5 * SoftDog 0.05: A Software Watchdog Device
7 * (c) Copyright 2015 Hewlett Packard Enterprise Development LP
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_timer_reg;
56 static unsigned long __iomem *hpwdt_timer_con;
58 static const struct pci_device_id hpwdt_devices[] = {
59 { PCI_DEVICE(PCI_VENDOR_ID_COMPAQ, 0xB203) }, /* iLO2 */
60 { PCI_DEVICE(PCI_VENDOR_ID_HP, 0x3306) }, /* iLO3 */
61 {0}, /* terminate list */
63 MODULE_DEVICE_TABLE(pci, hpwdt_devices);
65 #ifdef CONFIG_HPWDT_NMI_DECODING
66 #define PCI_BIOS32_SD_VALUE 0x5F32335F /* "_32_" */
67 #define CRU_BIOS_SIGNATURE_VALUE 0x55524324
68 #define PCI_BIOS32_PARAGRAPH_LEN 16
69 #define PCI_ROM_BASE1 0x000F0000
70 #define ROM_SIZE 0x10000
72 struct bios32_service_dir {
82 struct smbios_cru64_info {
91 #define SMBIOS_CRU64_INFORMATION 212
94 struct smbios_proliant_info {
103 #define SMBIOS_ICRU_INFORMATION 219
106 struct cmn_registers {
146 } __attribute__((packed));
148 static unsigned int hpwdt_nmi_decoding;
149 static unsigned int allow_kdump = 1;
150 static unsigned int is_icru;
151 static unsigned int is_uefi;
152 static DEFINE_SPINLOCK(rom_lock);
153 static void *cru_rom_addr;
154 static struct cmn_registers cmn_regs;
156 extern asmlinkage void asminline_call(struct cmn_registers *pi86Regs,
157 unsigned long *pRomEntry);
160 /* --32 Bit Bios------------------------------------------------------------ */
162 #define HPWDT_ARCH 32
166 ".globl asminline_call \n"
167 "asminline_call: \n\t"
169 "movl %esp, %ebp \n\t"
175 "movl 8(%ebp),%eax \n\t"
176 "movl 4(%eax),%ebx \n\t"
177 "movl 8(%eax),%ecx \n\t"
178 "movl 12(%eax),%edx \n\t"
179 "movl 16(%eax),%esi \n\t"
180 "movl 20(%eax),%edi \n\t"
181 "movl (%eax),%eax \n\t"
183 "call *12(%ebp) \n\t"
186 "movl 8(%ebp),%eax \n\t"
187 "movl %ebx,4(%eax) \n\t"
188 "movl %ecx,8(%eax) \n\t"
189 "movl %edx,12(%eax) \n\t"
190 "movl %esi,16(%eax) \n\t"
191 "movl %edi,20(%eax) \n\t"
192 "movw %ds,24(%eax) \n\t"
193 "movw %es,26(%eax) \n\t"
195 "movl %ebx,(%eax) \n\t"
197 "movl %ebx,28(%eax) \n\t"
209 * Routine Description:
210 * This function uses the 32-bit BIOS Service Directory record to
211 * search for a $CRU record.
217 static int cru_detect(unsigned long map_entry,
218 unsigned long map_offset)
221 unsigned long *bios32_entrypoint;
222 unsigned long cru_physical_address;
223 unsigned long cru_length;
224 unsigned long physical_bios_base = 0;
225 unsigned long physical_bios_offset = 0;
226 int retval = -ENODEV;
228 bios32_map = ioremap(map_entry, (2 * PAGE_SIZE));
230 if (bios32_map == NULL)
233 bios32_entrypoint = bios32_map + map_offset;
235 cmn_regs.u1.reax = CRU_BIOS_SIGNATURE_VALUE;
237 set_memory_x((unsigned long)bios32_map, 2);
238 asminline_call(&cmn_regs, bios32_entrypoint);
240 if (cmn_regs.u1.ral != 0) {
241 pr_warn("Call succeeded but with an error: 0x%x\n",
244 physical_bios_base = cmn_regs.u2.rebx;
245 physical_bios_offset = cmn_regs.u4.redx;
246 cru_length = cmn_regs.u3.recx;
247 cru_physical_address =
248 physical_bios_base + physical_bios_offset;
250 /* If the values look OK, then map it in. */
251 if ((physical_bios_base + physical_bios_offset)) {
253 ioremap(cru_physical_address, cru_length);
255 set_memory_x((unsigned long)cru_rom_addr & PAGE_MASK,
256 (cru_length + PAGE_SIZE - 1) >> PAGE_SHIFT);
261 pr_debug("CRU Base Address: 0x%lx\n", physical_bios_base);
262 pr_debug("CRU Offset Address: 0x%lx\n", physical_bios_offset);
263 pr_debug("CRU Length: 0x%lx\n", cru_length);
264 pr_debug("CRU Mapped Address: %p\n", &cru_rom_addr);
273 static int bios_checksum(const char __iomem *ptr, int len)
279 * calculate checksum of size bytes. This should add up
280 * to zero if we have a valid header.
282 for (i = 0; i < len; i++)
285 return ((sum == 0) && (len > 0));
291 * Routine Description:
292 * This function finds the 32-bit BIOS Service Directory
298 static int bios32_present(const char __iomem *p)
300 struct bios32_service_dir *bios_32_ptr;
302 unsigned long map_entry, map_offset;
304 bios_32_ptr = (struct bios32_service_dir *) p;
307 * Search for signature by checking equal to the swizzled value
308 * instead of calling another routine to perform a strcmp.
310 if (bios_32_ptr->signature == PCI_BIOS32_SD_VALUE) {
311 length = bios_32_ptr->length * PCI_BIOS32_PARAGRAPH_LEN;
312 if (bios_checksum(p, length)) {
314 * According to the spec, we're looking for the
315 * first 4KB-aligned address below the entrypoint
316 * listed in the header. The Service Directory code
317 * is guaranteed to occupy no more than 2 4KB pages.
319 map_entry = bios_32_ptr->entry_point & ~(PAGE_SIZE - 1);
320 map_offset = bios_32_ptr->entry_point - map_entry;
322 return cru_detect(map_entry, map_offset);
328 static int detect_cru_service(void)
334 * Search from 0x0f0000 through 0x0fffff, inclusive.
336 p = ioremap(PCI_ROM_BASE1, ROM_SIZE);
340 for (q = p; q < p + ROM_SIZE; q += 16) {
341 rc = bios32_present(q);
348 /* ------------------------------------------------------------------------- */
349 #endif /* CONFIG_X86_32 */
351 /* --64 Bit Bios------------------------------------------------------------ */
353 #define HPWDT_ARCH 64
357 ".globl asminline_call \n\t"
358 ".type asminline_call, @function \n\t"
359 "asminline_call: \n\t"
366 "movq %rsi, %r12 \n\t"
367 "movq %rdi, %r9 \n\t"
368 "movl 4(%r9),%ebx \n\t"
369 "movl 8(%r9),%ecx \n\t"
370 "movl 12(%r9),%edx \n\t"
371 "movl 16(%r9),%esi \n\t"
372 "movl 20(%r9),%edi \n\t"
373 "movl (%r9),%eax \n\t"
377 "movl %eax, (%r9) \n\t"
378 "movl %ebx, 4(%r9) \n\t"
379 "movl %ecx, 8(%r9) \n\t"
380 "movl %edx, 12(%r9) \n\t"
381 "movl %esi, 16(%r9) \n\t"
382 "movl %edi, 20(%r9) \n\t"
383 "movq %r12, %rax \n\t"
384 "movl %eax, 28(%r9) \n\t"
397 * Routine Description:
398 * This function checks whether or not a SMBIOS/DMI record is
399 * the 64bit CRU info or not
401 static void dmi_find_cru(const struct dmi_header *dm, void *dummy)
403 struct smbios_cru64_info *smbios_cru64_ptr;
404 unsigned long cru_physical_address;
406 if (dm->type == SMBIOS_CRU64_INFORMATION) {
407 smbios_cru64_ptr = (struct smbios_cru64_info *) dm;
408 if (smbios_cru64_ptr->signature == CRU_BIOS_SIGNATURE_VALUE) {
409 cru_physical_address =
410 smbios_cru64_ptr->physical_address +
411 smbios_cru64_ptr->double_offset;
412 cru_rom_addr = ioremap(cru_physical_address,
413 smbios_cru64_ptr->double_length);
414 set_memory_x((unsigned long)cru_rom_addr & PAGE_MASK,
415 smbios_cru64_ptr->double_length >> PAGE_SHIFT);
420 static int detect_cru_service(void)
424 dmi_walk(dmi_find_cru, NULL);
426 /* if cru_rom_addr has been set then we found a CRU service */
427 return ((cru_rom_addr != NULL) ? 0 : -ENODEV);
429 /* ------------------------------------------------------------------------- */
430 #endif /* CONFIG_X86_64 */
431 #endif /* CONFIG_HPWDT_NMI_DECODING */
434 * Watchdog operations
436 static void hpwdt_start(void)
438 reload = SECS_TO_TICKS(soft_margin);
439 iowrite16(reload, hpwdt_timer_reg);
440 iowrite8(0x85, hpwdt_timer_con);
443 static void hpwdt_stop(void)
447 data = ioread8(hpwdt_timer_con);
449 iowrite8(data, hpwdt_timer_con);
452 static void hpwdt_ping(void)
454 iowrite16(reload, hpwdt_timer_reg);
457 static int hpwdt_change_timer(int new_margin)
459 if (new_margin < 1 || new_margin > HPWDT_MAX_TIMER) {
460 pr_warn("New value passed in is invalid: %d seconds\n",
465 soft_margin = new_margin;
466 pr_debug("New timer passed in is %d seconds\n", new_margin);
467 reload = SECS_TO_TICKS(soft_margin);
472 static int hpwdt_time_left(void)
474 return TICKS_TO_SECS(ioread16(hpwdt_timer_reg));
477 #ifdef CONFIG_HPWDT_NMI_DECODING
481 static int hpwdt_pretimeout(unsigned int ulReason, struct pt_regs *regs)
483 unsigned long rom_pl;
484 static int die_nmi_called;
486 if (!hpwdt_nmi_decoding)
489 spin_lock_irqsave(&rom_lock, rom_pl);
490 if (!die_nmi_called && !is_icru && !is_uefi)
491 asminline_call(&cmn_regs, cru_rom_addr);
493 spin_unlock_irqrestore(&rom_lock, rom_pl);
498 if (!is_icru && !is_uefi) {
499 if (cmn_regs.u1.ral == 0) {
500 nmi_panic(regs, "An NMI occurred, but unable to determine source.\n");
504 nmi_panic(regs, "An NMI occurred. Depending on your system the reason "
505 "for the NMI is logged in any one of the following "
507 "1. Integrated Management Log (IML)\n"
509 "3. OA Forward Progress Log\n"
514 #endif /* CONFIG_HPWDT_NMI_DECODING */
517 * /dev/watchdog handling
519 static int hpwdt_open(struct inode *inode, struct file *file)
521 /* /dev/watchdog can only be opened once */
522 if (test_and_set_bit(0, &hpwdt_is_open))
525 /* Start the watchdog */
529 return nonseekable_open(inode, file);
532 static int hpwdt_release(struct inode *inode, struct file *file)
534 /* Stop the watchdog */
535 if (expect_release == 42) {
538 pr_crit("Unexpected close, not stopping watchdog!\n");
544 /* /dev/watchdog is being closed, make sure it can be re-opened */
545 clear_bit(0, &hpwdt_is_open);
550 static ssize_t hpwdt_write(struct file *file, const char __user *data,
551 size_t len, loff_t *ppos)
553 /* See if we got the magic character 'V' and reload the timer */
558 /* note: just in case someone wrote the magic character
559 * five months ago... */
562 /* scan to see whether or not we got the magic char. */
563 for (i = 0; i != len; i++) {
565 if (get_user(c, data + i))
572 /* someone wrote to us, we should reload the timer */
579 static const struct watchdog_info ident = {
580 .options = WDIOF_SETTIMEOUT |
581 WDIOF_KEEPALIVEPING |
583 .identity = "HPE iLO2+ HW Watchdog Timer",
586 static long hpwdt_ioctl(struct file *file, unsigned int cmd,
589 void __user *argp = (void __user *)arg;
590 int __user *p = argp;
591 int new_margin, options;
595 case WDIOC_GETSUPPORT:
597 if (copy_to_user(argp, &ident, sizeof(ident)))
601 case WDIOC_GETSTATUS:
602 case WDIOC_GETBOOTSTATUS:
603 ret = put_user(0, p);
606 case WDIOC_KEEPALIVE:
611 case WDIOC_SETOPTIONS:
612 ret = get_user(options, p);
616 if (options & WDIOS_DISABLECARD)
619 if (options & WDIOS_ENABLECARD) {
625 case WDIOC_SETTIMEOUT:
626 ret = get_user(new_margin, p);
630 ret = hpwdt_change_timer(new_margin);
636 case WDIOC_GETTIMEOUT:
637 ret = put_user(soft_margin, p);
640 case WDIOC_GETTIMELEFT:
641 ret = put_user(hpwdt_time_left(), p);
650 static const struct file_operations hpwdt_fops = {
651 .owner = THIS_MODULE,
653 .write = hpwdt_write,
654 .unlocked_ioctl = hpwdt_ioctl,
656 .release = hpwdt_release,
659 static struct miscdevice hpwdt_miscdev = {
660 .minor = WATCHDOG_MINOR,
669 #ifdef CONFIG_HPWDT_NMI_DECODING
670 #ifdef CONFIG_X86_LOCAL_APIC
671 static void hpwdt_check_nmi_decoding(struct pci_dev *dev)
674 * If nmi_watchdog is turned off then we can turn on
675 * our nmi decoding capability.
677 hpwdt_nmi_decoding = 1;
680 static void hpwdt_check_nmi_decoding(struct pci_dev *dev)
682 dev_warn(&dev->dev, "NMI decoding is disabled. "
683 "Your kernel does not support a NMI Watchdog.\n");
685 #endif /* CONFIG_X86_LOCAL_APIC */
690 * Routine Description:
691 * This function checks whether or not we are on an iCRU-based server.
692 * This check is independent of architecture and needs to be made for
693 * any ProLiant system.
695 static void dmi_find_icru(const struct dmi_header *dm, void *dummy)
697 struct smbios_proliant_info *smbios_proliant_ptr;
699 if (dm->type == SMBIOS_ICRU_INFORMATION) {
700 smbios_proliant_ptr = (struct smbios_proliant_info *) dm;
701 if (smbios_proliant_ptr->misc_features & 0x01)
703 if (smbios_proliant_ptr->misc_features & 0x408)
708 static int hpwdt_init_nmi_decoding(struct pci_dev *dev)
713 * On typical CRU-based systems we need to map that service in
714 * the BIOS. For 32 bit Operating Systems we need to go through
715 * the 32 Bit BIOS Service Directory. For 64 bit Operating
716 * Systems we get that service through SMBIOS.
718 * On systems that support the new iCRU service all we need to
719 * do is call dmi_walk to get the supported flag value and skip
720 * the old cru detect code.
722 dmi_walk(dmi_find_icru, NULL);
723 if (!is_icru && !is_uefi) {
726 * We need to map the ROM to get the CRU service.
727 * For 32 bit Operating Systems we need to go through the 32 Bit
728 * BIOS Service Directory
729 * For 64 bit Operating Systems we get that service through SMBIOS.
731 retval = detect_cru_service();
734 "Unable to detect the %d Bit CRU Service.\n",
740 * We know this is the only CRU call we need to make so lets keep as
741 * few instructions as possible once the NMI comes in.
743 cmn_regs.u1.rah = 0x0D;
744 cmn_regs.u1.ral = 0x02;
748 * Only one function can register for NMI_UNKNOWN
750 retval = register_nmi_handler(NMI_UNKNOWN, hpwdt_pretimeout, 0, "hpwdt");
753 retval = register_nmi_handler(NMI_SERR, hpwdt_pretimeout, 0, "hpwdt");
756 retval = register_nmi_handler(NMI_IO_CHECK, hpwdt_pretimeout, 0, "hpwdt");
761 "HPE Watchdog Timer Driver: NMI decoding initialized"
762 ", allow kernel dump: %s (default = 1/ON)\n",
763 (allow_kdump == 0) ? "OFF" : "ON");
767 unregister_nmi_handler(NMI_SERR, "hpwdt");
769 unregister_nmi_handler(NMI_UNKNOWN, "hpwdt");
772 "Unable to register a die notifier (err=%d).\n",
775 iounmap(cru_rom_addr);
779 static void hpwdt_exit_nmi_decoding(void)
781 unregister_nmi_handler(NMI_UNKNOWN, "hpwdt");
782 unregister_nmi_handler(NMI_SERR, "hpwdt");
783 unregister_nmi_handler(NMI_IO_CHECK, "hpwdt");
785 iounmap(cru_rom_addr);
787 #else /* !CONFIG_HPWDT_NMI_DECODING */
788 static void hpwdt_check_nmi_decoding(struct pci_dev *dev)
792 static int hpwdt_init_nmi_decoding(struct pci_dev *dev)
797 static void hpwdt_exit_nmi_decoding(void)
800 #endif /* CONFIG_HPWDT_NMI_DECODING */
802 static int hpwdt_init_one(struct pci_dev *dev,
803 const struct pci_device_id *ent)
808 * Check if we can do NMI decoding or not
810 hpwdt_check_nmi_decoding(dev);
813 * First let's find out if we are on an iLO2+ server. We will
814 * not run on a legacy ASM box.
815 * So we only support the G5 ProLiant servers and higher.
817 if (dev->subsystem_vendor != PCI_VENDOR_ID_HP &&
818 dev->subsystem_vendor != PCI_VENDOR_ID_HP_3PAR) {
820 "This server does not have an iLO2+ ASIC.\n");
825 * Ignore all auxilary iLO devices with the following PCI ID
827 if (dev->subsystem_vendor == PCI_VENDOR_ID_HP &&
828 dev->subsystem_device == 0x1979)
831 if (pci_enable_device(dev)) {
833 "Not possible to enable PCI Device: 0x%x:0x%x.\n",
834 ent->vendor, ent->device);
838 pci_mem_addr = pci_iomap(dev, 1, 0x80);
841 "Unable to detect the iLO2+ server memory.\n");
843 goto error_pci_iomap;
845 hpwdt_timer_reg = pci_mem_addr + 0x70;
846 hpwdt_timer_con = pci_mem_addr + 0x72;
848 /* Make sure that timer is disabled until /dev/watchdog is opened */
851 /* Make sure that we have a valid soft_margin */
852 if (hpwdt_change_timer(soft_margin))
853 hpwdt_change_timer(DEFAULT_MARGIN);
855 /* Initialize NMI Decoding functionality */
856 retval = hpwdt_init_nmi_decoding(dev);
858 goto error_init_nmi_decoding;
860 retval = misc_register(&hpwdt_miscdev);
863 "Unable to register miscdev on minor=%d (err=%d).\n",
864 WATCHDOG_MINOR, retval);
865 goto error_misc_register;
868 dev_info(&dev->dev, "HPE Watchdog Timer Driver: %s"
869 ", timer margin: %d seconds (nowayout=%d).\n",
870 HPWDT_VERSION, soft_margin, nowayout);
874 hpwdt_exit_nmi_decoding();
875 error_init_nmi_decoding:
876 pci_iounmap(dev, pci_mem_addr);
878 pci_disable_device(dev);
882 static void hpwdt_exit(struct pci_dev *dev)
887 misc_deregister(&hpwdt_miscdev);
888 hpwdt_exit_nmi_decoding();
889 pci_iounmap(dev, pci_mem_addr);
890 pci_disable_device(dev);
893 static struct pci_driver hpwdt_driver = {
895 .id_table = hpwdt_devices,
896 .probe = hpwdt_init_one,
897 .remove = hpwdt_exit,
900 MODULE_AUTHOR("Tom Mingarelli");
901 MODULE_DESCRIPTION("hp watchdog driver");
902 MODULE_LICENSE("GPL");
903 MODULE_VERSION(HPWDT_VERSION);
905 module_param(soft_margin, int, 0);
906 MODULE_PARM_DESC(soft_margin, "Watchdog timeout in seconds");
908 module_param(nowayout, bool, 0);
909 MODULE_PARM_DESC(nowayout, "Watchdog cannot be stopped once started (default="
910 __MODULE_STRING(WATCHDOG_NOWAYOUT) ")");
912 #ifdef CONFIG_HPWDT_NMI_DECODING
913 module_param(allow_kdump, int, 0);
914 MODULE_PARM_DESC(allow_kdump, "Start a kernel dump after NMI occurs");
915 #endif /* !CONFIG_HPWDT_NMI_DECODING */
917 module_pci_driver(hpwdt_driver);