5 * SoftDog 0.05: A Software Watchdog Device
7 * (c) Copyright 2007 Hewlett-Packard Development Company, L.P.
8 * Thomas Mingarelli <thomas.mingarelli@hp.com>
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * version 2 as published by the Free Software Foundation
16 #include <linux/device.h>
18 #include <linux/init.h>
19 #include <linux/interrupt.h>
21 #include <linux/irq.h>
22 #include <linux/nmi.h>
23 #include <linux/kernel.h>
24 #include <linux/miscdevice.h>
26 #include <linux/module.h>
27 #include <linux/kdebug.h>
28 #include <linux/moduleparam.h>
29 #include <linux/notifier.h>
30 #include <linux/pci.h>
31 #include <linux/pci_ids.h>
32 #include <linux/reboot.h>
33 #include <linux/sched.h>
34 #include <linux/timer.h>
35 #include <linux/types.h>
36 #include <linux/uaccess.h>
37 #include <linux/watchdog.h>
38 #include <linux/dmi.h>
39 #include <linux/efi.h>
40 #include <linux/string.h>
41 #include <linux/bootmem.h>
42 #include <linux/slab.h>
44 #include <asm/cacheflush.h>
46 #define PCI_BIOS32_SD_VALUE 0x5F32335F /* "_32_" */
47 #define CRU_BIOS_SIGNATURE_VALUE 0x55524324
48 #define PCI_BIOS32_PARAGRAPH_LEN 16
49 #define PCI_ROM_BASE1 0x000F0000
50 #define ROM_SIZE 0x10000
51 #define HPWDT_VERSION "1.1.1"
53 struct bios32_service_dir
{
63 struct smbios_cru64_info
{
72 #define SMBIOS_CRU64_INFORMATION 212
74 struct cmn_registers
{
114 } __attribute__((packed
));
116 #define DEFAULT_MARGIN 30
117 static unsigned int soft_margin
= DEFAULT_MARGIN
; /* in seconds */
118 static unsigned int reload
; /* the computed soft_margin */
119 static int nowayout
= WATCHDOG_NOWAYOUT
;
120 static char expect_release
;
121 static unsigned long hpwdt_is_open
;
122 static unsigned int allow_kdump
;
123 static unsigned int hpwdt_nmi_sourcing
;
124 static unsigned int priority
; /* hpwdt at end of die_notify list */
126 static void __iomem
*pci_mem_addr
; /* the PCI-memory address */
127 static unsigned long __iomem
*hpwdt_timer_reg
;
128 static unsigned long __iomem
*hpwdt_timer_con
;
130 static DEFINE_SPINLOCK(rom_lock
);
132 static void *cru_rom_addr
;
134 static struct cmn_registers cmn_regs
;
136 static struct pci_device_id hpwdt_devices
[] = {
137 { PCI_DEVICE(PCI_VENDOR_ID_COMPAQ
, 0xB203) },
138 { PCI_DEVICE(PCI_VENDOR_ID_HP
, 0x3306) },
139 {0}, /* terminate list */
141 MODULE_DEVICE_TABLE(pci
, hpwdt_devices
);
143 extern asmlinkage
void asminline_call(struct cmn_registers
*pi86Regs
,
144 unsigned long *pRomEntry
);
146 #ifndef CONFIG_X86_64
147 /* --32 Bit Bios------------------------------------------------------------ */
149 #define HPWDT_ARCH 32
153 "asminline_call: \n\t"
155 "movl %esp, %ebp \n\t"
161 "movl 8(%ebp),%eax \n\t"
162 "movl 4(%eax),%ebx \n\t"
163 "movl 8(%eax),%ecx \n\t"
164 "movl 12(%eax),%edx \n\t"
165 "movl 16(%eax),%esi \n\t"
166 "movl 20(%eax),%edi \n\t"
167 "movl (%eax),%eax \n\t"
169 "call *12(%ebp) \n\t"
172 "movl 8(%ebp),%eax \n\t"
173 "movl %ebx,4(%eax) \n\t"
174 "movl %ecx,8(%eax) \n\t"
175 "movl %edx,12(%eax) \n\t"
176 "movl %esi,16(%eax) \n\t"
177 "movl %edi,20(%eax) \n\t"
178 "movw %ds,24(%eax) \n\t"
179 "movw %es,26(%eax) \n\t"
181 "movl %ebx,(%eax) \n\t"
183 "movl %ebx,28(%eax) \n\t"
195 * Routine Description:
196 * This function uses the 32-bit BIOS Service Directory record to
197 * search for a $CRU record.
203 static int __devinit
cru_detect(unsigned long map_entry
,
204 unsigned long map_offset
)
207 unsigned long *bios32_entrypoint
;
208 unsigned long cru_physical_address
;
209 unsigned long cru_length
;
210 unsigned long physical_bios_base
= 0;
211 unsigned long physical_bios_offset
= 0;
212 int retval
= -ENODEV
;
214 bios32_map
= ioremap(map_entry
, (2 * PAGE_SIZE
));
216 if (bios32_map
== NULL
)
219 bios32_entrypoint
= bios32_map
+ map_offset
;
221 cmn_regs
.u1
.reax
= CRU_BIOS_SIGNATURE_VALUE
;
223 asminline_call(&cmn_regs
, bios32_entrypoint
);
225 if (cmn_regs
.u1
.ral
!= 0) {
227 "hpwdt: Call succeeded but with an error: 0x%x\n",
230 physical_bios_base
= cmn_regs
.u2
.rebx
;
231 physical_bios_offset
= cmn_regs
.u4
.redx
;
232 cru_length
= cmn_regs
.u3
.recx
;
233 cru_physical_address
=
234 physical_bios_base
+ physical_bios_offset
;
236 /* If the values look OK, then map it in. */
237 if ((physical_bios_base
+ physical_bios_offset
)) {
239 ioremap(cru_physical_address
, cru_length
);
244 printk(KERN_DEBUG
"hpwdt: CRU Base Address: 0x%lx\n",
246 printk(KERN_DEBUG
"hpwdt: CRU Offset Address: 0x%lx\n",
247 physical_bios_offset
);
248 printk(KERN_DEBUG
"hpwdt: CRU Length: 0x%lx\n",
250 printk(KERN_DEBUG
"hpwdt: CRU Mapped Address: 0x%x\n",
251 (unsigned int)&cru_rom_addr
);
260 static int __devinit
bios_checksum(const char __iomem
*ptr
, int len
)
266 * calculate checksum of size bytes. This should add up
267 * to zero if we have a valid header.
269 for (i
= 0; i
< len
; i
++)
272 return ((sum
== 0) && (len
> 0));
278 * Routine Description:
279 * This function finds the 32-bit BIOS Service Directory
285 static int __devinit
bios32_present(const char __iomem
*p
)
287 struct bios32_service_dir
*bios_32_ptr
;
289 unsigned long map_entry
, map_offset
;
291 bios_32_ptr
= (struct bios32_service_dir
*) p
;
294 * Search for signature by checking equal to the swizzled value
295 * instead of calling another routine to perform a strcmp.
297 if (bios_32_ptr
->signature
== PCI_BIOS32_SD_VALUE
) {
298 length
= bios_32_ptr
->length
* PCI_BIOS32_PARAGRAPH_LEN
;
299 if (bios_checksum(p
, length
)) {
301 * According to the spec, we're looking for the
302 * first 4KB-aligned address below the entrypoint
303 * listed in the header. The Service Directory code
304 * is guaranteed to occupy no more than 2 4KB pages.
306 map_entry
= bios_32_ptr
->entry_point
& ~(PAGE_SIZE
- 1);
307 map_offset
= bios_32_ptr
->entry_point
- map_entry
;
309 return cru_detect(map_entry
, map_offset
);
315 static int __devinit
detect_cru_service(void)
321 * Search from 0x0f0000 through 0x0fffff, inclusive.
323 p
= ioremap(PCI_ROM_BASE1
, ROM_SIZE
);
327 for (q
= p
; q
< p
+ ROM_SIZE
; q
+= 16) {
328 rc
= bios32_present(q
);
337 /* --64 Bit Bios------------------------------------------------------------ */
339 #define HPWDT_ARCH 64
343 "asminline_call: \n\t"
345 "movq %rsp, %rbp \n\t"
351 "movq %rsi, %r12 \n\t"
352 "movq %rdi, %r9 \n\t"
353 "movl 4(%r9),%ebx \n\t"
354 "movl 8(%r9),%ecx \n\t"
355 "movl 12(%r9),%edx \n\t"
356 "movl 16(%r9),%esi \n\t"
357 "movl 20(%r9),%edi \n\t"
358 "movl (%r9),%eax \n\t"
362 "movl %eax, (%r9) \n\t"
363 "movl %ebx, 4(%r9) \n\t"
364 "movl %ecx, 8(%r9) \n\t"
365 "movl %edx, 12(%r9) \n\t"
366 "movl %esi, 16(%r9) \n\t"
367 "movl %edi, 20(%r9) \n\t"
368 "movq %r12, %rax \n\t"
369 "movl %eax, 28(%r9) \n\t"
382 * Routine Description:
383 * This function checks whether or not a SMBIOS/DMI record is
384 * the 64bit CRU info or not
386 static void __devinit
dmi_find_cru(const struct dmi_header
*dm
, void *dummy
)
388 struct smbios_cru64_info
*smbios_cru64_ptr
;
389 unsigned long cru_physical_address
;
391 if (dm
->type
== SMBIOS_CRU64_INFORMATION
) {
392 smbios_cru64_ptr
= (struct smbios_cru64_info
*) dm
;
393 if (smbios_cru64_ptr
->signature
== CRU_BIOS_SIGNATURE_VALUE
) {
394 cru_physical_address
=
395 smbios_cru64_ptr
->physical_address
+
396 smbios_cru64_ptr
->double_offset
;
397 cru_rom_addr
= ioremap(cru_physical_address
,
398 smbios_cru64_ptr
->double_length
);
399 set_memory_x((unsigned long)cru_rom_addr
& PAGE_MASK
,
400 smbios_cru64_ptr
->double_length
>> PAGE_SHIFT
);
405 static int __devinit
detect_cru_service(void)
409 dmi_walk(dmi_find_cru
, NULL
);
411 /* if cru_rom_addr has been set then we found a CRU service */
412 return ((cru_rom_addr
!= NULL
) ? 0 : -ENODEV
);
415 /* ------------------------------------------------------------------------- */
420 * Watchdog operations
422 static void hpwdt_start(void)
424 reload
= (soft_margin
* 1000) / 128;
425 iowrite16(reload
, hpwdt_timer_reg
);
426 iowrite16(0x85, hpwdt_timer_con
);
429 static void hpwdt_stop(void)
433 data
= ioread16(hpwdt_timer_con
);
435 iowrite16(data
, hpwdt_timer_con
);
438 static void hpwdt_ping(void)
440 iowrite16(reload
, hpwdt_timer_reg
);
443 static int hpwdt_change_timer(int new_margin
)
445 /* Arbitrary, can't find the card's limits */
446 if (new_margin
< 30 || new_margin
> 600) {
448 "hpwdt: New value passed in is invalid: %d seconds.\n",
453 soft_margin
= new_margin
;
455 "hpwdt: New timer passed in is %d seconds.\n",
457 reload
= (soft_margin
* 1000) / 128;
465 static int hpwdt_pretimeout(struct notifier_block
*nb
, unsigned long ulReason
,
468 unsigned long rom_pl
;
469 static int die_nmi_called
;
471 if (ulReason
!= DIE_NMI
&& ulReason
!= DIE_NMI_IPI
)
474 if (hpwdt_nmi_sourcing
) {
475 spin_lock_irqsave(&rom_lock
, rom_pl
);
477 asminline_call(&cmn_regs
, cru_rom_addr
);
479 spin_unlock_irqrestore(&rom_lock
, rom_pl
);
480 if (cmn_regs
.u1
.ral
== 0) {
481 printk(KERN_WARNING
"hpwdt: An NMI occurred, "
482 "but unable to determine source.\n");
486 panic("An NMI occurred, please see the Integrated "
487 "Management Log for details.\n");
494 * /dev/watchdog handling
496 static int hpwdt_open(struct inode
*inode
, struct file
*file
)
498 /* /dev/watchdog can only be opened once */
499 if (test_and_set_bit(0, &hpwdt_is_open
))
502 /* Start the watchdog */
506 return nonseekable_open(inode
, file
);
509 static int hpwdt_release(struct inode
*inode
, struct file
*file
)
511 /* Stop the watchdog */
512 if (expect_release
== 42) {
516 "hpwdt: Unexpected close, not stopping watchdog!\n");
522 /* /dev/watchdog is being closed, make sure it can be re-opened */
523 clear_bit(0, &hpwdt_is_open
);
528 static ssize_t
hpwdt_write(struct file
*file
, const char __user
*data
,
529 size_t len
, loff_t
*ppos
)
531 /* See if we got the magic character 'V' and reload the timer */
536 /* note: just in case someone wrote the magic character
537 * five months ago... */
540 /* scan to see whether or not we got the magic char. */
541 for (i
= 0; i
!= len
; i
++) {
543 if (get_user(c
, data
+ i
))
550 /* someone wrote to us, we should reload the timer */
557 static struct watchdog_info ident
= {
558 .options
= WDIOF_SETTIMEOUT
|
559 WDIOF_KEEPALIVEPING
|
561 .identity
= "HP iLO2 HW Watchdog Timer",
564 static long hpwdt_ioctl(struct file
*file
, unsigned int cmd
,
567 void __user
*argp
= (void __user
*)arg
;
568 int __user
*p
= argp
;
573 case WDIOC_GETSUPPORT
:
575 if (copy_to_user(argp
, &ident
, sizeof(ident
)))
579 case WDIOC_GETSTATUS
:
580 case WDIOC_GETBOOTSTATUS
:
581 ret
= put_user(0, p
);
584 case WDIOC_KEEPALIVE
:
589 case WDIOC_SETTIMEOUT
:
590 ret
= get_user(new_margin
, p
);
594 ret
= hpwdt_change_timer(new_margin
);
600 case WDIOC_GETTIMEOUT
:
601 ret
= put_user(soft_margin
, p
);
610 static const struct file_operations hpwdt_fops
= {
611 .owner
= THIS_MODULE
,
613 .write
= hpwdt_write
,
614 .unlocked_ioctl
= hpwdt_ioctl
,
616 .release
= hpwdt_release
,
619 static struct miscdevice hpwdt_miscdev
= {
620 .minor
= WATCHDOG_MINOR
,
625 static struct notifier_block die_notifier
= {
626 .notifier_call
= hpwdt_pretimeout
,
634 #ifdef ARCH_HAS_NMI_WATCHDOG
635 static void __devinit
hpwdt_check_nmi_sourcing(struct pci_dev
*dev
)
638 * If nmi_watchdog is turned off then we can turn on
639 * our nmi sourcing capability.
641 if (!nmi_watchdog_active())
642 hpwdt_nmi_sourcing
= 1;
644 dev_warn(&dev
->dev
, "NMI sourcing is disabled. To enable this "
645 "functionality you must reboot with nmi_watchdog=0 "
646 "and load the hpwdt driver with priority=1.\n");
649 static void __devinit
hpwdt_check_nmi_sourcing(struct pci_dev
*dev
)
651 dev_warn(&dev
->dev
, "NMI sourcing is disabled. "
652 "Your kernel does not support a NMI Watchdog.\n");
656 static int __devinit
hpwdt_init_one(struct pci_dev
*dev
,
657 const struct pci_device_id
*ent
)
662 * Check if we can do NMI sourcing or not
664 hpwdt_check_nmi_sourcing(dev
);
667 * First let's find out if we are on an iLO2 server. We will
668 * not run on a legacy ASM box.
669 * So we only support the G5 ProLiant servers and higher.
671 if (dev
->subsystem_vendor
!= PCI_VENDOR_ID_HP
) {
673 "This server does not have an iLO2 ASIC.\n");
677 if (pci_enable_device(dev
)) {
679 "Not possible to enable PCI Device: 0x%x:0x%x.\n",
680 ent
->vendor
, ent
->device
);
684 pci_mem_addr
= pci_iomap(dev
, 1, 0x80);
687 "Unable to detect the iLO2 server memory.\n");
689 goto error_pci_iomap
;
691 hpwdt_timer_reg
= pci_mem_addr
+ 0x70;
692 hpwdt_timer_con
= pci_mem_addr
+ 0x72;
694 /* Make sure that we have a valid soft_margin */
695 if (hpwdt_change_timer(soft_margin
))
696 hpwdt_change_timer(DEFAULT_MARGIN
);
699 * We need to map the ROM to get the CRU service.
700 * For 32 bit Operating Systems we need to go through the 32 Bit
701 * BIOS Service Directory
702 * For 64 bit Operating Systems we get that service through SMBIOS.
704 retval
= detect_cru_service();
707 "Unable to detect the %d Bit CRU Service.\n",
713 * We know this is the only CRU call we need to make so lets keep as
714 * few instructions as possible once the NMI comes in.
716 cmn_regs
.u1
.rah
= 0x0D;
717 cmn_regs
.u1
.ral
= 0x02;
720 * If the priority is set to 1, then we will be put first on the
721 * die notify list to handle a critical NMI. The default is to
722 * be last so other users of the NMI signal can function.
725 die_notifier
.priority
= 0x7FFFFFFF;
727 retval
= register_die_notifier(&die_notifier
);
730 "Unable to register a die notifier (err=%d).\n",
732 goto error_die_notifier
;
735 retval
= misc_register(&hpwdt_miscdev
);
738 "Unable to register miscdev on minor=%d (err=%d).\n",
739 WATCHDOG_MINOR
, retval
);
740 goto error_misc_register
;
744 "hp Watchdog Timer Driver: %s"
745 ", timer margin: %d seconds (nowayout=%d)"
746 ", allow kernel dump: %s (default = 0/OFF)"
747 ", priority: %s (default = 0/LAST).\n",
748 HPWDT_VERSION
, soft_margin
, nowayout
,
749 (allow_kdump
== 0) ? "OFF" : "ON",
750 (priority
== 0) ? "LAST" : "FIRST");
755 unregister_die_notifier(&die_notifier
);
758 iounmap(cru_rom_addr
);
760 pci_iounmap(dev
, pci_mem_addr
);
762 pci_disable_device(dev
);
766 static void __devexit
hpwdt_exit(struct pci_dev
*dev
)
771 misc_deregister(&hpwdt_miscdev
);
772 unregister_die_notifier(&die_notifier
);
775 iounmap(cru_rom_addr
);
776 pci_iounmap(dev
, pci_mem_addr
);
777 pci_disable_device(dev
);
780 static struct pci_driver hpwdt_driver
= {
782 .id_table
= hpwdt_devices
,
783 .probe
= hpwdt_init_one
,
784 .remove
= __devexit_p(hpwdt_exit
),
787 static void __exit
hpwdt_cleanup(void)
789 pci_unregister_driver(&hpwdt_driver
);
792 static int __init
hpwdt_init(void)
794 return pci_register_driver(&hpwdt_driver
);
797 MODULE_AUTHOR("Tom Mingarelli");
798 MODULE_DESCRIPTION("hp watchdog driver");
799 MODULE_LICENSE("GPL");
800 MODULE_VERSION(HPWDT_VERSION
);
801 MODULE_ALIAS_MISCDEV(WATCHDOG_MINOR
);
803 module_param(soft_margin
, int, 0);
804 MODULE_PARM_DESC(soft_margin
, "Watchdog timeout in seconds");
806 module_param(allow_kdump
, int, 0);
807 MODULE_PARM_DESC(allow_kdump
, "Start a kernel dump after NMI occurs");
809 module_param(nowayout
, int, 0);
810 MODULE_PARM_DESC(nowayout
, "Watchdog cannot be stopped once started (default="
811 __MODULE_STRING(WATCHDOG_NOWAYOUT
) ")");
813 module_param(priority
, int, 0);
814 MODULE_PARM_DESC(priority
, "The hpwdt driver handles NMIs first or last"
815 " (default = 0/Last)\n");
817 module_init(hpwdt_init
);
818 module_exit(hpwdt_cleanup
);