE_AUTHOR("Hewlett-Packard Company");
55 MODULE_DESCRIPTION("Driver for HP Controller SA5xxx SA6xxx version 3.6.14");
56 MODULE_SUPPORTED_DEVICE("HP SA5i SA5i+ SA532 SA5300 SA5312 SA641 SA642 SA6400"
57 " SA6i P600 P800 P400 P400i E200 E200i E500");
58 MODULE_VERSION("3.6.14");
59 MODULE_LICENSE("GPL");
61 #include "cciss_cmd.h"
62 #include "cciss.h"
63 #include <linux/cciss_ioctl.h>
65 /* define the PCI info for the cards we can control */
66 static const struct pci_device_id cciss_pci_device_id[] = {
67 {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISS, 0x0E11, 0x4070},
68 {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4080},
69 {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4082},
70 {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4083},
71 {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x4091},
72 {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409A},
73 {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409B},
74 {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409C},
75 {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409D},
76 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSA, 0x103C, 0x3225},
77 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSC, 0x103C, 0x3223},
78 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSC, 0x103C, 0x3234},
79 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSC, 0x103C, 0x3235},
80 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSD, 0x103C, 0x3211},
81 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSD, 0x103C, 0x3212},
82 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSD, 0x103C, 0x3213},
83 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSD, 0x103C, 0x3214},
84 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSD, 0x103C, 0x3215},
85 {PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSC, 0x103C, 0x3237},
86 {PCI_VENDOR_ID_HP, PCI_ANY_ID, PCI_ANY_ID, PCI_ANY_ID,
87 PCI_CLASS_STORAGE_RAID << 8, 0xffff << 8, 0},
88 {0,}
91 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
93 /* board_id = Subsystem Device ID & Vendor ID
94 * product = Marketing Name for the board
95 * access = Address of the struct of function pointers
96 * nr_cmds = Number of commands supported by controller
98 static struct board_type products[] = {
99 {0x40700E11, "Smart Array 5300", &SA5_access, 512},
100 {0x40800E11, "Smart Array 5i", &SA5B_access, 512},
101 {0x40820E11, "Smart Array 532", &SA5B_access, 512},
102 {0x40830E11, "Smart Array 5312", &SA5B_access, 512},
103 {0x409A0E11, "Smart Array 641", &SA5_access, 512},
104 {0x409B0E11, "Smart Array 642", &SA5_access, 512},
105 {0x409C0E11, "Smart Array 6400", &SA5_access, 512},
106 {0x409D0E11, "Smart Array 6400 EM", &SA5_access, 512},
107 {0x40910E11, "Smart Array 6i", &SA5_access, 512},
108 {0x3225103C, "Smart Array P600", &SA5_access, 512},
109 {0x3223103C, "Smart Array P800", &SA5_access, 512},
110 {0x3234103C, "Smart Array P400", &SA5_access, 512},
111 {0x3235103C, "Smart Array P400i", &SA5_access, 512},
112 {0x3211103C, "Smart Array E200i", &SA5_access, 120},
113 {0x3212103C, "Smart Array E200", &SA5_access, 120},
114 {0x3213103C, "Smart Array E200i", &SA5_access, 120},
115 {0x3214103C, "Smart Array E200i", &SA5_access, 120},
116 {0x3215103C, "Smart Array E200i", &SA5_access, 120},
117 {0x3237103C, "Smart Array E500", &SA5_access, 512},
118 {0xFFFF103C, "Unknown Smart Array", &SA5_access, 120},
121 /* How long to wait (in milliseconds) for board to go into simple mode */
122 #define MAX_CONFIG_WAIT 30000
123 #define MAX_IOCTL_CONFIG_WAIT 1000
125 /*define how many times we will try a command because of bus resets */
126 #define MAX_CMD_RETRIES 3
128 #define READ_AHEAD 1024
129 #define MAX_CTLR 32
131 /* Originally cciss driver only supports 8 major numbers */
132 #define MAX_CTLR_ORIG 8
134 static ctlr_info_t *hba[MAX_CTLR];
136 static void do_cciss_request(request_queue_t *q);
137 static irqreturn_t do_cciss_intr(int irq, void *dev_id);
138 static int cciss_open(struct inode *inode, struct file *filep);
139 static int cciss_release(struct inode *inode, struct file *filep);
140 static int cciss_ioctl(struct inode *inode, struct file *filep,
141 unsigned int cmd, unsigned long arg);
142 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
144 static int cciss_revalidate(struct gendisk *disk);
145 static int rebuild_lun_table(ctlr_info_t *h, struct gendisk *del_disk);
146 static int deregister_disk(struct gendisk *disk, drive_info_struct *drv,
147 int clear_all);
149 static void cciss_read_capacity(int ctlr, int logvol, int withirq,
150 sector_t *total_size, unsigned int *block_size);
151 static void cciss_read_capacity_16(int ctlr, int logvol, int withirq,
152 sector_t *total_size, unsigned int *block_size);
153 static void cciss_geometry_inquiry(int ctlr, int logvol,
154 int withirq, sector_t total_size,
155 unsigned int block_size, InquiryData_struct *inq_buff,
156 drive_info_struct *drv);
157 static void cciss_getgeometry(int cntl_num);
158 static void __devinit cciss_interrupt_mode(ctlr_info_t *, struct pci_dev *,
159 __u32);
160 static void start_io(ctlr_info_t *h);
161 static int sendcmd(__u8 cmd, int ctlr, void *buff, size_t size,
162 unsigned int use_unit_num, unsigned int log_unit,
163 __u8 page_code, unsigned char *scsi3addr, int cmd_type);
164 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
165 unsigned int use_unit_num, unsigned int log_unit,
166 __u8 page_code, int cmd_type);
168 static void fail_all_cmds(unsigned long ctlr);
170 #ifdef CONFIG_PROC_FS
171 static int cciss_proc_get_info(char *buffer, char **start, off_t offset,
172 int length, int *eof, void *data);
173 static void cciss_procinit(int i);
174 #else
175 static void cciss_procinit(int i)
178 #endif /* CONFIG_PROC_FS */
180 #ifdef CONFIG_COMPAT
181 static long cciss_compat_ioctl(struct file *f, unsigned cmd, unsigned long arg);
182 #endif
184 static struct block_device_operations cciss_fops = {
185 .owner = THIS_MODULE,
186 .open = cciss_open,
187 .release = cciss_release,
188 .ioctl = cciss_ioctl,
189 .getgeo = cciss_getgeo,
190 #ifdef CONFIG_COMPAT
191 .compat_ioctl = cciss_compat_ioctl,
192 #endif
193 .revalidate_disk = cciss_revalidate,
197 * Enqueuing and dequeuing functions for cmdlists.
199 static inline void addQ(CommandList_struct **Qptr, CommandList_struct *c)
201 if (*Qptr == NULL) {
202 *Qptr = c;
203 c->next = c->prev = c;
204 } else {
205 c->prev = (*Qptr)->prev;
206 c->next = (*Qptr);
207 (*Qptr)->prev->next = c;
208 (*Qptr)->prev = c;
212 static inline CommandList_struct *removeQ(CommandList_struct **Qptr,
213 CommandList_struct *c)
215 if (c && c->next != c) {
216 if (*Qptr == c)
217 *Qptr = c->next;
218 c->prev->next = c->next;
219 c->next->prev = c->prev;
220 } else {
221 *Qptr = NULL;
223 return c;
226 #include "cciss_scsi.c" /* For SCSI tape support */
228 #define RAID_UNKNOWN 6
230 #ifdef CONFIG_PROC_FS
233 * Report information about this controller.
235 #define ENG_GIG 1000000000
236 #define ENG_GIG_FACTOR (ENG_GIG/512)
237 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
238 "UNKNOWN"
241 static struct proc_dir_entry *proc_cciss;
243 static int cciss_proc_get_info(char *buffer, char **start, off_t offset,
244 int length, int *eof, void *data)
246 off_t pos = 0;
247 off_t len = 0;
248 int size, i, ctlr;
249 ctlr_info_t *h = (ctlr_info_t *) data;
250 drive_info_struct *drv;
251 unsigned long flags;
252 sector_t vol_sz, vol_sz_frac;
254 ctlr = h->ctlr;
256 /* prevent displaying bogus info during configuration
257 * or deconfiguration of a logical volume
259 spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
260 if (h->busy_configuring) {
261 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
262 return -EBUSY;
264 h->busy_configuring = 1;
265 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
267 size = sprintf(buffer, "%s: HP %s Controller\n"
268 "Board ID: 0x%08lx\n"
269 "Firmware Version: %c%c%c%c\n"
270 "IRQ: %d\n"
271 "Logical drives: %d\n"
272 "Max sectors: %d\n"
273 "Current Q depth: %d\n"
274 "Current # commands on controller: %d\n"
275 "Max Q depth since init: %d\n"
276 "Max # commands on controller since init: %d\n"
277 "Max SG entries since init: %d\n\n",
278 h->devname,
279 h->product_name,
280 (unsigned long)h->board_id,
281 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
282 h->firm_ver[3], (unsigned int)h->intr[SIMPLE_MODE_INT],
283 h->num_luns,
284 h->cciss_max_sectors,
285 h->Qdepth, h->commands_outstanding,
286 h->maxQsinceinit, h->max_outstanding, h->maxSG);
288 pos += size;
289 len += size;
290 cciss_proc_tape_report(ctlr, buffer, &pos, &len);
291 for (i = 0; i <= h->highest_lun; i++) {
293 drv = &h->drv[i];
294 if (drv->heads == 0)
295 continue;
297 vol_sz = drv->nr_blocks;
298 vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
299 vol_sz_frac *= 100;
300 sector_div(vol_sz_frac, ENG_GIG_FACTOR);
302 if (drv->raid_level > 5)
303 drv->raid_level = RAID_UNKNOWN;
304 size = sprintf(buffer + len, "cciss/c%dd%d:"
305 "\t%4u.%02uGB\tRAID %s\n",
306 ctlr, i, (int)vol_sz, (int)vol_sz_frac,
307 raid_label[drv->raid_level]);
308 pos += size;
309 len += size;
312 *eof = 1;
313 *start = buffer + offset;
314 len -= offset;
315 if (len > length)
316 len = length;
317 h->busy_configuring = 0;
318 return len;
321 static int
322 cciss_proc_write(struct file *file, const char __user *buffer,
323 unsigned long count, void *data)
325 unsigned char cmd[80];
326 int len;
327 #ifdef CONFIG_CISS_SCSI_TAPE
328 ctlr_info_t *h = (ctlr_info_t *) data;
329 int rc;
330 #endif
332 if (count > sizeof(cmd) - 1)
333 return -EINVAL;
334 if (copy_from_user(cmd, buffer, count))
335 return -EFAULT;
336 cmd[count] = '\0';
337 len = strlen(cmd); // above 3 lines ensure safety
338 if (len && cmd[len - 1] == '\n')
339 cmd[--len] = '\0';
340 # ifdef CONFIG_CISS_SCSI_TAPE
341 if (strcmp("engage scsi", cmd) == 0) {
342 rc = cciss_engage_scsi(h->ctlr);
343 if (rc != 0)
344 return -rc;
345 return count;
347 /* might be nice to have "disengage" too, but it's not
348 safely possible. (only 1 module use count, lock issues.) */
349 # endif
350 return -EINVAL;
354 * Get us a file in /proc/cciss that says something about each controller.
355 * Create /proc/cciss if it doesn't exist yet.
357 static void __devinit cciss_procinit(int i)
359 struct proc_dir_entry *pde;
361 if (proc_cciss == NULL) {
362 proc_cciss = proc_mkdir("cciss", proc_root_driver);
363 if (!proc_cciss)
364 return;
367 pde = create_proc_read_entry(hba[i]->devname,
368 S_IWUSR | S_IRUSR | S_IRGRP | S_IROTH,
369 proc_cciss, cciss_proc_get_info, hba[i]);
370 pde->write_proc = cciss_proc_write;
372 #endif /* CONFIG_PROC_FS */
375 * For operations that cannot sleep, a command block is allocated at init,
376 * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
377 * which ones are free or in use. For operations that can wait for kmalloc
378 * to possible sleep, this routine can be called with get_from_pool set to 0.
379 * cmd_free() MUST be called with a got_from_pool set to 0 if cmd_alloc was.
381 static CommandList_struct *cmd_alloc(ctlr_info_t *h, int get_from_pool)
383 CommandList_struct *c;
384 int i;
385 u64bit temp64;
386 dma_addr_t cmd_dma_handle, err_dma_handle;
388 if (!get_from_pool) {
389 c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
390 sizeof(CommandList_struct), &cmd_dma_handle);
391 if (c == NULL)
392 return NULL;
393 memset(c, 0, sizeof(CommandList_struct));
395 c->cmdindex = -1;
397 c->err_info = (ErrorInfo_struct *)
398 pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
399 &err_dma_handle);
401 if (c->err_info == NULL) {
402 pci_free_consistent(h->pdev,
403 sizeof(CommandList_struct), c, cmd_dma_handle);
404 return NULL;
406 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
407 } else { /* get it out of the controllers pool */
409 do {
410 i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
411 if (i == h->nr_cmds)
412 return NULL;
413 } while (test_and_set_bit
414 (i & (BITS_PER_LONG - 1),
415 h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
416 #ifdef CCISS_DEBUG
417 printk(KERN_DEBUG "cciss: using command buffer %d\n", i);
418 #endif
419 c = h->cmd_pool + i;
420 memset(c, 0, sizeof(CommandList_struct));
421 cmd_dma_handle = h->cmd_pool_dhandle
422 + i * sizeof(CommandList_struct);
423 c->err_info = h->errinfo_pool + i;
424 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
425 err_dma_handle = h->errinfo_pool_dhandle
426 + i * sizeof(ErrorInfo_struct);
427 h->nr_allocs++;
429 c->cmdindex = i;
432 c->busaddr = (__u32) cmd_dma_handle;
433 temp64.val = (__u64) err_dma_handle;
434 c->ErrDesc.Addr.lower = temp64.val32.lower;
435 c->ErrDesc.Addr.upper = temp64.val32.upper;
436 c->ErrDesc.Len = sizeof(ErrorInfo_struct);
438 c->ctlr = h->ctlr;
439 return c;
443 * Frees a command block that was previously allocated with cmd_alloc().
445 static void cmd_free(ctlr_info_t *h, CommandList_struct *c, int got_from_pool)
447 int i;
448 u64bit temp64;
450 if (!got_from_pool) {
451 temp64.val32.lower = c->ErrDesc.Addr.lower;
452 temp64.val32.upper = c->ErrDesc.Addr.upper;
453 pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
454 c->err_info, (dma_addr_t) temp64.val);
455 pci_free_consistent(h->pdev, sizeof(CommandList_struct),
456 c, (dma_addr_t) c->busaddr);
457 } else {
458 i = c - h->cmd_pool;
459 clear_bit(i & (BITS_PER_LONG - 1),
460 h->cmd_pool_bits + (i / BITS_PER_LONG));
461 h->nr_frees++;
465 static inline ctlr_info_t *get_host(struct gendisk *disk)
467 return disk->queue->queuedata;
470 static inline drive_info_struct *get_drv(struct gendisk *disk)
472 return disk->private_data;
476 * Open. Make sure the device is really there.
478 static int cciss_open(struct inode *inode, struct file *filep)
480 ctlr_info_t *host = get_host(inode->i_bdev->bd_disk);
481 drive_info_struct *drv = get_drv(inode->i_bdev->bd_disk);
483 #ifdef CCISS_DEBUG
484 printk(KERN_DEBUG "cciss_open %s\n", inode->i_bdev->bd_disk->disk_name);
485 #endif /* CCISS_DEBUG */
487 if (host->busy_initializing || drv->busy_configuring)
488 return -EBUSY;
490 * Root is allowed to open raw volume zero even if it's not configured
491 * so array config can still work. Root is also allowed to open any
492 * volume that has a LUN ID, so it can issue IOCTL to reread the
493 * disk information. I don't think I really like this
494 * but I'm already using way to many device nodes to claim another one
495 * for "raw controller".
497 if (drv->heads == 0) {
498 if (iminor(inode) != 0) { /* not node 0? */
499 /* if not node 0 make sure it is a partition = 0 */
500 if (iminor(inode) & 0x0f) {
501 return -ENXIO;
502 /* if it is, make sure we have a LUN ID */
503 } else if (drv->LunID == 0) {
504 return -ENXIO;
507 if (!capable(CAP_SYS_ADMIN))
508 return -EPERM;
510 drv->usage_count++;
511 host->usage_count++;
512 return 0;
516 * Close. Sync first.
518 static int cciss_release(struct inode *inode, struct file *filep)
520 ctlr_info_t *host = get_host(inode->i_bdev->bd_disk);
521 drive_info_struct *drv = get_drv(inode->i_bdev->bd_disk);
523 #ifdef CCISS_DEBUG
524 printk(KERN_DEBUG "cciss_release %s\n",
525 inode->i_bdev->bd_disk->disk_name);
526 #endif /* CCISS_DEBUG */
528 drv->usage_count--;
529 host->usage_count--;
530 return 0;
533 #ifdef CONFIG_COMPAT
535 static int do_ioctl(struct file *f, unsigned cmd, unsigned long arg)
537 int ret;
538 lock_kernel();
539 ret = cciss_ioctl(f->f_path.dentry->d_inode, f, cmd, arg);
540 unlock_kernel();
541 return ret;
544 static int cciss_ioctl32_passthru(struct file *f, unsigned cmd,
545 unsigned long arg);
546 static int cciss_ioctl32_big_passthru(struct file *f, unsigned cmd,
547 unsigned long arg);
549 static long cciss_compat_ioctl(struct file *f, unsigned cmd, unsigned long arg)
551 switch (cmd) {
552 case CCISS_GETPCIINFO:
553 case CCISS_GETINTINFO:
554 case CCISS_SETINTINFO:
555 case CCISS_GETNODENAME:
556 case CCISS_SETNODENAME:
557 case CCISS_GETHEARTBEAT:
558 case CCISS_GETBUSTYPES:
559 case CCISS_GETFIRMVER:
560 case CCISS_GETDRIVVER:
561 case CCISS_REVALIDVOLS:
562 case CCISS_DEREGDISK:
563 case CCISS_REGNEWDISK:
564 case CCISS_REGNEWD:
565 case CCISS_RESCANDISK:
566 case CCISS_GETLUNINFO:
567 return do_ioctl(f, cmd, arg);
569 case CCISS_PASSTHRU32:
570 return cciss_ioctl32_passthru(f, cmd, arg);
571 case CCISS_BIG_PASSTHRU32:
572 return cciss_ioctl32_big_passthru(f, cmd, arg);
574 default:
575 return -ENOIOCTLCMD;
579 static int cciss_ioctl32_passthru(struct file *f, unsigned cmd,
580 unsigned long arg)
582 IOCTL32_Command_struct __user *arg32 =
583 (IOCTL32_Command_struct __user *) arg;
584 IOCTL_Command_struct arg64;
585 IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
586 int err;
587 u32 cp;
589 err = 0;
590 err |=
591 copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
592 sizeof(arg64.LUN_info));
593 err |=
594 copy_from_user(&arg64.Request, &arg32->Request,
595 sizeof(arg64.Request));
596 err |=
597 copy_from_user(&arg64.error_info, &arg32->error_info,
598 sizeof(arg64.error_info));
599 err |= get_user(arg64.buf_size, &arg32->buf_size);
600 err |= get_user(cp, &arg32->buf);
601 arg64.buf = compat_ptr(cp);
602 err |= copy_to_user(p, &arg64, sizeof(arg64));
604 if (err)
605 return -EFAULT;
607 err = do_ioctl(f, CCISS_PASSTHRU, (unsigned long)p);
608 if (err)
609 return err;
610 err |=
611 copy_in_user(&arg32->error_info, &p->error_info,
612 sizeof(arg32->error_info));
613 if (err)
614 return -EFAULT;
615 return err;
618 static int cciss_ioctl32_big_passthru(struct file *file, unsigned cmd,
619 unsigned long arg)
621 BIG_IOCTL32_Command_struct __user *arg32 =
622 (BIG_IOCTL32_Command_struct __user *) arg;
623 BIG_IOCTL_Command_struct arg64;
624 BIG_IOCTL_Command_struct __user *p =
625 compat_alloc_user_space(sizeof(arg64));
626 int err;
627 u32 cp;
629 err = 0;
630 err |=
631 copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
632 sizeof(arg64.LUN_info));
633 err |=
634 copy_from_user(&arg64.Request, &arg32->Request,
635 sizeof(arg64.Request));
636 err |=
637 copy_from_user(&arg64.error_info, &arg32->error_info,
638 sizeof(arg64.error_info));
639 err |= get_user(arg64.buf_size, &arg32->buf_size);
640 err |= get_user(arg64.malloc_size, &arg32->malloc_size);
641 err |= get_user(cp, &arg32->buf);
642 arg64.buf = compat_ptr(cp);
643 err |= copy_to_user(p, &arg64, sizeof(arg64));
645 if (err)
646 return -EFAULT;
648 err = do_ioctl(file, CCISS_BIG_PASSTHRU, (unsigned long)p);
649 if (err)
650 return err;
651 err |=
652 copy_in_user(&arg32->error_info, &p->error_info,
653 sizeof(arg32->error_info));
654 if (err)
655 return -EFAULT;
656 return err;
658 #endif
660 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
662 drive_info_struct *drv = get_drv(bdev->bd_disk);
664 if (!drv->cylinders)
665 return -ENXIO;
667 geo->heads = drv->heads;
668 geo->sectors = drv->sectors;
669 geo->cylinders = drv->cylinders;
670 return 0;
674 * ioctl
676 static int cciss_ioctl(struct inode *inode, struct file *filep,
677 unsigned int cmd, unsigned long arg)
679 struct block_device *bdev = inode->i_bdev;
680 struct gendisk *disk = bdev->bd_disk;
681 ctlr_info_t *host = get_host(disk);
682 drive_info_struct *drv = get_drv(disk);
683 int ctlr = host->ctlr;
684 void __user *argp = (void __user *)arg;
686 #ifdef CCISS_DEBUG
687 printk(KERN_DEBUG "cciss_ioctl: Called with cmd=%x %lx\n", cmd, arg);
688 #endif /* CCISS_DEBUG */
690 switch (cmd) {
691 case CCISS_GETPCIINFO:
693 cciss_pci_info_struct pciinfo;
695 if (!arg)
696 return -EINVAL;
697 pciinfo.domain = pci_domain_nr(host->pdev->bus);
698 pciinfo.bus = host->pdev->bus->number;
699 pciinfo.dev_fn = host->pdev->devfn;
700 pciinfo.board_id = host->board_id;
701 if (copy_to_user
702 (argp, &pciinfo, sizeof(cciss_pci_info_struct)))
703 return -EFAULT;
704 return 0;
706 case CCISS_GETINTINFO:
708 cciss_coalint_struct intinfo;
709 if (!arg)
710 return -EINVAL;
711 intinfo.delay =
712 readl(&host->cfgtable->HostWrite.CoalIntDelay);
713 intinfo.count =
714 readl(&host->cfgtable->HostWrite.CoalIntCount);
715 if (copy_to_user
716 (argp, &intinfo, sizeof(cciss_coalint_struct)))
717 return -EFAULT;
718 return 0;
720 case CCISS_SETINTINFO:
722 cciss_coalint_struct intinfo;
723 unsigned long flags;
724 int i;
726 if (!arg)
727 return -EINVAL;
728 if (!capable(CAP_SYS_ADMIN))
729 return -EPERM;
730 if (copy_from_user
731 (&intinfo, argp, sizeof(cciss_coalint_struct)))
732 return -EFAULT;
733 if ((intinfo.delay == 0) && (intinfo.count == 0))
735 // printk("cciss_ioctl: delay and count cannot be 0\n");
736 return -EINVAL;
738 spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
739 /* Update the field, and then ring the doorbell */
740 writel(intinfo.delay,
741 &(host->cfgtable->HostWrite.CoalIntDelay));
742 writel(intinfo.count,
743 &(host->cfgtable->HostWrite.CoalIntCount));
744 writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
746 for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
747 if (!(readl(host->vaddr + SA5_DOORBELL)
748 & CFGTBL_ChangeReq))
749 break;
750 /* delay and try again */
751 udelay(1000);
753 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
754 if (i >= MAX_IOCTL_CONFIG_WAIT)
755 return -EAGAIN;
756 return 0;
758 case CCISS_GETNODENAME:
760 NodeName_type NodeName;
761 int i;
763 if (!arg)
764 return -EINVAL;
765 for (i = 0; i < 16; i++)
766 NodeName[i] =
767 readb(&host->cfgtable->ServerName[i]);
768 if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
769 return -EFAULT;
770 return 0;
772 case CCISS_SETNODENAME:
774 NodeName_type NodeName;
775 unsigned long flags;
776 int i;
778 if (!arg)
779 return -EINVAL;
780 if (!capable(CAP_SYS_ADMIN))
781 return -EPERM;
783 if (copy_from_user
784 (NodeName, argp, sizeof(NodeName_type)))
785 return -EFAULT;
787 spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
789 /* Update the field, and then ring the doorbell */
790 for (i = 0; i < 16; i++)
791 writeb(NodeName[i],
792 &host->cfgtable->ServerName[i]);
794 writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
796 for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
797 if (!(readl(host->vaddr + SA5_DOORBELL)
798 & CFGTBL_ChangeReq))
799 break;
800 /* delay and try again */
801 udelay(1000);
803 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
804 if (i >= MAX_IOCTL_CONFIG_WAIT)
805 return -EAGAIN;
806 return 0;
809 case CCISS_GETHEARTBEAT:
811 Heartbeat_type heartbeat;
813 if (!arg)
814 return -EINVAL;
815 heartbeat = readl(&host->cfgtable->HeartBeat);
816 if (copy_to_user
817 (argp, &heartbeat, sizeof(Heartbeat_type)))
818 return -EFAULT;
819 return 0;
821 case CCISS_GETBUSTYPES:
823 BusTypes_type BusTypes;
825 if (!arg)
826 return -EINVAL;
827 BusTypes = readl(&host->cfgtable->BusTypes);
828 if (copy_to_user
829 (argp, &BusTypes, sizeof(BusTypes_type)))
830 return -EFAULT;
831 return 0;
833 case CCISS_GETFIRMVER:
835 FirmwareVer_type firmware;
837 if (!arg)
838 return -EINVAL;
839 memcpy(firmware, host->firm_ver, 4);
841 if (copy_to_user
842 (argp, firmware, sizeof(FirmwareVer_type)))
843 return -EFAULT;
844 return 0;
846 case CCISS_GETDRIVVER:
848 DriverVer_type DriverVer = DRIVER_VERSION;
850 if (!arg)
851 return -EINVAL;
853 if (copy_to_user
854 (argp, &DriverVer, sizeof(DriverVer_type)))
855 return -EFAULT;
856 return 0;
859 case CCISS_REVALIDVOLS:
860 return rebuild_lun_table(host, NULL);
862 case CCISS_GETLUNINFO:{
863 LogvolInfo_struct luninfo;
865 luninfo.LunID = drv->LunID;
866 luninfo.num_opens = drv->usage_count;
867 luninfo.num_parts = 0;
868 if (copy_to_user(argp, &luninfo,
869 sizeof(LogvolInfo_struct)))
870 return -EFAULT;
871 return 0;
873 case CCISS_DEREGDISK:
874 return rebuild_lun_table(host, disk);
876 case CCISS_REGNEWD:
877 return rebuild_lun_table(host, NULL);
879 case CCISS_PASSTHRU:
881 IOCTL_Command_struct iocommand;
882 CommandList_struct *c;
883 char *buff = NULL;
884 u64bit temp64;
885 unsigned long flags;
886 DECLARE_COMPLETION_ONSTACK(wait);
888 if (!arg)
889 return -EINVAL;
891 if (!capable(CAP_SYS_RAWIO))
892 return -EPERM;
894 if (copy_from_user
895 (&iocommand, argp, sizeof(IOCTL_Command_struct)))
896 return -EFAULT;
897 if ((iocommand.buf_size < 1) &&
898 (iocommand.Request.Type.Direction != XFER_NONE)) {
899 return -EINVAL;
901 #if 0 /* 'buf_size' member is 16-bits, and always smaller than kmalloc limit */
902 /* Check kmalloc limits */
903 if (iocommand.buf_size > 128000)
904 return -EINVAL;
905 #endif
906 if (iocommand.buf_size > 0) {
907 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
908 if (buff == NULL)
909 return -EFAULT;
911 if (iocommand.Request.Type.Direction == XFER_WRITE) {
912 /* Copy the data into the buffer we created */
913 if (copy_from_user
914 (buff, iocommand.buf, iocommand.buf_size)) {
915 kfree(buff);
916 return -EFAULT;
918 } else {
919 memset(buff, 0, iocommand.buf_size);
921 if ((c = cmd_alloc(host, 0)) == NULL) {
922 kfree(buff);
923 return -ENOMEM;
925 // Fill in the command type
926 c->cmd_type = CMD_IOCTL_PEND;
927 // Fill in Command Header
928 c->Header.ReplyQueue = 0; // unused in simple mode
929 if (iocommand.buf_size > 0) // buffer to fill
931 c->Header.SGList = 1;
932 c->Header.SGTotal = 1;
933 } else // no buffers to fill
935 c->Header.SGList = 0;
936 c->Header.SGTotal = 0;
938 c->Header.LUN = iocommand.LUN_info;
939 c->Header.Tag.lower = c->busaddr; // use the kernel address the cmd block for tag
941 // Fill in Request block
942 c->Request = iocommand.Request;
944 // Fill in the scatter gather information
945 if (iocommand.buf_size > 0) {
946 temp64.val = pci_map_single(host->pdev, buff,
947 iocommand.buf_size,
948 PCI_DMA_BIDIRECTIONAL);
949 c->SG[0].Addr.lower = temp64.val32.lower;
950 c->SG[0].Addr.upper = temp64.val32.upper;
951 c->SG[0].Len = iocommand.buf_size;
952 c->SG[0].Ext = 0; // we are not chaining
954 c->waiting = &wait;
956 /* Put the request on the tail of the request queue */
957 spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
958 addQ(&host->reqQ, c);
959 host->Qdepth++;
960 start_io(host);
961 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
963 wait_for_completion(&wait);
965 /* unlock the buffers from DMA */
966 temp64.val32.lower = c->SG[0].Addr.lower;
967 temp64.val32.upper = c->SG[0].Addr.upper;
968 pci_unmap_single(host->pdev, (dma_addr_t) temp64.val,
969 iocommand.buf_size,
970 PCI_DMA_BIDIRECTIONAL);
972 /* Copy the error information out */
973 iocommand.error_info = *(c->err_info);
974 if (copy_to_user
975 (argp, &iocommand, sizeof(IOCTL_Command_struct))) {
976 kfree(buff);
977 cmd_free(host, c, 0);
978 return -EFAULT;
981 if (iocommand.Request.Type.Direction == XFER_READ) {
982 /* Copy the data out of the buffer we created */
983 if (copy_to_user
984 (iocommand.buf, buff, iocommand.buf_size)) {
985 kfree(buff);
986 cmd_free(host, c, 0);
987 return -EFAULT;
990 kfree(buff);
991 cmd_free(host, c, 0);
992 return 0;
994 case CCISS_BIG_PASSTHRU:{
995 BIG_IOCTL_Command_struct *ioc;
996 CommandList_struct *c;
997 unsigned char **buff = NULL;
998 int *buff_size = NULL;
999 u64bit temp64;
1000 unsigned long flags;
1001 BYTE sg_used = 0;
1002 int status = 0;
1003 int i;
1004 DECLARE_COMPLETION_ONSTACK(wait);
1005 __u32 left;
1006 __u32 sz;
1007 BYTE __user *data_ptr;
1009 if (!arg)
1010 return -EINVAL;
1011 if (!capable(CAP_SYS_RAWIO))
1012 return -EPERM;
1013 ioc = (BIG_IOCTL_Command_struct *)
1014 kmalloc(sizeof(*ioc), GFP_KERNEL);
1015 if (!ioc) {
1016 status = -ENOMEM;
1017 goto cleanup1;
1019 if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1020 status = -EFAULT;
1021 goto cleanup1;
1023 if ((ioc->buf_size < 1) &&
1024 (ioc->Request.Type.Direction != XFER_NONE)) {
1025 status = -EINVAL;
1026 goto cleanup1;
1028 /* Check kmalloc limits using all SGs */
1029 if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1030 status = -EINVAL;
1031 goto cleanup1;
1033 if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1034 status = -EINVAL;
1035 goto cleanup1;
1037 buff =
1038 kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1039 if (!buff) {
1040 status = -ENOMEM;
1041 goto cleanup1;
1043 buff_size = kmalloc(MAXSGENTRIES * sizeof(int),
1044 GFP_KERNEL);
1045 if (!buff_size) {
1046 status = -ENOMEM;
1047 goto cleanup1;
1049 left = ioc->buf_size;
1050 data_ptr = ioc->buf;
1051 while (left) {
1052 sz = (left >
1053 ioc->malloc_size) ? ioc->
1054 malloc_size : left;
1055 buff_size[sg_used] = sz;
1056 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1057 if (buff[sg_used] == NULL) {
1058 status = -ENOMEM;
1059 goto cleanup1;
1061 if (ioc->Request.Type.Direction == XFER_WRITE) {
1062 if (copy_from_user
1063 (buff[sg_used], data_ptr, sz)) {
1064 status = -ENOMEM;
1065 goto cleanup1;
1067 } else {
1068 memset(buff[sg_used], 0, sz);
1070 left -= sz;
1071 data_ptr += sz;
1072 sg_used++;
1074 if ((c = cmd_alloc(host, 0)) == NULL) {
1075 status = -ENOMEM;
1076 goto cleanup1;
1078 c->cmd_type = CMD_IOCTL_PEND;
1079 c->Header.ReplyQueue = 0;
1081 if (ioc->buf_size > 0) {
1082 c->Header.SGList = sg_used;
1083 c->Header.SGTotal = sg_used;
1084 } else {
1085 c->Header.SGList = 0;
1086 c->Header.SGTotal = 0;
1088 c->Header.LUN = ioc->LUN_info;
1089 c->Header.Tag.lower = c->busaddr;
1091 c->Request = ioc->Request;
1092 if (ioc->buf_size > 0) {
1093 int i;
1094 for (i = 0; i < sg_used; i++) {
1095 temp64.val =
1096 pci_map_single(host->pdev, buff[i],
1097 buff_size[i],
1098 PCI_DMA_BIDIRECTIONAL);
1099 c->SG[i].Addr.lower =
1100 temp64.val32.lower;
1101 c->SG[i].Addr.upper =
1102 temp64.val32.upper;
1103 c->SG[i].Len = buff_size[i];
1104 c->SG[i].Ext = 0; /* we are not chaining */
1107 c->waiting = &wait;
1108 /* Put the request on the tail of the request queue */
1109 spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1110 addQ(&host->reqQ, c);
1111 host->Qdepth++;
1112 start_io(host);
1113 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1114 wait_for_completion(&wait);
1115 /* unlock the buffers from DMA */
1116 for (i = 0; i < sg_used; i++) {
1117 temp64.val32.lower = c->SG[i].Addr.lower;
1118 temp64.val32.upper = c->SG[i].Addr.upper;
1119 pci_unmap_single(host->pdev,
1120 (dma_addr_t) temp64.val, buff_size[i],
1121 PCI_DMA_BIDIRECTIONAL);
1123 /* Copy the error information out */
1124 ioc->error_info = *(c->err_info);
1125 if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1126 cmd_free(host, c, 0);
1127 status = -EFAULT;
1128 goto cleanup1;
1130 if (ioc->Request.Type.Direction == XFER_READ) {
1131 /* Copy the data out of the buffer we created */
1132 BYTE __user *ptr = ioc->buf;
1133 for (i = 0; i < sg_used; i++) {
1134 if (copy_to_user
1135 (ptr, buff[i], buff_size[i])) {
1136 cmd_free(host, c, 0);
1137 status = -EFAULT;
1138 goto cleanup1;
1140 ptr += buff_size[i];
1143 cmd_free(host, c, 0);
1144 status = 0;
1145 cleanup1:
1146 if (buff) {
1147 for (i = 0; i < sg_used; i++)
1148 kfree(buff[i]);
1149 kfree(buff);
1151 kfree(buff_size);
1152 kfree(ioc);
1153 return status;
1155 default:
1156 return -ENOTTY;
1160 static inline void complete_buffers(struct bio *bio, int status)
1162 while (bio) {
1163 struct bio *xbh = bio->bi_next;
1164 int nr_sectors = bio_sectors(bio);
1166 bio->bi_next = NULL;
1167 bio_endio(bio, nr_sectors << 9, status ? 0 : -EIO);
1168 bio = xbh;
1172 static void cciss_check_queues(ctlr_info_t *h)
1174 int start_queue = h->next_to_run;
1175 int i;
1177 /* check to see if we have maxed out the number of commands that can
1178 * be placed on the queue. If so then exit. We do this check here
1179 * in case the interrupt we serviced was from an ioctl and did not
1180 * free any new commands.
1182 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1183 return;
1185 /* We have room on the queue for more commands. Now we need to queue
1186 * them up. We will also keep track of the next queue to run so
1187 * that every queue gets a chance to be started first.
1189 for (i = 0; i < h->highest_lun + 1; i++) {
1190 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1191 /* make sure the disk has been added and the drive is real
1192 * because this can be called from the middle of init_one.
1194 if (!(h->drv[curr_queue].queue) || !(h->drv[curr_queue].heads))
1195 continue;
1196 blk_start_queue(h->gendisk[curr_queue]->queue);
1198 /* check to see if we have maxed out the number of commands
1199 * that can be placed on the queue.
1201 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1202 if (curr_queue == start_queue) {
1203 h->next_to_run =
1204 (start_queue + 1) % (h->highest_lun + 1);
1205 break;
1206 } else {
1207 h->next_to_run = curr_queue;
1208 break;
1210 } else {
1211 curr_queue = (curr_queue + 1) % (h->highest_lun + 1);
1216 static void cciss_softirq_done(struct request *rq)
1218 CommandList_struct *cmd = rq->completion_data;
1219 ctlr_info_t *h = hba[cmd->ctlr];
1220 unsigned long flags;
1221 u64bit temp64;
1222 int i, ddir;
1224 if (cmd->Request.Type.Direction == XFER_READ)
1225 ddir = PCI_DMA_FROMDEVICE;
1226 else
1227 ddir = PCI_DMA_TODEVICE;
1229 /* command did not need to be retried */
1230 /* unmap the DMA mapping for all the scatter gather elements */
1231 for (i = 0; i < cmd->Header.SGList; i++) {
1232 temp64.val32.lower = cmd->SG[i].Addr.lower;
1233 temp64.val32.upper = cmd->SG[i].Addr.upper;
1234 pci_unmap_page(h->pdev, temp64.val, cmd->SG[i].Len, ddir);
1237 complete_buffers(rq->bio, rq->errors);
1239 if (blk_fs_request(rq)) {
1240 const int rw = rq_data_dir(rq);
1242 disk_stat_add(rq->rq_disk, sectors[rw], rq->nr_sectors);
1245 #ifdef CCISS_DEBUG
1246 printk("Done with %p\n", rq);
1247 #endif /* CCISS_DEBUG */
1249 add_disk_randomness(rq->rq_disk);
1250 spin_lock_irqsave(&h->lock, flags);
1251 end_that_request_last(rq, rq->errors);
1252 cmd_free(h, cmd, 1);
1253 cciss_check_queues(h);
1254 spin_unlock_irqrestore(&h->lock, flags);
1257 /* This function will check the usage_count of the drive to be updated/added.
1258 * If the usage_count is zero then the drive information will be updated and
1259 * the disk will be re-registered with the kernel. If not then it will be
1260 * left alone for the next reboot. The exception to this is disk 0 which
1261 * will always be left registered with the kernel since it is also the
1262 * controller node. Any changes to disk 0 will show up on the next
1263 * reboot.
1265 static void cciss_update_drive_info(int ctlr, int drv_index)
1267 ctlr_info_t *h = hba[ctlr];
1268 struct gendisk *disk;
1269 InquiryData_struct *inq_buff = NULL;
1270 unsigned int block_size;
1271 sector_t total_size;
1272 unsigned long flags = 0;
1273 int ret = 0;
1275 /* if the disk already exists then deregister it before proceeding */
1276 if (h->drv[drv_index].raid_level != -1) {
1277 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1278 h->drv[drv_index].busy_configuring = 1;
1279 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1280 ret = deregister_disk(h->gendisk[drv_index],
1281 &h->drv[drv_index], 0);
1282 h->drv[drv_index].busy_configuring = 0;
1285 /* If the disk is in use return */
1286 if (ret)
1287 return;
1289 /* Get information about the disk and modify the driver structure */
1290 inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1291 if (inq_buff == NULL)
1292 goto mem_msg;
1294 /* testing to see if 16-byte CDBs are already being used */
1295 if (h->cciss_read == CCISS_READ_16) {
1296 cciss_read_capacity_16(h->ctlr, drv_index, 1,
1297 &total_size, &block_size);
1298 goto geo_inq;
1301 cciss_read_capacity(ctlr, drv_index, 1,
1302 &total_size, &block_size);
1304 /* if read_capacity returns all F's this volume is >2TB in size */
1305 /* so we switch to 16-byte CDB's for all read/write ops */
1306 if (total_size == 0xFFFFFFFFULL) {
1307 cciss_read_capacity_16(ctlr, drv_index, 1,
1308 &total_size, &block_size);
1309 h->cciss_read = CCISS_READ_16;
1310 h->cciss_write = CCISS_WRITE_16;
1311 } else {
1312 h->cciss_read = CCISS_READ_10;
1313 h->cciss_write = CCISS_WRITE_10;
1315 geo_inq:
1316 cciss_geometry_inquiry(ctlr, drv_index, 1, total_size, block_size,
1317 inq_buff, &h->drv[drv_index]);
1319 ++h->num_luns;
1320 disk = h->gendisk[drv_index];
1321 set_capacity(disk, h->drv[drv_index].nr_blocks);
1323 /* if it's the controller it's already added */
1324 if (drv_index) {
1325 disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1326 sprintf(disk->disk_name, "cciss/c%dd%d", ctlr, drv_index);
1327 disk->major = h->major;
1328 disk->first_minor = drv_index << NWD_SHIFT;
1329 disk->fops = &cciss_fops;
1330 disk->private_data = &h->drv[drv_index];
1332 /* Set up queue information */
1333 disk->queue->backing_dev_info.ra_pages = READ_AHEAD;
1334 blk_queue_bounce_limit(disk->queue, hba[ctlr]->pdev->dma_mask);
1336 /* This is a hardware imposed limit. */
1337 blk_queue_max_hw_segments(disk->queue, MAXSGENTRIES);
1339 /* This is a limit in the driver and could be eliminated. */
1340 blk_queue_max_phys_segments(disk->queue, MAXSGENTRIES);
1342 blk_queue_max_sectors(disk->queue, h->cciss_max_sectors);
1344 blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1346 disk->queue->queuedata = hba[ctlr];
1348 blk_queue_hardsect_size(disk->queue,
1349 hba[ctlr]->drv[drv_index].block_size);
1351 h->drv[drv_index].queue = disk->queue;
1352 add_disk(disk);
1355 freeret:
1356 kfree(inq_buff);
1357 return;
1358 mem_msg:
1359 printk(KERN_ERR "cciss: out of memory\n");
1360 goto freeret;
1363 /* This function will find the first index of the controllers drive array
1364 * that has a -1 for the raid_level and will return that index. This is
1365 * where new drives will be added. If the index to be returned is greater
1366 * than the highest_lun index for the controller then highest_lun is set
1367 * to this new index. If there are no available indexes then -1 is returned.
1369 static int cciss_find_free_drive_index(int ctlr)
1371 int i;
1373 for (i = 0; i < CISS_MAX_LUN; i++) {
1374 if (hba[ctlr]->drv[i].raid_level == -1) {
1375 if (i > hba[ctlr]->highest_lun)
1376 hba[ctlr]->highest_lun = i;
1377 return i;
1380 return -1;
1383 /* This function will add and remove logical drives from the Logical
1384 * drive array of the controller and maintain persistency of ordering
1385 * so that mount points are preserved until the next reboot. This allows
1386 * for the removal of logical drives in the middle of the drive array
1387 * without a re-ordering of those drives.
1388 * INPUT
1389 * h = The controller to perform the operations on
1390 * del_disk = The disk to remove if specified. If the value given
1391 * is NULL then no disk is removed.
1393 static int rebuild_lun_table(ctlr_info_t *h, struct gendisk *del_disk)
1395 int ctlr = h->ctlr;
1396 int num_luns;
1397 ReportLunData_struct *ld_buff = NULL;
1398 drive_info_struct *drv = NULL;
1399 int return_code;
1400 int listlength = 0;
1401 int i;
1402 int drv_found;
1403 int drv_index = 0;
1404 __u32 lunid = 0;
1405 unsigned long flags;
1407 /* Set busy_configuring flag for this operation */
1408 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1409 if (h->busy_configuring) {
1410 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1411 return -EBUSY;
1413 h->busy_configuring = 1;
1415 /* if del_disk is NULL then we are being called to add a new disk
1416 * and update the logical drive table. If it is not NULL then
1417 * we will check if the disk is in use or not.
1419 if (del_disk != NULL) {
1420 drv = get_drv(del_disk);
1421 drv->busy_configuring = 1;
1422 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1423 return_code = deregister_disk(del_disk, drv, 1);
1424 drv->busy_configuring = 0;
1425 h->busy_configuring = 0;
1426 return return_code;
1427 } else {
1428 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1429 if (!capable(CAP_SYS_RAWIO))
1430 return -EPERM;
1432 ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
1433 if (ld_buff == NULL)
1434 goto mem_msg;
1436 return_code = sendcmd_withirq(CISS_REPORT_LOG, ctlr, ld_buff,
1437 sizeof(ReportLunData_struct), 0,
1438 0, 0, TYPE_CMD);
1440 if (return_code == IO_OK) {
1441 listlength =
1442 be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
1443 } else { /* reading number of logical volumes failed */
1444 printk(KERN_WARNING "cciss: report logical volume"
1445 " command failed\n");
1446 listlength = 0;
1447 goto freeret;
1450 num_luns = listlength / 8; /* 8 bytes per entry */
1451 if (num_luns > CISS_MAX_LUN) {
1452 num_luns = CISS_MAX_LUN;
1453 printk(KERN_WARNING "cciss: more luns configured"
1454 " on controller than can be handled by"
1455 " this driver.\n");
1458 /* Compare controller drive array to drivers drive array.
1459 * Check for updates in the drive information and any new drives
1460 * on the controller.
1462 for (i = 0; i < num_luns; i++) {
1463 int j;
1465 drv_found = 0;
1467 lunid = (0xff &
1468 (unsigned int)(ld_buff->LUN[i][3])) << 24;
1469 lunid |= (0xff &
1470 (unsigned int)(ld_buff->LUN[i][2])) << 16;
1471 lunid |= (0xff &
1472 (unsigned int)(ld_buff->LUN[i][1])) << 8;
1473 lunid |= 0xff & (unsigned int)(ld_buff->LUN[i][0]);
1475 /* Find if the LUN is already in the drive array
1476 * of the controller. If so then update its info
1477 * if not is use. If it does not exist then find
1478 * the first free index and add it.
1480 for (j = 0; j <= h->highest_lun; j++) {
1481 if (h->drv[j].LunID == lunid) {
1482 drv_index = j;
1483 drv_found = 1;
1487 /* check if the drive was found already in the array */
1488 if (!drv_found) {
1489 drv_index = cciss_find_free_drive_index(ctlr);
1490 if (drv_index == -1)
1491 goto freeret;
1493 /*Check if the gendisk needs to be allocated */
1494 if (!h->gendisk[drv_index]){
1495 h->gendisk[drv_index] = alloc_disk(1 << NWD_SHIFT);
1496 if (!h->gendisk[drv_index]){
1497 printk(KERN_ERR "cciss: could not allocate new disk %d\n", drv_index);
1498 goto mem_msg;
1502 h->drv[drv_index].LunID = lunid;
1503 cciss_update_drive_info(ctlr, drv_index);
1504 } /* end for */
1505 } /* end else */
1507 freeret:
1508 kfree(ld_buff);
1509 h->busy_configuring = 0;
1510 /* We return -1 here to tell the ACU that we have registered/updated
1511 * all of the drives that we can and to keep it from calling us
1512 * additional times.
1514 return -1;
1515 mem_msg:
1516 printk(KERN_ERR "cciss: out of memory\n");
1517 goto freeret;
1520 /* This function will deregister the disk and it's queue from the
1521 * kernel. It must be called with the controller lock held and the
1522 * drv structures busy_configuring flag set. It's parameters are:
1524 * disk = This is the disk to be deregistered
1525 * drv = This is the drive_info_struct associated with the disk to be
1526 * deregistered. It contains information about the disk used
1527 * by the driver.
1528 * clear_all = This flag determines whether or not the disk information
1529 * is going to be completely cleared out and the highest_lun
1530 * reset. Sometimes we want to clear out information about
1531 * the disk in preparation for re-adding it. In this case
1532 * the highest_lun should be left unchanged and the LunID
1533 * should not be cleared.
1535 static int deregister_disk(struct gendisk *disk, drive_info_struct *drv,
1536 int clear_all)
1538 int i;
1539 ctlr_info_t *h = get_host(disk);
1541 if (!capable(CAP_SYS_RAWIO))
1542 return -EPERM;
1544 /* make sure logical volume is NOT is use */
1545 if (clear_all || (h->gendisk[0] == disk)) {
1546 if (drv->usage_count > 1)
1547 return -EBUSY;
1548 } else if (drv->usage_count > 0)
1549 return -EBUSY;
1551 /* invalidate the devices and deregister the disk. If it is disk
1552 * zero do not deregister it but just zero out it's values. This
1553 * allows us to delete disk zero but keep the controller registered.
1555 if (h->gendisk[0] != disk) {
1556 if (disk) {
1557 request_queue_t *q = disk->queue;
1558 if (disk->flags & GENHD_FL_UP)
1559 del_gendisk(disk);
1560 if (q) {
1561 blk_cleanup_queue(q);
1562 /* Set drv->queue to NULL so that we do not try
1563 * to call blk_start_queue on this queue in the
1564 * interrupt handler
1566 drv->queue = NULL;
1568 /* If clear_all is set then we are deleting the logical
1569 * drive, not just refreshing its info. For drives
1570 * other than disk 0 we will call put_disk. We do not
1571 * do this for disk 0 as we need it to be able to
1572 * configure the controller.
1574 if (clear_all){
1575 /* This isn't pretty, but we need to find the
1576 * disk in our array and NULL our the pointer.
1577 * This is so that we will call alloc_disk if
1578 * this index is used again later.
1580 for (i=0; i < CISS_MAX_LUN; i++){
1581 if(h->gendisk[i] == disk){
1582 h->gendisk[i] = NULL;
1583 break;
1586 put_disk(disk);
1589 } else {
1590 set_capacity(disk, 0);
1593 --h->num_luns;
1594 /* zero out the disk size info */
1595 drv->nr_blocks = 0;
1596 drv->block_size = 0;
1597 drv->heads = 0;
1598 drv->sectors = 0;
1599 drv->cylinders = 0;
1600 drv->raid_level = -1; /* This can be used as a flag variable to
1601 * indicate that this element of the drive
1602 * array is free.
1605 if (clear_all) {
1606 /* check to see if it was the last disk */
1607 if (drv == h->drv + h->highest_lun) {
1608 /* if so, find the new hightest lun */
1609 int i, newhighest = -1;
1610 for (i = 0; i < h->highest_lun; i++) {
1611 /* if the disk has size > 0, it is available */
1612 if (h->drv[i].heads)
1613 newhighest = i;
1615 h->highest_lun = newhighest;
1618 drv->LunID = 0;
1620 return 0;
1623 static int fill_cmd(CommandList_struct *c, __u8 cmd, int ctlr, void *buff, size_t size, unsigned int use_unit_num, /* 0: address the controller,
1624 1: address logical volume log_unit,
1625 2: periph device address is scsi3addr */
1626 unsigned int log_unit, __u8 page_code,
1627 unsigned char *scsi3addr, int cmd_type)
1629 ctlr_info_t *h = hba[ctlr];
1630 u64bit buff_dma_handle;
1631 int status = IO_OK;
1633 c->cmd_type = CMD_IOCTL_PEND;
1634 c->Header.ReplyQueue = 0;
1635 if (buff != NULL) {
1636 c->Header.SGList = 1;
1637 c->Header.SGTotal = 1;
1638 } else {
1639 c->Header.SGList = 0;
1640 c->Header.SGTotal = 0;
1642 c->Header.Tag.lower = c->busaddr;
1644 c->Request.Type.Type = cmd_type;
1645 if (cmd_type == TYPE_CMD) {
1646 switch (cmd) {
1647 case CISS_INQUIRY:
1648 /* If the logical unit number is 0 then, this is going
1649 to controller so It's a physical command
1650 mode = 0 target = 0. So we have nothing to write.
1651 otherwise, if use_unit_num == 1,
1652 mode = 1(volume set addressing) target = LUNID
1653 otherwise, if use_unit_num == 2,
1654 mode = 0(periph dev addr) target = scsi3addr */
1655 if (use_unit_num == 1) {
1656 c->Header.LUN.LogDev.VolId =
1657 h->drv[log_unit].LunID;
1658 c->Header.LUN.LogDev.Mode = 1;
1659 } else if (use_unit_num == 2) {
1660 memcpy(c->Header.LUN.LunAddrBytes, scsi3addr,
1662 c->Header.LUN.LogDev.Mode = 0;
1664 /* are we trying to read a vital product page */
1665 if (page_code != 0) {
1666 c->Request.CDB[1] = 0x01;
1667 c->Request.CDB[2] = page_code;
1669 c->Request.CDBLen = 6;
1670 c->Request.Type.Attribute = ATTR_SIMPLE;
1671 c->Request.Type.Direction = XFER_READ;
1672 c->Request.Timeout = 0;
1673 c->Request.CDB[0] = CISS_INQUIRY;
1674 c->Request.CDB[4] = size & 0xFF;
1675 break;
1676 case CISS_REPORT_LOG:
1677 case CISS_REPORT_PHYS:
1678 /* Talking to controller so It's a physical command
1679 mode = 00 target = 0. Nothing to write.
1681 c->Request.CDBLen = 12;
1682 c->Request.Type.Attribute = ATTR_SIMPLE;
1683 c->Request.Type.Direction = XFER_READ;
1684 c->Request.Timeout = 0;
1685 c->Request.CDB[0] = cmd;
1686 c->Request.CDB[6] = (size >> 24) & 0xFF; //MSB
1687 c->Request.CDB[7] = (size >> 16) & 0xFF;
1688 c->Request.CDB[8] = (size >> 8) & 0xFF;
1689 c->Request.CDB[9] = size & 0xFF;
1690 break;
1692 case CCISS_READ_CAPACITY:
1693 c->Header.LUN.LogDev.VolId = h->drv[log_unit].LunID;
1694 c->Header.LUN.LogDev.Mode = 1;
1695 c->Request.CDBLen = 10;
1696 c->Request.Type.Attribute = ATTR_SIMPLE;
1697 c->Request.Type.Direction = XFER_READ;
1698 c->Request.Timeout = 0;
1699 c->Request.CDB[0] = cmd;
1700 break;
1701 case CCISS_READ_CAPACITY_16:
1702 c->Header.LUN.LogDev.VolId = h->drv[log_unit].LunID;
1703 c->Header.LUN.LogDev.Mode = 1;
1704 c->Request.CDBLen = 16;
1705 c->Request.Type.Attribute = ATTR_SIMPLE;
1706 c->Request.Type.Direction = XFER_READ;
1707 c->Request.Timeout = 0;
1708 c->Request.CDB[0] = cmd;
1709 c->Request.CDB[1] = 0x10;
1710 c->Request.CDB[10] = (size >> 24) & 0xFF;
1711 c->Request.CDB[11] = (size >> 16) & 0xFF;
1712 c->Request.CDB[12] = (size >> 8) & 0xFF;
1713 c->Request.CDB[13] = size & 0xFF;
1714 c->Request.Timeout = 0;
1715 c->Request.CDB[0] = cmd;
1716 break;
1717 case CCISS_CACHE_FLUSH:
1718 c->Request.CDBLen = 12;
1719 c->Request.Type.Attribute = ATTR_SIMPLE;
1720 c->Request.Type.Direction = XFER_WRITE;
1721 c->Request.Timeout = 0;
1722 c->Request.CDB[0] = BMIC_WRITE;
1723 c->Request.CDB[6] = BMIC_CACHE_FLUSH;
1724 break;
1725 default:
1726 printk(KERN_WARNING
1727 "cciss%d: Unknown Command 0x%c\n", ctlr, cmd);
1728 return IO_ERROR;
1730 } else if (cmd_type == TYPE_MSG) {
1731 switch (cmd) {
1732 case 0: /* ABORT message */
1733 c->Request.CDBLen = 12;
1734 c->Request.Type.Attribute = ATTR_SIMPLE;
1735 c->Request.Type.Direction = XFER_WRITE;
1736 c->Request.Timeout = 0;
1737 c->Request.CDB[0] = cmd; /* abort */
1738 c->Request.CDB[1] = 0; /* abort a command */
1739 /* buff contains the tag of the command to abort */
1740 memcpy(&c->Request.CDB[4], buff, 8);
1741 break;
1742 case 1: /* RESET message */
1743 c->Request.CDBLen = 12;
1744 c->Request.Type.Attribute = ATTR_SIMPLE;
1745 c->Request.Type.Direction = XFER_WRITE;
1746 c->Request.Timeout = 0;
1747 memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
1748 c->Request.CDB[0] = cmd; /* reset */
1749 c->Request.CDB[1] = 0x04; /* reset a LUN */
1750 break;
1751 case 3: /* No-Op message */
1752 c->Request.CDBLen = 1;
1753 c->Request.Type.Attribute = ATTR_SIMPLE;
1754 c->Request.Type.Direction = XFER_WRITE;
1755 c->Request.Timeout = 0;
1756 c->Request.CDB[0] = cmd;
1757 break;
1758 default:
1759 printk(KERN_WARNING
1760 "cciss%d: unknown message type %d\n", ctlr, cmd);
1761 return IO_ERROR;
1763 } else {
1764 printk(KERN_WARNING
1765 "cciss%d: unknown command type %d\n", ctlr, cmd_type);
1766 return IO_ERROR;
1768 /* Fill in the scatter gather information */
1769 if (size > 0) {
1770 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
1771 buff, size,
1772 PCI_DMA_BIDIRECTIONAL);
1773 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
1774 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
1775 c->SG[0].Len = size;
1776 c->SG[0].Ext = 0; /* we are not chaining */
1778 return status;
1781 static int sendcmd_withirq(__u8 cmd,
1782 int ctlr,
1783 void *buff,
1784 size_t size,
1785 unsigned int use_unit_num,
1786 unsigned int log_unit, __u8 page_code, int cmd_type)
1788 ctlr_info_t *h = hba[ctlr];
1789 CommandList_struct *c;
1790 u64bit buff_dma_handle;
1791 unsigned long flags;
1792 int return_status;
1793 DECLARE_COMPLETION_ONSTACK(wait);
1795 if ((c = cmd_alloc(h, 0)) == NULL)
1796 return -ENOMEM;
1797 return_status = fill_cmd(c, cmd, ctlr, buff, size, use_unit_num,
1798 log_unit, page_code, NULL, cmd_type);
1799 if (return_status != IO_OK) {
1800 cmd_free(h, c, 0);
1801 return return_status;
1803 resend_cmd2:
1804 c->waiting = &wait;
1806 /* Put the request on the tail of the queue and send it */
1807 spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1808 addQ(&h->reqQ, c);
1809 h->Qdepth++;
1810 start_io(h);
1811 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1813 wait_for_completion(&wait);
1815 if (c->err_info->CommandStatus != 0) { /* an error has occurred */
1816 switch (c->err_info->CommandStatus) {
1817 case CMD_TARGET_STATUS:
1818 printk(KERN_WARNING "cciss: cmd %p has "
1819 " completed with errors\n", c);
1820 if (c->err_info->ScsiStatus) {
1821 printk(KERN_WARNING "cciss: cmd %p "
1822 "has SCSI Status = %x\n",
1823 c, c->err_info->ScsiStatus);
1826 break;
1827 case CMD_DATA_UNDERRUN:
1828 case CMD_DATA_OVERRUN:
1829 /* expected for inquire and report lun commands */
1830 break;
1831 case CMD_INVALID:
1832 printk(KERN_WARNING "cciss: Cmd %p is "
1833 "reported invalid\n", c);
1834 return_status = IO_ERROR;
1835 break;
1836 case CMD_PROTOCOL_ERR:
1837 printk(KERN_WARNING "cciss: cmd %p has "
1838 "protocol error \n", c);
1839 return_status = IO_ERROR;
1840 break;
1841 case CMD_HARDWARE_ERR:
1842 printk(KERN_WARNING "cciss: cmd %p had "
1843 " hardware error\n", c);
1844 return_status = IO_ERROR;
1845 break;
1846 case CMD_CONNECTION_LOST:
1847 printk(KERN_WARNING "cciss: cmd %p had "
1848 "connection lost\n", c);
1849 return_status = IO_ERROR;
1850 break;
1851 case CMD_ABORTED:
1852 printk(KERN_WARNING "cciss: cmd %p was "
1853 "aborted\n", c);
1854 return_status = IO_ERROR;
1855 break;
1856 case CMD_ABORT_FAILED:
1857 printk(KERN_WARNING "cciss: cmd %p reports "
1858 "abort failed\n", c);
1859 return_status = IO_ERROR;
1860 break;
1861 case CMD_UNSOLICITED_ABORT:
1862 printk(KERN_WARNING
1863 "cciss%d: unsolicited abort %p\n", ctlr, c);
1864 if (c->retry_count < MAX_CMD_RETRIES) {
1865 printk(KERN_WARNING
1866 "cciss%d: retrying %p\n", ctlr, c);
1867 c->retry_count++;
1868 /* erase the old error information */
1869 memset(c->err_info, 0,
1870 sizeof(ErrorInfo_struct));
1871 return_status = IO_OK;
1872 INIT_COMPLETION(wait);
1873 goto resend_cmd2;
1875 return_status = IO_ERROR;
1876 break;
1877 default:
1878 printk(KERN_WARNING "cciss: cmd %p returned "
1879 "unknown status %x\n", c,
1880 c->err_info->CommandStatus);
1881 return_status = IO_ERROR;
1884 /* unlock the buffers from DMA */
1885 buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
1886 buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
1887 pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
1888 c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
1889 cmd_free(h, c, 0);
1890 return return_status;
1893 static void cciss_geometry_inquiry(int ctlr, int logvol,
1894 int withirq, sector_t total_size,
1895 unsigned int block_size,
1896 InquiryData_struct *inq_buff,
1897 drive_info_struct *drv)
1899 int return_code;
1900 unsigned long t;
1902 memset(inq_buff, 0, sizeof(InquiryData_struct));
1903 if (withirq)
1904 return_code = sendcmd_withirq(CISS_INQUIRY, ctlr,
1905 inq_buff, sizeof(*inq_buff), 1,
1906 logvol, 0xC1, TYPE_CMD);
1907 else
1908 return_code = sendcmd(CISS_INQUIRY, ctlr, inq_buff,
1909 sizeof(*inq_buff), 1, logvol, 0xC1, NULL,
1910 TYPE_CMD);
1911 if (return_code == IO_OK) {
1912 if (inq_buff->data_byte[8] == 0xFF) {
1913 printk(KERN_WARNING
1914 "cciss: reading geometry failed, volume "
1915 "does not support reading geometry\n");
1916 drv->heads = 255;
1917 drv->sectors = 32; // Sectors per track
1918 drv->cylinders = total_size + 1;
1919 drv->raid_level = RAID_UNKNOWN;
1920 } else {
1921 drv->heads = inq_buff->data_byte[6];
1922 drv->sectors = inq_buff->data_byte[7];
1923 drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
1924 drv->cylinders += inq_buff->data_byte[5];
1925 drv->raid_level = inq_buff->data_byte[8];
1927 drv->block_size = block_size;
1928 drv->nr_blocks = total_size + 1;
1929 t = drv->heads * drv->sectors;
1930 if (t > 1) {
1931 sector_t real_size = total_size + 1;
1932 unsigned long rem = sector_div(real_size, t);
1933 if (rem)
1934 real_size++;
1935 drv->cylinders = real_size;
1937 } else { /* Get geometry failed */
1938 printk(KERN_WARNING "cciss: reading geometry failed\n");
1940 printk(KERN_INFO " heads=%d, sectors=%d, cylinders=%d\n\n",
1941 drv->heads, drv->sectors, drv->cylinders);
1944 static void
1945 cciss_read_capacity(int ctlr, int logvol, int withirq, sector_t *total_size,
1946 unsigned int *block_size)
1948 ReadCapdata_struct *buf;
1949 int return_code;
1950 buf = kmalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
1951 if (buf == NULL) {
1952 printk(KERN_WARNING "cciss: out of memory\n");
1953 return;
1955 memset(buf, 0, sizeof(ReadCapdata_struct));
1956 if (withirq)
1957 return_code = sendcmd_withirq(CCISS_READ_CAPACITY,
1958 ctlr, buf, sizeof(ReadCapdata_struct),
1959 1, logvol, 0, TYPE_CMD);
1960 else
1961 return_code = sendcmd(CCISS_READ_CAPACITY,
1962 ctlr, buf, sizeof(ReadCapdata_struct),
1963 1, logvol, 0, NULL, TYPE_CMD);
1964 if (return_code == IO_OK) {
1965 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
1966 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
1967 } else { /* read capacity command failed */
1968 printk(KERN_WARNING "cciss: read capacity failed\n");
1969 *total_size = 0;
1970 *block_size = BLOCK_SIZE;
1972 if (*total_size != 0)
1973 printk(KERN_INFO " blocks= %llu block_size= %d\n",
1974 (unsigned long long)*total_size+1, *block_size);
1975 kfree(buf);
1976 return;
1979 static void
1980 cciss_read_capacity_16(int ctlr, int logvol, int withirq, sector_t *total_size, unsigned int *block_size)
1982 ReadCapdata_struct_16 *buf;
1983 int return_code;
1984 buf = kmalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
1985 if (buf == NULL) {
1986 printk(KERN_WARNING "cciss: out of memory\n");
1987 return;
1989 memset(buf, 0, sizeof(ReadCapdata_struct_16));
1990 if (withirq) {
1991 return_code = sendcmd_withirq(CCISS_READ_CAPACITY_16,
1992 ctlr, buf, sizeof(ReadCapdata_struct_16),
1993 1, logvol, 0, TYPE_CMD);
1995 else {
1996 return_code = sendcmd(CCISS_READ_CAPACITY_16,
1997 ctlr, buf, sizeof(ReadCapdata_struct_16),
1998 1, logvol, 0, NULL, TYPE_CMD);
2000 if (return_code == IO_OK) {
2001 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2002 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2003 } else { /* read capacity command failed */
2004 printk(KERN_WARNING "cciss: read capacity failed\n");
2005 *total_size = 0;
2006 *block_size = BLOCK_SIZE;
2008 printk(KERN_INFO " blocks= %llu block_size= %d\n",
2009 (unsigned long long)*total_size+1, *block_size);
2010 kfree(buf);
2011 return;
2014 static int cciss_revalidate(struct gendisk *disk)
2016 ctlr_info_t *h = get_host(disk);
2017 drive_info_struct *drv = get_drv(disk);
2018 int logvol;
2019 int FOUND = 0;
2020 unsigned int block_size;
2021 sector_t total_size;
2022 InquiryData_struct *inq_buff = NULL;
2024 for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
2025 if (h->drv[logvol].LunID == drv->LunID) {
2026 FOUND = 1;
2027 break;
2031 if (!FOUND)
2032 return 1;
2034 inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2035 if (inq_buff == NULL) {
2036 printk(KERN_WARNING "cciss: out of memory\n");
2037 return 1;
2039 if (h->cciss_read == CCISS_READ_10) {
2040 cciss_read_capacity(h->ctlr, logvol, 1,
2041 &total_size, &block_size);
2042 } else {
2043 cciss_read_capacity_16(h->ctlr, logvol, 1,
2044 &total_size, &block_size);
2046 cciss_geometry_inquiry(h->ctlr, logvol, 1, total_size, block_size,
2047 inq_buff, drv);
2049 blk_queue_hardsect_size(drv->queue, drv->block_size);
2050 set_capacity(disk, drv->nr_blocks);
2052 kfree(inq_buff);
2053 return 0;
2057 * Wait polling for a command to complete.
2058 * The memory mapped FIFO is polled for the completion.
2059 * Used only at init time, interrupts from the HBA are disabled.
2061 static unsigned long pollcomplete(int ctlr)
2063 unsigned long done;
2064 int i;
2066 /* Wait (up to 20 seconds) for a command to complete */
2068 for (i = 20 * HZ; i > 0; i--) {
2069 done = hba[ctlr]->access.command_completed(hba[ctlr]);
2070 if (done == FIFO_EMPTY)
2071 schedule_timeout_uninterruptible(1);
2072 else
2073 return done;
2075 /* Invalid address to tell caller we ran out of time */
2076 return 1;
2079 static int add_sendcmd_reject(__u8 cmd, int ctlr, unsigned long complete)
2081 /* We get in here if sendcmd() is polling for completions
2082 and gets some command back that it wasn't expecting --
2083 something other than that which it just sent down.
2084 Ordinarily, that shouldn't happen, but it can happen when
2085 the scsi tape stuff gets into error handling mode, and
2086 starts using sendcmd() to try to abort commands and
2087 reset tape drives. In that case, sendcmd may pick up
2088 completions of commands that were sent to logical drives
2089 through the block i/o system, or cciss ioctls completing, etc.
2090 In that case, we need to save those completions for later
2091 processing by the interrupt handler.
2094 #ifdef CONFIG_CISS_SCSI_TAPE
2095 struct sendcmd_reject_list *srl = &hba[ctlr]->scsi_rejects;
2097 /* If it's not the scsi tape stuff doing error handling, (abort */
2098 /* or reset) then we don't expect anything weird. */
2099 if (cmd != CCISS_RESET_MSG && cmd != CCISS_ABORT_MSG) {
2100 #endif
2101 printk(KERN_WARNING "cciss cciss%d: SendCmd "
2102 "Invalid command list address returned! (%lx)\n",
2103 ctlr, complete);
2104 /* not much we can do. */
2105 #ifdef CONFIG_CISS_SCSI_TAPE
2106 return 1;
2109 /* We've sent down an abort or reset, but something else
2110 has completed */
2111 if (srl->ncompletions >= (hba[ctlr]->nr_cmds + 2)) {
2112 /* Uh oh. No room to save it for later... */
2113 printk(KERN_WARNING "cciss%d: Sendcmd: Invalid command addr, "
2114 "reject list overflow, command lost!\n", ctlr);
2115 return 1;
2117 /* Save it for later */
2118 srl->complete[srl->ncompletions] = complete;
2119 srl->ncompletions++;
2120 #endif
2121 return 0;
2125 * Send a command to the controller, and wait for it to complete.
2126 * Only used at init time.
2128 static int sendcmd(__u8 cmd, int ctlr, void *buff, size_t size, unsigned int use_unit_num, /* 0: address the controller,
2129 1: address logical volume log_unit,
2130 2: periph device address is scsi3addr */
2131 unsigned int log_unit,
2132 __u8 page_code, unsigned char *scsi3addr, int cmd_type)
2134 CommandList_struct *c;
2135 int i;
2136 unsigned long complete;
2137 ctlr_info_t *info_p = hba[ctlr];
2138 u64bit buff_dma_handle;
2139 int status, done = 0;
2141 if ((c = cmd_alloc(info_p, 1)) == NULL) {
2142 printk(KERN_WARNING "cciss: unable to get memory");
2143 return IO_ERROR;
2145 status = fill_cmd(c, cmd, ctlr, buff, size, use_unit_num,
2146 log_unit, page_code, scsi3addr, cmd_type);
2147 if (status != IO_OK) {
2148 cmd_free(info_p, c, 1);
2149 return status;
2151 resend_cmd1:
2153 * Disable interrupt
2155 #ifdef CCISS_DEBUG
2156 printk(KERN_DEBUG "cciss: turning intr off\n");
2157 #endif /* CCISS_DEBUG */
2158 info_p->access.set_intr_mask(info_p, CCISS_INTR_OFF);
2160 /* Make sure there is room in the command FIFO */
2161 /* Actually it should be completely empty at this time */
2162 /* unless we are in here doing error handling for the scsi */
2163 /* tape side of the driver. */
2164 for (i = 200000; i > 0; i--) {
2165 /* if fifo isn't full go */
2166 if (!(info_p->access.fifo_full(info_p))) {
2168 break;
2170 udelay(10);
2171 printk(KERN_WARNING "cciss cciss%d: SendCmd FIFO full,"
2172 " waiting!\n", ctlr);
2175 * Send the cmd
2177 info_p->access.submit_command(info_p, c);
2178 done = 0;
2179 do {
2180 complete = pollcomplete(ctlr);
2182 #ifdef CCISS_DEBUG
2183 printk(KERN_DEBUG "cciss: command completed\n");
2184 #endif /* CCISS_DEBUG */
2186 if (complete == 1) {
2187 printk(KERN_WARNING
2188 "cciss cciss%d: SendCmd Timeout out, "
2189 "No command list address returned!\n", ctlr);
2190 status = IO_ERROR;
2191 done = 1;
2192 break;
2195 /* This will need to change for direct lookup completions */
2196 if ((complete & CISS_ERROR_BIT)
2197 && (complete & ~CISS_ERROR_BIT) == c->busaddr) {
2198 /* if data overrun or underun on Report command
2199 ignore it
2201 if (((c->Request.CDB[0] == CISS_REPORT_LOG) ||
2202 (c->Request.CDB[0] == CISS_REPORT_PHYS) ||
2203 (c->Request.CDB[0] == CISS_INQUIRY)) &&
2204 ((c->err_info->CommandStatus ==
2205 CMD_DATA_OVERRUN) ||
2206 (c->err_info->CommandStatus == CMD_DATA_UNDERRUN)
2207 )) {
2208 complete = c->busaddr;
2209 } else {
2210 if (c->err_info->CommandStatus ==
2211 CMD_UNSOLICITED_ABORT) {
2212 printk(KERN_WARNING "cciss%d: "
2213 "unsolicited abort %p\n",
2214 ctlr, c);
2215 if (c->retry_count < MAX_CMD_RETRIES) {
2216 printk(KERN_WARNING
2217 "cciss%d: retrying %p\n",
2218 ctlr, c);
2219 c->retry_count++;
2220 /* erase the old error */
2221 /* information */
2222 memset(c->err_info, 0,
2223 sizeof
2224 (ErrorInfo_struct));
2225 goto resend_cmd1;
2226 } else {
2227 printk(KERN_WARNING
2228 "cciss%d: retried %p too "
2229 "many times\n", ctlr, c);
2230 status = IO_ERROR;
2231 goto cleanup1;
2233 } else if (c->err_info->CommandStatus ==
2234 CMD_UNABORTABLE) {
2235 printk(KERN_WARNING
2236 "cciss%d: command could not be aborted.\n",
2237 ctlr);
2238 status = IO_ERROR;
2239 goto cleanup1;
2241 printk(KERN_WARNING "ciss ciss%d: sendcmd"
2242 " Error %x \n", ctlr,
2243 c->err_info->CommandStatus);
2244 printk(KERN_WARNING "ciss ciss%d: sendcmd"
2245 " offensive info\n"
2246 " size %x\n num %x value %x\n",
2247 ctlr,
2248 c->err_info->MoreErrInfo.Invalid_Cmd.
2249 offense_size,
2250 c->err_info->MoreErrInfo.Invalid_Cmd.
2251 offense_num,
2252 c->err_info->MoreErrInfo.Invalid_Cmd.
2253 offense_value);
2254 status = IO_ERROR;
2255 goto cleanup1;
2258 /* This will need changing for direct lookup completions */
2259 if (complete != c->busaddr) {
2260 if (add_sendcmd_reject(cmd, ctlr, complete) != 0) {
2261 BUG(); /* we are pretty much hosed if we get here. */
2263 continue;
2264 } else
2265 done = 1;
2266 } while (!done);
2268 cleanup1:
2269 /* unlock the data buffer from DMA */
2270 buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2271 buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2272 pci_unmap_single(info_p->pdev, (dma_addr_t) buff_dma_handle.val,
2273 c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2274 #ifdef CONFIG_CISS_SCSI_TAPE
2275 /* if we saved some commands for later, process them now. */
2276 if (info_p->scsi_rejects.ncompletions > 0)
2277 do_cciss_intr(0, info_p);
2278 #endif
2279 cmd_free(info_p, c, 1);
2280 return status;
2284 * Map (physical) PCI mem into (virtual) kernel space
2286 static void __iomem *remap_pci_mem(ulong base, ulong size)
2288 ulong page_base = ((ulong) base) & PAGE_MASK;
2289 ulong page_offs = ((ulong) base) - page_base;
2290 void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2292 return page_remapped ? (page_remapped + page_offs) : NULL;
2296 * Takes jobs of the Q and sends them to the hardware, then puts it on
2297 * the Q to wait for completion.
2299 static void start_io(ctlr_info_t *h)
2301 CommandList_struct *c;
2303 while ((c = h->reqQ) != NULL) {
2304 /* can't do anything if fifo is full */
2305 if ((h->access.fifo_full(h))) {
2306 printk(KERN_WARNING "cciss: fifo full\n");
2307 break;
2310 /* Get the first entry from the Request Q */
2311 removeQ(&(h->reqQ), c);
2312 h->Qdepth--;
2314 /* Tell the controller execute command */
2315 h->access.submit_command(h, c);
2317 /* Put job onto the completed Q */
2318 addQ(&(h->cmpQ), c);
2322 /* Assumes that CCISS_LOCK(h->ctlr) is held. */
2323 /* Zeros out the error record and then resends the command back */
2324 /* to the controller */
2325 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2327 /* erase the old error information */
2328 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2330 /* add it to software queue and then send it to the controller */
2331 addQ(&(h->reqQ), c);
2332 h->Qdepth++;
2333 if (h->Qdepth > h->maxQsinceinit)
2334 h->maxQsinceinit = h->Qdepth;
2336 start_io(h);
2339 /* checks the status of the job and calls complete buffers to mark all
2340 * buffers for the completed job. Note that this function does not need
2341 * to hold the hba/queue lock.
2343 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
2344 int timeout)
2346 int status = 1;
2347 int retry_cmd = 0;
2349 if (timeout)
2350 status = 0;
2352 if (cmd->err_info->CommandStatus != 0) { /* an error has occurred */
2353 switch (cmd->err_info->CommandStatus) {
2354 unsigned char sense_key;
2355 case CMD_TARGET_STATUS:
2356 status = 0;
2358 if (cmd->err_info->ScsiStatus == 0x02) {
2359 printk(KERN_WARNING "cciss: cmd %p "
2360 "has CHECK CONDITION "
2361 " byte 2 = 0x%x\n", cmd,
2362 cmd->err_info->SenseInfo[2]
2364 /* check the sense key */
2365 sense_key = 0xf & cmd->err_info->SenseInfo[2];
2366 /* no status or recovered error */
2367 if ((sense_key == 0x0) || (sense_key == 0x1)) {
2368 status = 1;
2370 } else {
2371 printk(KERN_WARNING "cciss: cmd %p "
2372 "has SCSI Status 0x%x\n",
2373 cmd, cmd->err_info->ScsiStatus);
2375 break;
2376 case CMD_DATA_UNDERRUN:
2377 printk(KERN_WARNING "cciss: cmd %p has"
2378 " completed with data underrun "
2379 "reported\n", cmd);
2380 break;
2381 case CMD_DATA_OVERRUN:
2382 printk(KERN_WARNING "cciss: cmd %p has"
2383 " completed with data overrun "
2384 "reported\n", cmd);
2385 break;
2386 case CMD_INVALID:
2387 printk(KERN_WARNING "cciss: cmd %p is "
2388 "reported invalid\n", cmd);
2389 status = 0;
2390 break;
2391 case CMD_PROTOCOL_ERR:
2392 printk(KERN_WARNING "cciss: cmd %p has "
2393 "protocol error \n", cmd);
2394 status = 0;
2395 break;
2396 case CMD_HARDWARE_ERR:
2397 printk(KERN_WARNING "cciss: cmd %p had "
2398 " hardware error\n", cmd);
2399 status = 0;
2400 break;
2401 case CMD_CONNECTION_LOST:
2402 printk(KERN_WARNING "cciss: cmd %p had "
2403 "connection lost\n", cmd);
2404 status = 0;
2405 break;
2406 case CMD_ABORTED:
2407 printk(KERN_WARNING "cciss: cmd %p was "
2408 "aborted\n", cmd);
2409 status = 0;
2410 break;
2411 case CMD_ABORT_FAILED:
2412 printk(KERN_WARNING "cciss: cmd %p reports "
2413 "abort failed\n", cmd);
2414 status = 0;
2415 break;
2416 case CMD_UNSOLICITED_ABORT:
2417 printk(KERN_WARNING "cciss%d: unsolicited "
2418 "abort %p\n", h->ctlr, cmd);
2419 if (cmd->retry_count < MAX_CMD_RETRIES) {
2420 retry_cmd = 1;
2421 printk(KERN_WARNING
2422 "cciss%d: retrying %p\n", h->ctlr, cmd);
2423 cmd->retry_count++;
2424 } else
2425 printk(KERN_WARNING
2426 "cciss%d: %p retried too "
2427 "many times\n", h->ctlr, cmd);
2428 status = 0;
2429 break;
2430 case CMD_TIMEOUT:
2431 printk(KERN_WARNING "cciss: cmd %p timedout\n", cmd);
2432 status = 0;
2433 break;
2434 default:
2435 printk(KERN_WARNING "cciss: cmd %p returned "
2436 "unknown status %x\n", cmd,
2437 cmd->err_info->CommandStatus);
2438 status = 0;
2441 /* We need to return this command */
2442 if (retry_cmd) {
2443 resend_cciss_cmd(h, cmd);
2444 return;
2447 cmd->rq->completion_data = cmd;
2448 cmd->rq->errors = status;
2449 blk_add_trace_rq(cmd->rq->q, cmd->rq, BLK_TA_COMPLETE);
2450 blk_complete_request(cmd->rq);
2454 * Get a request and submit it to the controller.
2456 static void do_cciss_request(request_queue_t *q)
2458 ctlr_info_t *h = q->queuedata;
2459 CommandList_struct *c;
2460 sector_t start_blk;
2461 int seg;
2462 struct request *creq;
2463 u64bit temp64;
2464 struct scatterlist tmp_sg[MAXSGENTRIES];
2465 drive_info_struct *drv;
2466 int i, dir;
2468 /* We call start_io here in case there is a command waiting on the
2469 * queue that has not been sent.
2471 if (blk_queue_plugged(q))
2472 goto startio;
2474 queue:
2475 creq = elv_next_request(q);
2476 if (!creq)
2477 goto startio;
2479 BUG_ON(creq->nr_phys_segments > MAXSGENTRIES);
2481 if ((c = cmd_alloc(h, 1)) == NULL)
2482 goto full;
2484 blkdev_dequeue_request(creq);
2486 spin_unlock_irq(q->queue_lock);
2488 c->cmd_type = CMD_RWREQ;
2489 c->rq = creq;
2491 /* fill in the request */
2492 drv = creq->rq_disk->private_data;
2493 c->Header.ReplyQueue = 0; // unused in simple mode
2494 /* got command from pool, so use the command block index instead */
2495 /* for direct lookups. */
2496 /* The first 2 bits are reserved for controller error reporting. */
2497 c->Header.Tag.lower = (c->cmdindex << 3);
2498 c->Header.Tag.lower |= 0x04; /* flag for direct lookup. */
2499 c->Header.LUN.LogDev.VolId = drv->LunID;
2500 c->Header.LUN.LogDev.Mode = 1;
2501 c->Request.CDBLen = 10; // 12 byte commands not in FW yet;
2502 c->Request.Type.Type = TYPE_CMD; // It is a command.
2503 c->Request.Type.Attribute = ATTR_SIMPLE;
2504 c->Request.Type.Direction =
2505 (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
2506 c->Request.Timeout = 0; // Don't time out
2507 c->Request.CDB[0] =
2508 (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
2509 start_blk = creq->sector;
2510 #ifdef CCISS_DEBUG
2511 printk(KERN_DEBUG "ciss: sector =%d nr_sectors=%d\n", (int)creq->sector,
2512 (int)creq->nr_sectors);
2513 #endif /* CCISS_DEBUG */
2515 seg = blk_rq_map_sg(q, creq, tmp_sg);
2517 /* get the DMA records for the setup */
2518 if (c->Request.Type.Direction == XFER_READ)
2519 dir = PCI_DMA_FROMDEVICE;
2520 else
2521 dir = PCI_DMA_TODEVICE;
2523 for (i = 0; i < seg; i++) {
2524 c->SG[i].Len = tmp_sg[i].length;
2525 temp64.val = (__u64) pci_map_page(h->pdev, tmp_sg[i].page,
2526 tmp_sg[i].offset,
2527 tmp_sg[i].length, dir);
2528 c->SG[i].Addr.lower = temp64.val32.lower;
2529 c->SG[i].Addr.upper = temp64.val32.upper;
2530 c->SG[i].Ext = 0; // we are not chaining
2532 /* track how many SG entries we are using */
2533 if (seg > h->maxSG)
2534 h->maxSG = seg;
2536 #ifdef CCISS_DEBUG
2537 printk(KERN_DEBUG "cciss: Submitting %d sectors in %d segments\n",
2538 creq->nr_sectors, seg);
2539 #endif /* CCISS_DEBUG */
2541 c->Header.SGList = c->Header.SGTotal = seg;
2542 if(h->cciss_read == CCISS_READ_10) {
2543 c->Request.CDB[1] = 0;
2544 c->Request.CDB[2] = (start_blk >> 24) & 0xff; //MSB
2545 c->Request.CDB[3] = (start_blk >> 16) & 0xff;
2546 c->Request.CDB[4] = (start_blk >> 8) & 0xff;
2547 c->Request.CDB[5] = start_blk & 0xff;
2548 c->Request.CDB[6] = 0; // (sect >> 24) & 0xff; MSB
2549 c->Request.CDB[7] = (creq->nr_sectors >> 8) & 0xff;
2550 c->Request.CDB[8] = creq->nr_sectors & 0xff;
2551 c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
2552 } else {
2553 c->Request.CDBLen = 16;
2554 c->Request.CDB[1]= 0;
2555 c->Request.CDB[2]= (start_blk >> 56) & 0xff; //MSB
2556 c->Request.CDB[3]= (start_blk >> 48) & 0xff;
2557 c->Request.CDB[4]= (start_blk >> 40) & 0xff;
2558 c->Request.CDB[5]= (start_blk >> 32) & 0xff;
2559 c->Request.CDB[6]= (start_blk >> 24) & 0xff;
2560 c->Request.CDB[7]= (start_blk >> 16) & 0xff;
2561 c->Request.CDB[8]= (start_blk >> 8) & 0xff;
2562 c->Request.CDB[9]= start_blk & 0xff;
2563 c->Request.CDB[10]= (creq->nr_sectors >> 24) & 0xff;
2564 c->Request.CDB[11]= (creq->nr_sectors >> 16) & 0xff;
2565 c->Request.CDB[12]= (creq->nr_sectors >> 8) & 0xff;
2566 c->Request.CDB[13]= creq->nr_sectors & 0xff;
2567 c->Request.CDB[14] = c->Request.CDB[15] = 0;
2570 spin_lock_irq(q->queue_lock);
2572 addQ(&(h->reqQ), c);
2573 h->Qdepth++;
2574 if (h->Qdepth > h->maxQsinceinit)
2575 h->maxQsinceinit = h->Qdepth;
2577 goto queue;
2578 full:
2579 blk_stop_queue(q);
2580 startio:
2581 /* We will already have the driver lock here so not need
2582 * to lock it.
2584 start_io(h);
2587 static inline unsigned long get_next_completion(ctlr_info_t *h)
2589 #ifdef CONFIG_CISS_SCSI_TAPE
2590 /* Any rejects from sendcmd() lying around? Process them first */
2591 if (h->scsi_rejects.ncompletions == 0)
2592 return h->access.command_completed(h);
2593 else {
2594 struct sendcmd_reject_list *srl;
2595 int n;
2596 srl = &h->scsi_rejects;
2597 n = --srl->ncompletions;
2598 /* printk("cciss%d: processing saved reject\n", h->ctlr); */
2599 printk("p");
2600 return srl->complete[n];
2602 #else
2603 return h->access.command_completed(h);
2604 #endif
2607 static inline int interrupt_pending(ctlr_info_t *h)
2609 #ifdef CONFIG_CISS_SCSI_TAPE
2610 return (h->access.intr_pending(h)
2611 || (h->scsi_rejects.ncompletions > 0));
2612 #else
2613 return h->access.intr_pending(h);
2614 #endif
2617 static inline long interrupt_not_for_us(ctlr_info_t *h)
2619 #ifdef CONFIG_CISS_SCSI_TAPE
2620 return (((h->access.intr_pending(h) == 0) ||
2621 (h->interrupts_enabled == 0))
2622 && (h->scsi_rejects.ncompletions == 0));
2623 #else
2624 return (((h->access.intr_pending(h) == 0) ||
2625 (h->interrupts_enabled == 0)));
2626 #endif
2629 static irqreturn_t do_cciss_intr(int irq, void *dev_id)
2631 ctlr_info_t *h = dev_id;
2632 CommandList_struct *c;
2633 unsigned long flags;
2634 __u32 a, a1, a2;
2636 if (interrupt_not_for_us(h))
2637 return IRQ_NONE;
2639 * If there are completed commands in the completion queue,
2640 * we had better do something about it.
2642 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2643 while (interrupt_pending(h)) {
2644 while ((a = get_next_completion(h)) != FIFO_EMPTY) {
2645 a1 = a;
2646 if ((a & 0x04)) {
2647 a2 = (a >> 3);
2648 if (a2 >= h->nr_cmds) {
2649 printk(KERN_WARNING
2650 "cciss: controller cciss%d failed, stopping.\n",
2651 h->ctlr);
2652 fail_all_cmds(h->ctlr);
2653 return IRQ_HANDLED;
2656 c = h->cmd_pool + a2;
2657 a = c->busaddr;
2659 } else {
2660 a &= ~3;
2661 if ((c = h->cmpQ) == NULL) {
2662 printk(KERN_WARNING
2663 "cciss: Completion of %08x ignored\n",
2664 a1);
2665 continue;
2667 while (c->busaddr != a) {
2668 c = c->next;
2669 if (c == h->cmpQ)
2670 break;
2674 * If we've found the command, take it off the
2675 * completion Q and free it
2677 if (c->busaddr == a) {
2678 removeQ(&h->cmpQ, c);
2679 if (c->cmd_type == CMD_RWREQ) {
2680 complete_command(h, c, 0);
2681 } else if (c->cmd_type == CMD_IOCTL_PEND) {
2682 complete(c->waiting);
2684 # ifdef CONFIG_CISS_SCSI_TAPE
2685 else if (c->cmd_type == CMD_SCSI)
2686 complete_scsi_command(c, 0, a1);
2687 # endif
2688 continue;
2693 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2694 return IRQ_HANDLED;
2698 * We cannot read the structure directly, for portability we must use
2699 * the io functions.
2700 * This is for debug only.
2702 #ifdef CCISS_DEBUG
2703 static void print_cfg_table(CfgTable_struct *tb)
2705 int i;
2706 char temp_name[17];
2708 printk("Controller Configuration information\n");
2709 printk("------------------------------------\n");
2710 for (i = 0; i < 4; i++)
2711 temp_name[i] = readb(&(tb->Signature[i]));
2712 temp_name[4] = '\0';
2713 printk(" Signature = %s\n", temp_name);
2714 printk(" Spec Number = %d\n", readl(&(tb->SpecValence)));
2715 printk(" Transport methods supported = 0x%x\n",
2716 readl(&(tb->TransportSupport)));
2717 printk(" Transport methods active = 0x%x\n",
2718 readl(&(tb->TransportActive)));
2719 printk(" Requested transport Method = 0x%x\n",
2720 readl(&(tb->HostWrite.TransportRequest)));
2721 printk(" Coalesce Interrupt Delay = 0x%x\n",
2722 readl(&(tb->HostWrite.CoalIntDelay)));
2723 printk(" Coalesce Interrupt Count = 0x%x\n",
2724 readl(&(tb->HostWrite.CoalIntCount)));
2725 printk(" Max outstanding commands = 0x%d\n",
2726 readl(&(tb->CmdsOutMax)));
2727 printk(" Bus Types = 0x%x\n", readl(&(tb->BusTypes)));
2728 for (i = 0; i < 16; i++)
2729 temp_name[i] = readb(&(tb->ServerName[i]));
2730 temp_name[16] = '\0';
2731 printk(" Server Name = %s\n", temp_name);
2732 printk(" Heartbeat Counter = 0x%x\n\n\n", readl(&(tb->HeartBeat)));
2734 #endif /* CCISS_DEBUG */
2736 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
2738 int i, offset, mem_type, bar_type;
2739 if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
2740 return 0;
2741 offset = 0;
2742 for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
2743 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
2744 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
2745 offset += 4;
2746 else {
2747 mem_type = pci_resource_flags(pdev, i) &
2748 PCI_BASE_ADDRESS_MEM_TYPE_MASK;
2749 switch (mem_type) {
2750 case PCI_BASE_ADDRESS_MEM_TYPE_32:
2751 case PCI_BASE_ADDRESS_MEM_TYPE_1M:
2752 offset += 4; /* 32 bit */
2753 break;
2754 case PCI_BASE_ADDRESS_MEM_TYPE_64:
2755 offset += 8;
2756 break;
2757 default: /* reserved in PCI 2.2 */
2758 printk(KERN_WARNING
2759 "Base address is invalid\n");
2760 return -1;
2761 break;
2764 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
2765 return i + 1;
2767 return -1;
2770 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
2771 * controllers that are capable. If not, we use IO-APIC mode.
2774 static void __devinit cciss_interrupt_mode(ctlr_info_t *c,
2775 struct pci_dev *pdev, __u32 board_id)
2777 #ifdef CONFIG_PCI_MSI
2778 int err;
2779 struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
2780 {0, 2}, {0, 3}
2783 /* Some boards advertise MSI but don't really support it */
2784 if ((board_id == 0x40700E11) ||
2785 (board_id == 0x40800E11) ||
2786 (board_id == 0x40820E11) || (board_id == 0x40830E11))
2787 goto default_int_mode;
2789 if (pci_find_capability(pdev, PCI_CAP_ID_MSIX)) {
2790 err = pci_enable_msix(pdev, cciss_msix_entries, 4);
2791 if (!err) {
2792 c->intr[0] = cciss_msix_entries[0].vector;
2793 c->intr[1] = cciss_msix_entries[1].vector;
2794 c->intr[2] = cciss_msix_entries[2].vector;
2795 c->intr[3] = cciss_msix_entries[3].vector;
2796 c->msix_vector = 1;
2797 return;
2799 if (err > 0) {
2800 printk(KERN_WARNING "cciss: only %d MSI-X vectors "
2801 "available\n", err);
2802 goto default_int_mode;
2803 } else {
2804 printk(KERN_WARNING "cciss: MSI-X init failed %d\n",
2805 err);
2806 goto default_int_mode;
2809 if (pci_find_capability(pdev, PCI_CAP_ID_MSI)) {
2810 if (!pci_enable_msi(pdev)) {
2811 c->msi_vector = 1;
2812 } else {
2813 printk(KERN_WARNING "cciss: MSI init failed\n");
2816 default_int_mode:
2817 #endif /* CONFIG_PCI_MSI */
2818 /* if we get here we're going to use the default interrupt mode */
2819 c->intr[SIMPLE_MODE_INT] = pdev->irq;
2820 return;
2823 static int cciss_pci_init(ctlr_info_t *c, struct pci_dev *pdev)
2825 ushort subsystem_vendor_id, subsystem_device_id, command;
2826 __u32 board_id, scratchpad = 0;
2827 __u64 cfg_offset;
2828 __u32 cfg_base_addr;
2829 __u64 cfg_base_addr_index;
2830 int i, err;
2832 /* check to see if controller has been disabled */
2833 /* BEFORE trying to enable it */
2834 (void)pci_read_config_word(pdev, PCI_COMMAND, &command);
2835 if (!(command & 0x02)) {
2836 printk(KERN_WARNING
2837 "cciss: controller appears to be disabled\n");
2838 return -ENODEV;
2841 err = pci_enable_device(pdev);
2842 if (err) {
2843 printk(KERN_ERR "cciss: Unable to Enable PCI device\n");
2844 return err;
2847 err = pci_request_regions(pdev, "cciss");
2848 if (err) {
2849 printk(KERN_ERR "cciss: Cannot obtain PCI resources, "
2850 "aborting\n");
2851 return err;
2854 subsystem_vendor_id = pdev->subsystem_vendor;
2855 subsystem_device_id = pdev->subsystem_device;
2856 board_id = (((__u32) (subsystem_device_id << 16) & 0xffff0000) |
2857 subsystem_vendor_id);
2859 #ifdef CCISS_DEBUG
2860 printk("command = %x\n", command);
2861 printk("irq = %x\n", pdev->irq);
2862 printk("board_id = %x\n", board_id);
2863 #endif /* CCISS_DEBUG */
2865 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
2866 * else we use the IO-APIC interrupt assigned to us by system ROM.
2868 cciss_interrupt_mode(c, pdev, board_id);
2871 * Memory base addr is first addr , the second points to the config
2872 * table
2875 c->paddr = pci_resource_start(pdev, 0); /* addressing mode bits already removed */
2876 #ifdef CCISS_DEBUG
2877 printk("address 0 = %x\n", c->paddr);
2878 #endif /* CCISS_DEBUG */
2879 c->vaddr = remap_pci_mem(c->paddr, 0x250);
2881 /* Wait for the board to become ready. (PCI hotplug needs this.)
2882 * We poll for up to 120 secs, once per 100ms. */
2883 for (i = 0; i < 1200; i++) {
2884 scratchpad = readl(c->vaddr + SA5_SCRATCHPAD_OFFSET);
2885 if (scratchpad == CCISS_FIRMWARE_READY)
2886 break;
2887 set_current_state(TASK_INTERRUPTIBLE);
2888 schedule_timeout(HZ / 10); /* wait 100ms */
2890 if (scratchpad != CCISS_FIRMWARE_READY) {
2891 printk(KERN_WARNING "cciss: Board not ready. Timed out.\n");
2892 err = -ENODEV;
2893 goto err_out_free_res;
2896 /* get the address index number */
2897 cfg_base_addr = readl(c->vaddr + SA5_CTCFG_OFFSET);
2898 cfg_base_addr &= (__u32) 0x0000ffff;
2899 #ifdef CCISS_DEBUG
2900 printk("cfg base address = %x\n", cfg_base_addr);
2901 #endif /* CCISS_DEBUG */
2902 cfg_base_addr_index = find_PCI_BAR_index(pdev, cfg_base_addr);
2903 #ifdef CCISS_DEBUG
2904 printk("cfg base address index = %x\n", cfg_base_addr_index);
2905 #endif /* CCISS_DEBUG */
2906 if (cfg_base_addr_index == -1) {
2907 printk(KERN_WARNING "cciss: Cannot find cfg_base_addr_index\n");
2908 err = -ENODEV;
2909 goto err_out_free_res;
2912 cfg_offset = readl(c->vaddr + SA5_CTMEM_OFFSET);
2913 #ifdef CCISS_DEBUG
2914 printk("cfg offset = %x\n", cfg_offset);
2915 #endif /* CCISS_DEBUG */
2916 c->cfgtable = remap_pci_mem(pci_resource_start(pdev,
2917 cfg_base_addr_index) +
2918 cfg_offset, sizeof(CfgTable_struct));
2919 c->board_id = board_id;
2921 #ifdef CCISS_DEBUG
2922 print_cfg_table(c->cfgtable);
2923 #endif /* CCISS_DEBUG */
2925 for (i = 0; i < ARRAY_SIZE(products); i++) {
2926 if (board_id == products[i].board_id) {
2927 c->product_name = products[i].product_name;
2928 c->access = *(products[i].access);
2929 c->nr_cmds = products[i].nr_cmds;
2930 break;
2933 if ((readb(&c->cfgtable->Signature[0]) != 'C') ||
2934 (readb(&c->cfgtable->Signature[1]) != 'I') ||
2935 (readb(&c->cfgtable->Signature[2]) != 'S') ||
2936 (readb(&c->cfgtable->Signature[3]) != 'S')) {
2937 printk("Does not appear to be a valid CISS config table\n");
2938 err = -ENODEV;
2939 goto err_out_free_res;
2941 /* We didn't find the controller in our list. We know the
2942 * signature is valid. If it's an HP device let's try to
2943 * bind to the device and fire it up. Otherwise we bail.
2945 if (i == ARRAY_SIZE(products)) {
2946 if (subsystem_vendor_id == PCI_VENDOR_ID_HP) {
2947 c->product_name = products[i-1].product_name;
2948 c->access = *(products[i-1].access);
2949 c->nr_cmds = products[i-1].nr_cmds;
2950 printk(KERN_WARNING "cciss: This is an unknown "
2951 "Smart Array controller.\n"
2952 "cciss: Please update to the latest driver "
2953 "available from www.hp.com.\n");
2954 } else {
2955 printk(KERN_WARNING "cciss: Sorry, I don't know how"
2956 " to access the Smart Array controller %08lx\n"
2957 , (unsigned long)board_id);
2958 err = -ENODEV;
2959 goto err_out_free_res;
2962 #ifdef CONFIG_X86
2964 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
2965 __u32 prefetch;
2966 prefetch = readl(&(c->cfgtable->SCSI_Prefetch));
2967 prefetch |= 0x100;
2968 writel(prefetch, &(c->cfgtable->SCSI_Prefetch));
2970 #endif
2972 /* Disabling DMA prefetch for the P600
2973 * An ASIC bug may result in a prefetch beyond
2974 * physical memory.
2976 if(board_id == 0x3225103C) {
2977 __u32 dma_prefetch;
2978 dma_prefetch = readl(c->vaddr + I2O_DMA1_CFG);
2979 dma_prefetch |= 0x8000;
2980 writel(dma_prefetch, c->vaddr + I2O_DMA1_CFG);
2983 #ifdef CCISS_DEBUG
2984 printk("Trying to put board into Simple mode\n");
2985 #endif /* CCISS_DEBUG */
2986 c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
2987 /* Update the field, and then ring the doorbell */
2988 writel(CFGTBL_Trans_Simple, &(c->cfgtable->HostWrite.TransportRequest));
2989 writel(CFGTBL_ChangeReq, c->vaddr + SA5_DOORBELL);
2991 /* under certain very rare conditions, this can take awhile.
2992 * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
2993 * as we enter this code.) */
2994 for (i = 0; i < MAX_CONFIG_WAIT; i++) {
2995 if (!(readl(c->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
2996 break;
2997 /* delay and try again */
2998 set_current_state(TASK_INTERRUPTIBLE);
2999 schedule_timeout(10);
3002 #ifdef CCISS_DEBUG
3003 printk(KERN_DEBUG "I counter got to %d %x\n", i,
3004 readl(c->vaddr + SA5_DOORBELL));
3005 #endif /* CCISS_DEBUG */
3006 #ifdef CCISS_DEBUG
3007 print_cfg_table(c->cfgtable);
3008 #endif /* CCISS_DEBUG */
3010 if (!(readl(&(c->cfgtable->TransportActive)) & CFGTBL_Trans_Simple)) {
3011 printk(KERN_WARNING "cciss: unable to get board into"
3012 " simple mode\n");
3013 err = -ENODEV;
3014 goto err_out_free_res;
3016 return 0;
3018 err_out_free_res:
3020 * Deliberately omit pci_disable_device(): it does something nasty to
3021 * Smart Array controllers that pci_enable_device does not undo
3023 pci_release_regions(pdev);
3024 return err;
3028 * Gets information about the local volumes attached to the controller.
3030 static void cciss_getgeometry(int cntl_num)
3032 ReportLunData_struct *ld_buff;
3033 InquiryData_struct *inq_buff;
3034 int return_code;
3035 int i;
3036 int listlength = 0;
3037 __u32 lunid = 0;
3038 int block_size;
3039 sector_t total_size;
3041 ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
3042 if (ld_buff == NULL) {
3043 printk(KERN_ERR "cciss: out of memory\n");
3044 return;
3046 inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
3047 if (inq_buff == NULL) {
3048 printk(KERN_ERR "cciss: out of memory\n");
3049 kfree(ld_buff);
3050 return;
3052 /* Get the firmware version */
3053 return_code = sendcmd(CISS_INQUIRY, cntl_num, inq_buff,
3054 sizeof(InquiryData_struct), 0, 0, 0, NULL,
3055 TYPE_CMD);
3056 if (return_code == IO_OK) {
3057 hba[cntl_num]->firm_ver[0] = inq_buff->data_byte[32];
3058 hba[cntl_num]->firm_ver[1] = inq_buff->data_byte[33];
3059 hba[cntl_num]->firm_ver[2] = inq_buff->data_byte[34];
3060 hba[cntl_num]->firm_ver[3] = inq_buff->data_byte[35];
3061 } else { /* send command failed */
3063 printk(KERN_WARNING "cciss: unable to determine firmware"
3064 " version of controller\n");
3066 /* Get the number of logical volumes */
3067 return_code = sendcmd(CISS_REPORT_LOG, cntl_num, ld_buff,
3068 sizeof(ReportLunData_struct), 0, 0, 0, NULL,
3069 TYPE_CMD);
3071 if (return_code == IO_OK) {
3072 #ifdef CCISS_DEBUG
3073 printk("LUN Data\n--------------------------\n");
3074 #endif /* CCISS_DEBUG */
3076 listlength |=
3077 (0xff & (unsigned int)(ld_buff->LUNListLength[0])) << 24;
3078 listlength |=
3079 (0xff & (unsigned int)(ld_buff->LUNListLength[1])) << 16;
3080 listlength |=
3081 (0xff & (unsigned int)(ld_buff->LUNListLength[2])) << 8;
3082 listlength |= 0xff & (unsigned int)(ld_buff->LUNListLength[3]);
3083 } else { /* reading number of logical volumes failed */
3085 printk(KERN_WARNING "cciss: report logical volume"
3086 " command failed\n");
3087 listlength = 0;
3089 hba[cntl_num]->num_luns = listlength / 8; // 8 bytes pre entry
3090 if (hba[cntl_num]->num_luns > CISS_MAX_LUN) {
3091 printk(KERN_ERR
3092 "ciss: only %d number of logical volumes supported\n",
3093 CISS_MAX_LUN);
3094 hba[cntl_num]->num_luns = CISS_MAX_LUN;
3096 #ifdef CCISS_DEBUG
3097 printk(KERN_DEBUG "Length = %x %x %x %x = %d\n",
3098 ld_buff->LUNListLength[0], ld_buff->LUNListLength[1],
3099 ld_buff->LUNListLength[2], ld_buff->LUNListLength[3],
3100 hba[cntl_num]->num_luns);
3101 #endif /* CCISS_DEBUG */
3103 hba[cntl_num]->highest_lun = hba[cntl_num]->num_luns - 1;
3104 for (i = 0; i < CISS_MAX_LUN; i++) {
3105 if (i < hba[cntl_num]->num_luns) {
3106 lunid = (0xff & (unsigned int)(ld_buff->LUN[i][3]))
3107 << 24;
3108 lunid |= (0xff & (unsigned int)(ld_buff->LUN[i][2]))
3109 << 16;
3110 lunid |= (0xff & (unsigned int)(ld_buff->LUN[i][1]))
3111 << 8;
3112 lunid |= 0xff & (unsigned int)(ld_buff->LUN[i][0]);
3114 hba[cntl_num]->drv[i].LunID = lunid;
3116 #ifdef CCISS_DEBUG
3117 printk(KERN_DEBUG "LUN[%d]: %x %x %x %x = %x\n", i,
3118 ld_buff->LUN[i][0], ld_buff->LUN[i][1],
3119 ld_buff->LUN[i][2], ld_buff->LUN[i][3],
3120 hba[cntl_num]->drv[i].LunID);
3121 #endif /* CCISS_DEBUG */
3123 /* testing to see if 16-byte CDBs are already being used */
3124 if(hba[cntl_num]->cciss_read == CCISS_READ_16) {
3125 cciss_read_capacity_16(cntl_num, i, 0,
3126 &total_size, &block_size);
3127 goto geo_inq;
3129 cciss_read_capacity(cntl_num, i, 0, &total_size, &block_size);
3131 /* If read_capacity returns all F's the logical is >2TB */
3132 /* so we switch to 16-byte CDBs for all read/write ops */
3133 if(total_size == 0xFFFFFFFFULL) {
3134 cciss_read_capacity_16(cntl_num, i, 0,
3135 &total_size, &block_size);
3136 hba[cntl_num]->cciss_read = CCISS_READ_16;
3137 hba[cntl_num]->cciss_write = CCISS_WRITE_16;
3138 } else {
3139 hba[cntl_num]->cciss_read = CCISS_READ_10;
3140 hba[cntl_num]->cciss_write = CCISS_WRITE_10;
3142 geo_inq:
3143 cciss_geometry_inquiry(cntl_num, i, 0, total_size,
3144 block_size, inq_buff,
3145 &hba[cntl_num]->drv[i]);
3146 } else {
3147 /* initialize raid_level to indicate a free space */
3148 hba[cntl_num]->drv[i].raid_level = -1;
3151 kfree(ld_buff);
3152 kfree(inq_buff);
3155 /* Function to find the first free pointer into our hba[] array */
3156 /* Returns -1 if no free entries are left. */
3157 static int alloc_cciss_hba(void)
3159 int i;
3161 for (i = 0; i < MAX_CTLR; i++) {
3162 if (!hba[i]) {
3163 ctlr_info_t *p;
3164 p = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
3165 if (!p)
3166 goto Enomem;
3167 p->gendisk[0] = alloc_disk(1 << NWD_SHIFT);
3168 if (!p->gendisk[0])
3169 goto Enomem;
3170 hba[i] = p;
3171 return i;
3174 printk(KERN_WARNING "cciss: This driver supports a maximum"
3175 " of %d controllers.\n", MAX_CTLR);
3176 return -1;
3177 Enomem:
3178 printk(KERN_ERR "cciss: out of memory.\n");
3179 return -1;
3182 static void free_hba(int i)
3184 ctlr_info_t *p = hba[i];
3185 int n;
3187 hba[i] = NULL;
3188 for (n = 0; n < CISS_MAX_LUN; n++)
3189 put_disk(p->gendisk[n]);
3190 kfree(p);
3194 * This is it. Find all the controllers and register them. I really hate
3195 * stealing all these major device numbers.
3196 * returns the number of block devices registered.
3198 static int __devinit cciss_init_one(struct pci_dev *pdev,
3199 const struct pci_device_id *ent)
3201 int i;
3202 int j = 0;
3203 int rc;
3204 int dac;
3206 i = alloc_cciss_hba();
3207 if (i < 0)
3208 return -1;
3210 hba[i]->busy_initializing = 1;
3212 if (cciss_pci_init(hba[i], pdev) != 0)
3213 goto clean1;
3215 sprintf(hba[i]->devname, "cciss%d", i);
3216 hba[i]->ctlr = i;
3217 hba[i]->pdev = pdev;
3219 /* configure PCI DMA stuff */
3220 if (!pci_set_dma_mask(pdev, DMA_64BIT_MASK))
3221 dac = 1;
3222 else if (!pci_set_dma_mask(pdev, DMA_32BIT_MASK))
3223 dac = 0;
3224 else {
3225 printk(KERN_ERR "cciss: no suitable DMA available\n");
3226 goto clean1;
3230 * register with the major number, or get a dynamic major number
3231 * by passing 0 as argument. This is done for greater than
3232 * 8 controller support.
3234 if (i < MAX_CTLR_ORIG)
3235 hba[i]->major = COMPAQ_CISS_MAJOR + i;
3236 rc = register_blkdev(hba[i]->major, hba[i]->devname);
3237 if (rc == -EBUSY || rc == -EINVAL) {
3238 printk(KERN_ERR
3239 "cciss: Unable to get major number %d for %s "
3240 "on hba %d\n", hba[i]->major, hba[i]->devname, i);
3241 goto clean1;
3242 } else {
3243 if (i >= MAX_CTLR_ORIG)
3244 hba[i]->major = rc;
3247 /* make sure the board interrupts are off */
3248 hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_OFF);
3249 if (request_irq(hba[i]->intr[SIMPLE_MODE_INT], do_cciss_intr,
3250 IRQF_DISABLED | IRQF_SHARED, hba[i]->devname, hba[i])) {
3251 printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
3252 hba[i]->intr[SIMPLE_MODE_INT], hba[i]->devname);
3253 goto clean2;
3256 printk(KERN_INFO "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
3257 hba[i]->devname, pdev->device, pci_name(pdev),
3258 hba[i]->intr[SIMPLE_MODE_INT], dac ? "" : " not");
3260 hba[i]->cmd_pool_bits =
3261 kmalloc(((hba[i]->nr_cmds + BITS_PER_LONG -
3262 1) / BITS_PER_LONG) * sizeof(unsigned long), GFP_KERNEL);
3263 hba[i]->cmd_pool = (CommandList_struct *)
3264 pci_alloc_consistent(hba[i]->pdev,
3265 hba[i]->nr_cmds * sizeof(CommandList_struct),
3266 &(hba[i]->cmd_pool_dhandle));
3267 hba[i]->errinfo_pool = (ErrorInfo_struct *)
3268 pci_alloc_consistent(hba[i]->pdev,
3269 hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
3270 &(hba[i]->errinfo_pool_dhandle));
3271 if ((hba[i]->cmd_pool_bits == NULL)
3272 || (hba[i]->cmd_pool == NULL)
3273 || (hba[i]->errinfo_pool == NULL)) {
3274 printk(KERN_ERR "cciss: out of memory");
3275 goto clean4;
3277 #ifdef CONFIG_CISS_SCSI_TAPE
3278 hba[i]->scsi_rejects.complete =
3279 kmalloc(sizeof(hba[i]->scsi_rejects.complete[0]) *
3280 (hba[i]->nr_cmds + 5), GFP_KERNEL);
3281 if (hba[i]->scsi_rejects.complete == NULL) {
3282 printk(KERN_ERR "cciss: out of memory");
3283 goto clean4;
3285 #endif
3286 spin_lock_init(&hba[i]->lock);
3288 /* Initialize the pdev driver private data.
3289 have it point to hba[i]. */
3290 pci_set_drvdata(pdev, hba[i]);
3291 /* command and error info recs zeroed out before
3292 they are used */
3293 memset(hba[i]->cmd_pool_bits, 0,
3294 ((hba[i]->nr_cmds + BITS_PER_LONG -
3295 1) / BITS_PER_LONG) * sizeof(unsigned long));
3297 #ifdef CCISS_DEBUG
3298 printk(KERN_DEBUG "Scanning for drives on controller cciss%d\n", i);
3299 #endif /* CCISS_DEBUG */
3301 cciss_getgeometry(i);
3303 cciss_scsi_setup(i);
3305 /* Turn the interrupts on so we can service requests */
3306 hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_ON);
3308 cciss_procinit(i);
3310 hba[i]->cciss_max_sectors = 2048;
3312 hba[i]->busy_initializing = 0;
3314 do {
3315 drive_info_struct *drv = &(hba[i]->drv[j]);
3316 struct gendisk *disk = hba[i]->gendisk[j];
3317 request_queue_t *q;
3319 /* Check if the disk was allocated already */
3320 if (!disk){
3321 hba[i]->gendisk[j] = alloc_disk(1 << NWD_SHIFT);
3322 disk = hba[i]->gendisk[j];
3325 /* Check that the disk was able to be allocated */
3326 if (!disk) {
3327 printk(KERN_ERR "cciss: unable to allocate memory for disk %d\n", j);
3328 goto clean4;
3331 q = blk_init_queue(do_cciss_request, &hba[i]->lock);
3332 if (!q) {
3333 printk(KERN_ERR
3334 "cciss: unable to allocate queue for disk %d\n",
3336 goto clean4;
3338 drv->queue = q;
3340 q->backing_dev_info.ra_pages = READ_AHEAD;
3341 blk_queue_bounce_limit(q, hba[i]->pdev->dma_mask);
3343 /* This is a hardware imposed limit. */
3344 blk_queue_max_hw_segments(q, MAXSGENTRIES);
3346 /* This is a limit in the driver and could be eliminated. */
3347 blk_queue_max_phys_segments(q, MAXSGENTRIES);
3349 blk_queue_max_sectors(q, hba[i]->cciss_max_sectors);
3351 blk_queue_softirq_done(q, cciss_softirq_done);
3353 q->queuedata = hba[i];
3354 sprintf(disk->disk_name, "cciss/c%dd%d", i, j);
3355 disk->major = hba[i]->major;
3356 disk->first_minor = j << NWD_SHIFT;
3357 disk->fops = &cciss_fops;
3358 disk->queue = q;
3359 disk->private_data = drv;
3360 disk->driverfs_dev = &pdev->dev;
3361 /* we must register the controller even if no disks exist */
3362 /* this is for the online array utilities */
3363 if (!drv->heads && j)
3364 continue;
3365 blk_queue_hardsect_size(q, drv->block_size);
3366 set_capacity(disk, drv->nr_blocks);
3367 add_disk(disk);
3368 j++;
3369 } while (j <= hba[i]->highest_lun);
3371 return 1;
3373 clean4:
3374 #ifdef CONFIG_CISS_SCSI_TAPE
3375 kfree(hba[i]->scsi_rejects.complete);
3376 #endif
3377 kfree(hba[i]->cmd_pool_bits);
3378 if (hba[i]->cmd_pool)
3379 pci_free_consistent(hba[i]->pdev,
3380 hba[i]->nr_cmds * sizeof(CommandList_struct),
3381 hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
3382 if (hba[i]->errinfo_pool)
3383 pci_free_consistent(hba[i]->pdev,
3384 hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
3385 hba[i]->errinfo_pool,
3386 hba[i]->errinfo_pool_dhandle);
3387 free_irq(hba[i]->intr[SIMPLE_MODE_INT], hba[i]);
3388 clean2:
3389 unregister_blkdev(hba[i]->major, hba[i]->devname);
3390 clean1:
3391 hba[i]->busy_initializing = 0;
3392 /* cleanup any queues that may have been initialized */
3393 for (j=0; j <= hba[i]->highest_lun; j++){
3394 drive_info_struct *drv = &(hba[i]->drv[j]);
3395 if (drv->queue)
3396 blk_cleanup_queue(drv->queue);
3399 * Deliberately omit pci_disable_device(): it does something nasty to
3400 * Smart Array controllers that pci_enable_device does not undo
3402 pci_release_regions(pdev);
3403 pci_set_drvdata(pdev, NULL);
3404 free_hba(i);
3405 return -1;
3408 static void cciss_remove_one(struct pci_dev *pdev)
3410 ctlr_info_t *tmp_ptr;
3411 int i, j;
3412 char flush_buf[4];
3413 int return_code;
3415 if (pci_get_drvdata(pdev) == NULL) {
3416 printk(KERN_ERR "cciss: Unable to remove device \n");
3417 return;
3419 tmp_ptr = pci_get_drvdata(pdev);
3420 i = tmp_ptr->ctlr;
3421 if (hba[i] == NULL) {
3422 printk(KERN_ERR "cciss: device appears to "
3423 "already be removed \n");
3424 return;
3427 remove_proc_entry(hba[i]->devname, proc_cciss);
3428 unregister_blkdev(hba[i]->major, hba[i]->devname);
3430 /* remove it from the disk list */
3431 for (j = 0; j < CISS_MAX_LUN; j++) {
3432 struct gendisk *disk = hba[i]->gendisk[j];
3433 if (disk) {
3434 request_queue_t *q = disk->queue;
3436 if (disk->flags & GENHD_FL_UP)
3437 del_gendisk(disk);
3438 if (q)
3439 blk_cleanup_queue(q);
3443 cciss_unregister_scsi(i); /* unhook from SCSI subsystem */
3445 /* Turn board interrupts off and send the flush cache command */
3446 /* sendcmd will turn off interrupt, and send the flush...
3447 * To write all data in the battery backed cache to disks */
3448 memset(flush_buf, 0, 4);
3449 return_code = sendcmd(CCISS_CACHE_FLUSH, i, flush_buf, 4, 0, 0, 0, NULL,
3450 TYPE_CMD);
3451 if (return_code == IO_OK) {
3452 printk(KERN_INFO "Completed flushing cache on controller %d\n", i);
3453 } else {
3454 printk(KERN_WARNING "Error flushing cache on controller %d\n", i);
3456 free_irq(hba[i]->intr[2], hba[i]);
3458 #ifdef CONFIG_PCI_MSI
3459 if (hba[i]->msix_vector)
3460 pci_disable_msix(hba[i]->pdev);
3461 else if (hba[i]->msi_vector)
3462 pci_disable_msi(hba[i]->pdev);
3463 #endif /* CONFIG_PCI_MSI */
3465 iounmap(hba[i]->vaddr);
3467 pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(CommandList_struct),
3468 hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
3469 pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
3470 hba[i]->errinfo_pool, hba[i]->errinfo_pool_dhandle);
3471 kfree(hba[i]->cmd_pool_bits);
3472 #ifdef CONFIG_CISS_SCSI_TAPE
3473 kfree(hba[i]->scsi_rejects.complete);
3474 #endif
3476 * Deliberately omit pci_disable_device(): it does something nasty to
3477 * Smart Array controllers that pci_enable_device does not undo
3479 pci_release_regions(pdev);
3480 pci_set_drvdata(pdev, NULL);
3481 free_hba(i);
3484 static struct pci_driver cciss_pci_driver = {
3485 .name = "cciss",
3486 .probe = cciss_init_one,
3487 .remove = __devexit_p(cciss_remove_one),
3488 .id_table = cciss_pci_device_id, /* id_table */
3489 .shutdown = cciss_remove_one,
3493 * This is it. Register the PCI driver information for the cards we control
3494 * the OS will call our registered routines when it finds one of our cards.
3496 static int __init cciss_init(void)
3498 printk(KERN_INFO DRIVER_NAME "\n");
3500 /* Register for our PCI devices */
3501 return pci_register_driver(&cciss_pci_driver);
3504 static void __exit cciss_cleanup(void)
3506 int i;
3508 pci_unregister_driver(&cciss_pci_driver);
3509 /* double check that all controller entrys have been removed */
3510 for (i = 0; i < MAX_CTLR; i++) {
3511 if (hba[i] != NULL) {
3512 printk(KERN_WARNING "cciss: had to remove"
3513 " controller %d\n", i);
3514 cciss_remove_one(hba[i]->pdev);
3517 remove_proc_entry("cciss", proc_root_driver);
3520 static void fail_all_cmds(unsigned long ctlr)
3522 /* If we get here, the board is apparently dead. */
3523 ctlr_info_t *h = hba[ctlr];
3524 CommandList_struct *c;
3525 unsigned long flags;
3527 printk(KERN_WARNING "cciss%d: controller not responding.\n", h->ctlr);
3528 h->alive = 0; /* the controller apparently died... */
3530 spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
3532 pci_disable_device(h->pdev); /* Make sure it is really dead. */
3534 /* move everything off the request queue onto the completed queue */
3535 while ((c = h->reqQ) != NULL) {
3536 removeQ(&(h->reqQ), c);
3537 h->Qdepth--;
3538 addQ(&(h->cmpQ), c);
3541 /* Now, fail everything on the completed queue with a HW error */
3542 while ((c = h->cmpQ) != NULL) {
3543 removeQ(&h->cmpQ, c);
3544 c->err_info->CommandStatus = CMD_HARDWARE_ERR;
3545 if (c->cmd_type == CMD_RWREQ) {
3546 complete_command(h, c, 0);
3547 } else if (c->cmd_type == CMD_IOCTL_PEND)
3548 complete(c->waiting);
3549 #ifdef CONFIG_CISS_SCSI_TAPE
3550 else if (c->cmd_type == CMD_SCSI)
3551 complete_scsi_command(c, 0, 0);
3552 #endif
3554 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
3555 return;
3558 module_init(cciss_init);
3559 module_exit(cciss_cleanup);