5 * SoftDog 0.05: A Software Watchdog Device
7 * (c) Copyright 2015 Hewlett Packard Enterprise Development LP
8 * Thomas Mingarelli <thomas.mingarelli@hpe.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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
18 #include <linux/device.h>
21 #include <linux/bitops.h>
22 #include <linux/kernel.h>
23 #include <linux/miscdevice.h>
24 #include <linux/module.h>
25 #include <linux/moduleparam.h>
26 #include <linux/pci.h>
27 #include <linux/pci_ids.h>
28 #include <linux/types.h>
29 #include <linux/uaccess.h>
30 #include <linux/watchdog.h>
31 #ifdef CONFIG_HPWDT_NMI_DECODING
32 #include <linux/dmi.h>
33 #include <linux/spinlock.h>
34 #include <linux/nmi.h>
35 #include <linux/kdebug.h>
36 #include <linux/notifier.h>
37 #include <asm/set_memory.h>
38 #endif /* CONFIG_HPWDT_NMI_DECODING */
40 #include <asm/frame.h>
42 #define HPWDT_VERSION "1.4.0"
43 #define SECS_TO_TICKS(secs) ((secs) * 1000 / 128)
44 #define TICKS_TO_SECS(ticks) ((ticks) * 128 / 1000)
45 #define HPWDT_MAX_TIMER TICKS_TO_SECS(65535)
46 #define DEFAULT_MARGIN 30
48 static unsigned int soft_margin
= DEFAULT_MARGIN
; /* in seconds */
49 static unsigned int reload
; /* the computed soft_margin */
50 static bool nowayout
= WATCHDOG_NOWAYOUT
;
51 static char expect_release
;
52 static unsigned long hpwdt_is_open
;
54 static void __iomem
*pci_mem_addr
; /* the PCI-memory address */
55 static unsigned long __iomem
*hpwdt_nmistat
;
56 static unsigned long __iomem
*hpwdt_timer_reg
;
57 static unsigned long __iomem
*hpwdt_timer_con
;
59 static const struct pci_device_id hpwdt_devices
[] = {
60 { PCI_DEVICE(PCI_VENDOR_ID_COMPAQ
, 0xB203) }, /* iLO2 */
61 { PCI_DEVICE(PCI_VENDOR_ID_HP
, 0x3306) }, /* iLO3 */
62 {0}, /* terminate list */
64 MODULE_DEVICE_TABLE(pci
, hpwdt_devices
);
66 #ifdef CONFIG_HPWDT_NMI_DECODING
67 #define PCI_BIOS32_SD_VALUE 0x5F32335F /* "_32_" */
68 #define CRU_BIOS_SIGNATURE_VALUE 0x55524324
69 #define PCI_BIOS32_PARAGRAPH_LEN 16
70 #define PCI_ROM_BASE1 0x000F0000
71 #define ROM_SIZE 0x10000
73 struct bios32_service_dir
{
83 struct smbios_cru64_info
{
92 #define SMBIOS_CRU64_INFORMATION 212
95 struct smbios_proliant_info
{
104 #define SMBIOS_ICRU_INFORMATION 219
107 struct cmn_registers
{
147 } __attribute__((packed
));
149 static unsigned int hpwdt_nmi_decoding
;
150 static unsigned int allow_kdump
= 1;
151 static unsigned int is_icru
;
152 static unsigned int is_uefi
;
153 static DEFINE_SPINLOCK(rom_lock
);
154 static void *cru_rom_addr
;
155 static struct cmn_registers cmn_regs
;
157 extern asmlinkage
void asminline_call(struct cmn_registers
*pi86Regs
,
158 unsigned long *pRomEntry
);
161 /* --32 Bit Bios------------------------------------------------------------ */
163 #define HPWDT_ARCH 32
167 ".globl asminline_call \n"
168 "asminline_call: \n\t"
170 "movl %esp, %ebp \n\t"
176 "movl 8(%ebp),%eax \n\t"
177 "movl 4(%eax),%ebx \n\t"
178 "movl 8(%eax),%ecx \n\t"
179 "movl 12(%eax),%edx \n\t"
180 "movl 16(%eax),%esi \n\t"
181 "movl 20(%eax),%edi \n\t"
182 "movl (%eax),%eax \n\t"
184 "call *12(%ebp) \n\t"
187 "movl 8(%ebp),%eax \n\t"
188 "movl %ebx,4(%eax) \n\t"
189 "movl %ecx,8(%eax) \n\t"
190 "movl %edx,12(%eax) \n\t"
191 "movl %esi,16(%eax) \n\t"
192 "movl %edi,20(%eax) \n\t"
193 "movw %ds,24(%eax) \n\t"
194 "movw %es,26(%eax) \n\t"
196 "movl %ebx,(%eax) \n\t"
198 "movl %ebx,28(%eax) \n\t"
210 * Routine Description:
211 * This function uses the 32-bit BIOS Service Directory record to
212 * search for a $CRU record.
218 static int cru_detect(unsigned long map_entry
,
219 unsigned long map_offset
)
222 unsigned long *bios32_entrypoint
;
223 unsigned long cru_physical_address
;
224 unsigned long cru_length
;
225 unsigned long physical_bios_base
= 0;
226 unsigned long physical_bios_offset
= 0;
227 int retval
= -ENODEV
;
229 bios32_map
= ioremap(map_entry
, (2 * PAGE_SIZE
));
231 if (bios32_map
== NULL
)
234 bios32_entrypoint
= bios32_map
+ map_offset
;
236 cmn_regs
.u1
.reax
= CRU_BIOS_SIGNATURE_VALUE
;
238 set_memory_x((unsigned long)bios32_map
, 2);
239 asminline_call(&cmn_regs
, bios32_entrypoint
);
241 if (cmn_regs
.u1
.ral
!= 0) {
242 pr_warn("Call succeeded but with an error: 0x%x\n",
245 physical_bios_base
= cmn_regs
.u2
.rebx
;
246 physical_bios_offset
= cmn_regs
.u4
.redx
;
247 cru_length
= cmn_regs
.u3
.recx
;
248 cru_physical_address
=
249 physical_bios_base
+ physical_bios_offset
;
251 /* If the values look OK, then map it in. */
252 if ((physical_bios_base
+ physical_bios_offset
)) {
254 ioremap(cru_physical_address
, cru_length
);
256 set_memory_x((unsigned long)cru_rom_addr
& PAGE_MASK
,
257 (cru_length
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
);
262 pr_debug("CRU Base Address: 0x%lx\n", physical_bios_base
);
263 pr_debug("CRU Offset Address: 0x%lx\n", physical_bios_offset
);
264 pr_debug("CRU Length: 0x%lx\n", cru_length
);
265 pr_debug("CRU Mapped Address: %p\n", &cru_rom_addr
);
274 static int bios_checksum(const char __iomem
*ptr
, int len
)
280 * calculate checksum of size bytes. This should add up
281 * to zero if we have a valid header.
283 for (i
= 0; i
< len
; i
++)
286 return ((sum
== 0) && (len
> 0));
292 * Routine Description:
293 * This function finds the 32-bit BIOS Service Directory
299 static int bios32_present(const char __iomem
*p
)
301 struct bios32_service_dir
*bios_32_ptr
;
303 unsigned long map_entry
, map_offset
;
305 bios_32_ptr
= (struct bios32_service_dir
*) p
;
308 * Search for signature by checking equal to the swizzled value
309 * instead of calling another routine to perform a strcmp.
311 if (bios_32_ptr
->signature
== PCI_BIOS32_SD_VALUE
) {
312 length
= bios_32_ptr
->length
* PCI_BIOS32_PARAGRAPH_LEN
;
313 if (bios_checksum(p
, length
)) {
315 * According to the spec, we're looking for the
316 * first 4KB-aligned address below the entrypoint
317 * listed in the header. The Service Directory code
318 * is guaranteed to occupy no more than 2 4KB pages.
320 map_entry
= bios_32_ptr
->entry_point
& ~(PAGE_SIZE
- 1);
321 map_offset
= bios_32_ptr
->entry_point
- map_entry
;
323 return cru_detect(map_entry
, map_offset
);
329 static int detect_cru_service(void)
335 * Search from 0x0f0000 through 0x0fffff, inclusive.
337 p
= ioremap(PCI_ROM_BASE1
, ROM_SIZE
);
341 for (q
= p
; q
< p
+ ROM_SIZE
; q
+= 16) {
342 rc
= bios32_present(q
);
349 /* ------------------------------------------------------------------------- */
350 #endif /* CONFIG_X86_32 */
352 /* --64 Bit Bios------------------------------------------------------------ */
354 #define HPWDT_ARCH 64
358 ".globl asminline_call \n\t"
359 ".type asminline_call, @function \n\t"
360 "asminline_call: \n\t"
367 "movq %rsi, %r12 \n\t"
368 "movq %rdi, %r9 \n\t"
369 "movl 4(%r9),%ebx \n\t"
370 "movl 8(%r9),%ecx \n\t"
371 "movl 12(%r9),%edx \n\t"
372 "movl 16(%r9),%esi \n\t"
373 "movl 20(%r9),%edi \n\t"
374 "movl (%r9),%eax \n\t"
378 "movl %eax, (%r9) \n\t"
379 "movl %ebx, 4(%r9) \n\t"
380 "movl %ecx, 8(%r9) \n\t"
381 "movl %edx, 12(%r9) \n\t"
382 "movl %esi, 16(%r9) \n\t"
383 "movl %edi, 20(%r9) \n\t"
384 "movq %r12, %rax \n\t"
385 "movl %eax, 28(%r9) \n\t"
398 * Routine Description:
399 * This function checks whether or not a SMBIOS/DMI record is
400 * the 64bit CRU info or not
402 static void dmi_find_cru(const struct dmi_header
*dm
, void *dummy
)
404 struct smbios_cru64_info
*smbios_cru64_ptr
;
405 unsigned long cru_physical_address
;
407 if (dm
->type
== SMBIOS_CRU64_INFORMATION
) {
408 smbios_cru64_ptr
= (struct smbios_cru64_info
*) dm
;
409 if (smbios_cru64_ptr
->signature
== CRU_BIOS_SIGNATURE_VALUE
) {
410 cru_physical_address
=
411 smbios_cru64_ptr
->physical_address
+
412 smbios_cru64_ptr
->double_offset
;
413 cru_rom_addr
= ioremap(cru_physical_address
,
414 smbios_cru64_ptr
->double_length
);
415 set_memory_x((unsigned long)cru_rom_addr
& PAGE_MASK
,
416 smbios_cru64_ptr
->double_length
>> PAGE_SHIFT
);
421 static int detect_cru_service(void)
425 dmi_walk(dmi_find_cru
, NULL
);
427 /* if cru_rom_addr has been set then we found a CRU service */
428 return ((cru_rom_addr
!= NULL
) ? 0 : -ENODEV
);
430 /* ------------------------------------------------------------------------- */
431 #endif /* CONFIG_X86_64 */
432 #endif /* CONFIG_HPWDT_NMI_DECODING */
435 * Watchdog operations
437 static void hpwdt_start(void)
439 reload
= SECS_TO_TICKS(soft_margin
);
440 iowrite16(reload
, hpwdt_timer_reg
);
441 iowrite8(0x85, hpwdt_timer_con
);
444 static void hpwdt_stop(void)
448 data
= ioread8(hpwdt_timer_con
);
450 iowrite8(data
, hpwdt_timer_con
);
453 static void hpwdt_ping(void)
455 iowrite16(reload
, hpwdt_timer_reg
);
458 static int hpwdt_change_timer(int new_margin
)
460 if (new_margin
< 1 || new_margin
> HPWDT_MAX_TIMER
) {
461 pr_warn("New value passed in is invalid: %d seconds\n",
466 soft_margin
= new_margin
;
467 pr_debug("New timer passed in is %d seconds\n", new_margin
);
468 reload
= SECS_TO_TICKS(soft_margin
);
473 static int hpwdt_time_left(void)
475 return TICKS_TO_SECS(ioread16(hpwdt_timer_reg
));
478 #ifdef CONFIG_HPWDT_NMI_DECODING
479 static int hpwdt_my_nmi(void)
481 return ioread8(hpwdt_nmistat
) & 0x6;
487 static int hpwdt_pretimeout(unsigned int ulReason
, struct pt_regs
*regs
)
489 unsigned long rom_pl
;
490 static int die_nmi_called
;
492 if (!hpwdt_nmi_decoding
)
495 if ((ulReason
== NMI_UNKNOWN
) && !hpwdt_my_nmi())
498 spin_lock_irqsave(&rom_lock
, rom_pl
);
499 if (!die_nmi_called
&& !is_icru
&& !is_uefi
)
500 asminline_call(&cmn_regs
, cru_rom_addr
);
502 spin_unlock_irqrestore(&rom_lock
, rom_pl
);
507 if (!is_icru
&& !is_uefi
) {
508 if (cmn_regs
.u1
.ral
== 0) {
509 nmi_panic(regs
, "An NMI occurred, but unable to determine source.\n");
513 nmi_panic(regs
, "An NMI occurred. Depending on your system the reason "
514 "for the NMI is logged in any one of the following "
516 "1. Integrated Management Log (IML)\n"
518 "3. OA Forward Progress Log\n"
523 #endif /* CONFIG_HPWDT_NMI_DECODING */
526 * /dev/watchdog handling
528 static int hpwdt_open(struct inode
*inode
, struct file
*file
)
530 /* /dev/watchdog can only be opened once */
531 if (test_and_set_bit(0, &hpwdt_is_open
))
534 /* Start the watchdog */
538 return nonseekable_open(inode
, file
);
541 static int hpwdt_release(struct inode
*inode
, struct file
*file
)
543 /* Stop the watchdog */
544 if (expect_release
== 42) {
547 pr_crit("Unexpected close, not stopping watchdog!\n");
553 /* /dev/watchdog is being closed, make sure it can be re-opened */
554 clear_bit(0, &hpwdt_is_open
);
559 static ssize_t
hpwdt_write(struct file
*file
, const char __user
*data
,
560 size_t len
, loff_t
*ppos
)
562 /* See if we got the magic character 'V' and reload the timer */
567 /* note: just in case someone wrote the magic character
568 * five months ago... */
571 /* scan to see whether or not we got the magic char. */
572 for (i
= 0; i
!= len
; i
++) {
574 if (get_user(c
, data
+ i
))
581 /* someone wrote to us, we should reload the timer */
588 static const struct watchdog_info ident
= {
589 .options
= WDIOF_SETTIMEOUT
|
590 WDIOF_KEEPALIVEPING
|
592 .identity
= "HPE iLO2+ HW Watchdog Timer",
595 static long hpwdt_ioctl(struct file
*file
, unsigned int cmd
,
598 void __user
*argp
= (void __user
*)arg
;
599 int __user
*p
= argp
;
600 int new_margin
, options
;
604 case WDIOC_GETSUPPORT
:
606 if (copy_to_user(argp
, &ident
, sizeof(ident
)))
610 case WDIOC_GETSTATUS
:
611 case WDIOC_GETBOOTSTATUS
:
612 ret
= put_user(0, p
);
615 case WDIOC_KEEPALIVE
:
620 case WDIOC_SETOPTIONS
:
621 ret
= get_user(options
, p
);
625 if (options
& WDIOS_DISABLECARD
)
628 if (options
& WDIOS_ENABLECARD
) {
634 case WDIOC_SETTIMEOUT
:
635 ret
= get_user(new_margin
, p
);
639 ret
= hpwdt_change_timer(new_margin
);
645 case WDIOC_GETTIMEOUT
:
646 ret
= put_user(soft_margin
, p
);
649 case WDIOC_GETTIMELEFT
:
650 ret
= put_user(hpwdt_time_left(), p
);
659 static const struct file_operations hpwdt_fops
= {
660 .owner
= THIS_MODULE
,
662 .write
= hpwdt_write
,
663 .unlocked_ioctl
= hpwdt_ioctl
,
665 .release
= hpwdt_release
,
668 static struct miscdevice hpwdt_miscdev
= {
669 .minor
= WATCHDOG_MINOR
,
678 #ifdef CONFIG_HPWDT_NMI_DECODING
679 #ifdef CONFIG_X86_LOCAL_APIC
680 static void hpwdt_check_nmi_decoding(struct pci_dev
*dev
)
683 * If nmi_watchdog is turned off then we can turn on
684 * our nmi decoding capability.
686 hpwdt_nmi_decoding
= 1;
689 static void hpwdt_check_nmi_decoding(struct pci_dev
*dev
)
691 dev_warn(&dev
->dev
, "NMI decoding is disabled. "
692 "Your kernel does not support a NMI Watchdog.\n");
694 #endif /* CONFIG_X86_LOCAL_APIC */
699 * Routine Description:
700 * This function checks whether or not we are on an iCRU-based server.
701 * This check is independent of architecture and needs to be made for
702 * any ProLiant system.
704 static void dmi_find_icru(const struct dmi_header
*dm
, void *dummy
)
706 struct smbios_proliant_info
*smbios_proliant_ptr
;
708 if (dm
->type
== SMBIOS_ICRU_INFORMATION
) {
709 smbios_proliant_ptr
= (struct smbios_proliant_info
*) dm
;
710 if (smbios_proliant_ptr
->misc_features
& 0x01)
712 if (smbios_proliant_ptr
->misc_features
& 0x1400)
717 static int hpwdt_init_nmi_decoding(struct pci_dev
*dev
)
722 * On typical CRU-based systems we need to map that service in
723 * the BIOS. For 32 bit Operating Systems we need to go through
724 * the 32 Bit BIOS Service Directory. For 64 bit Operating
725 * Systems we get that service through SMBIOS.
727 * On systems that support the new iCRU service all we need to
728 * do is call dmi_walk to get the supported flag value and skip
729 * the old cru detect code.
731 dmi_walk(dmi_find_icru
, NULL
);
732 if (!is_icru
&& !is_uefi
) {
735 * We need to map the ROM to get the CRU service.
736 * For 32 bit Operating Systems we need to go through the 32 Bit
737 * BIOS Service Directory
738 * For 64 bit Operating Systems we get that service through SMBIOS.
740 retval
= detect_cru_service();
743 "Unable to detect the %d Bit CRU Service.\n",
749 * We know this is the only CRU call we need to make so lets keep as
750 * few instructions as possible once the NMI comes in.
752 cmn_regs
.u1
.rah
= 0x0D;
753 cmn_regs
.u1
.ral
= 0x02;
757 * Only one function can register for NMI_UNKNOWN
759 retval
= register_nmi_handler(NMI_UNKNOWN
, hpwdt_pretimeout
, 0, "hpwdt");
762 retval
= register_nmi_handler(NMI_SERR
, hpwdt_pretimeout
, 0, "hpwdt");
765 retval
= register_nmi_handler(NMI_IO_CHECK
, hpwdt_pretimeout
, 0, "hpwdt");
770 "HPE Watchdog Timer Driver: NMI decoding initialized"
771 ", allow kernel dump: %s (default = 1/ON)\n",
772 (allow_kdump
== 0) ? "OFF" : "ON");
776 unregister_nmi_handler(NMI_SERR
, "hpwdt");
778 unregister_nmi_handler(NMI_UNKNOWN
, "hpwdt");
781 "Unable to register a die notifier (err=%d).\n",
784 iounmap(cru_rom_addr
);
788 static void hpwdt_exit_nmi_decoding(void)
790 unregister_nmi_handler(NMI_UNKNOWN
, "hpwdt");
791 unregister_nmi_handler(NMI_SERR
, "hpwdt");
792 unregister_nmi_handler(NMI_IO_CHECK
, "hpwdt");
794 iounmap(cru_rom_addr
);
796 #else /* !CONFIG_HPWDT_NMI_DECODING */
797 static void hpwdt_check_nmi_decoding(struct pci_dev
*dev
)
801 static int hpwdt_init_nmi_decoding(struct pci_dev
*dev
)
806 static void hpwdt_exit_nmi_decoding(void)
809 #endif /* CONFIG_HPWDT_NMI_DECODING */
811 static int hpwdt_init_one(struct pci_dev
*dev
,
812 const struct pci_device_id
*ent
)
817 * Check if we can do NMI decoding or not
819 hpwdt_check_nmi_decoding(dev
);
822 * First let's find out if we are on an iLO2+ server. We will
823 * not run on a legacy ASM box.
824 * So we only support the G5 ProLiant servers and higher.
826 if (dev
->subsystem_vendor
!= PCI_VENDOR_ID_HP
&&
827 dev
->subsystem_vendor
!= PCI_VENDOR_ID_HP_3PAR
) {
829 "This server does not have an iLO2+ ASIC.\n");
834 * Ignore all auxilary iLO devices with the following PCI ID
836 if (dev
->subsystem_vendor
== PCI_VENDOR_ID_HP
&&
837 dev
->subsystem_device
== 0x1979)
840 if (pci_enable_device(dev
)) {
842 "Not possible to enable PCI Device: 0x%x:0x%x.\n",
843 ent
->vendor
, ent
->device
);
847 pci_mem_addr
= pci_iomap(dev
, 1, 0x80);
850 "Unable to detect the iLO2+ server memory.\n");
852 goto error_pci_iomap
;
854 hpwdt_nmistat
= pci_mem_addr
+ 0x6e;
855 hpwdt_timer_reg
= pci_mem_addr
+ 0x70;
856 hpwdt_timer_con
= pci_mem_addr
+ 0x72;
858 /* Make sure that timer is disabled until /dev/watchdog is opened */
861 /* Make sure that we have a valid soft_margin */
862 if (hpwdt_change_timer(soft_margin
))
863 hpwdt_change_timer(DEFAULT_MARGIN
);
865 /* Initialize NMI Decoding functionality */
866 retval
= hpwdt_init_nmi_decoding(dev
);
868 goto error_init_nmi_decoding
;
870 retval
= misc_register(&hpwdt_miscdev
);
873 "Unable to register miscdev on minor=%d (err=%d).\n",
874 WATCHDOG_MINOR
, retval
);
875 goto error_misc_register
;
878 dev_info(&dev
->dev
, "HPE Watchdog Timer Driver: %s"
879 ", timer margin: %d seconds (nowayout=%d).\n",
880 HPWDT_VERSION
, soft_margin
, nowayout
);
884 hpwdt_exit_nmi_decoding();
885 error_init_nmi_decoding
:
886 pci_iounmap(dev
, pci_mem_addr
);
888 pci_disable_device(dev
);
892 static void hpwdt_exit(struct pci_dev
*dev
)
897 misc_deregister(&hpwdt_miscdev
);
898 hpwdt_exit_nmi_decoding();
899 pci_iounmap(dev
, pci_mem_addr
);
900 pci_disable_device(dev
);
903 static struct pci_driver hpwdt_driver
= {
905 .id_table
= hpwdt_devices
,
906 .probe
= hpwdt_init_one
,
907 .remove
= hpwdt_exit
,
910 MODULE_AUTHOR("Tom Mingarelli");
911 MODULE_DESCRIPTION("hp watchdog driver");
912 MODULE_LICENSE("GPL");
913 MODULE_VERSION(HPWDT_VERSION
);
915 module_param(soft_margin
, int, 0);
916 MODULE_PARM_DESC(soft_margin
, "Watchdog timeout in seconds");
918 module_param(nowayout
, bool, 0);
919 MODULE_PARM_DESC(nowayout
, "Watchdog cannot be stopped once started (default="
920 __MODULE_STRING(WATCHDOG_NOWAYOUT
) ")");
922 #ifdef CONFIG_HPWDT_NMI_DECODING
923 module_param(allow_kdump
, int, 0);
924 MODULE_PARM_DESC(allow_kdump
, "Start a kernel dump after NMI occurs");
925 #endif /* !CONFIG_HPWDT_NMI_DECODING */
927 module_pci_driver(hpwdt_driver
);