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>
20 #include <linux/bitops.h>
21 #include <linux/kernel.h>
22 #include <linux/miscdevice.h>
23 #include <linux/module.h>
24 #include <linux/moduleparam.h>
25 #include <linux/pci.h>
26 #include <linux/pci_ids.h>
27 #include <linux/types.h>
28 #include <linux/uaccess.h>
29 #include <linux/watchdog.h>
30 #include <linux/dmi.h>
31 #include <linux/spinlock.h>
32 #include <linux/nmi.h>
33 #include <linux/kdebug.h>
34 #include <linux/notifier.h>
35 #include <asm/cacheflush.h>
37 #define HPWDT_VERSION "1.1.1"
38 #define SECS_TO_TICKS(secs) ((secs) * 1000 / 128)
39 #define DEFAULT_MARGIN 30
41 static unsigned int soft_margin = DEFAULT_MARGIN; /* in seconds */
42 static unsigned int reload; /* the computed soft_margin */
43 static int nowayout = WATCHDOG_NOWAYOUT;
44 static char expect_release;
45 static unsigned long hpwdt_is_open;
47 static void __iomem *pci_mem_addr; /* the PCI-memory address */
48 static unsigned long __iomem *hpwdt_timer_reg;
49 static unsigned long __iomem *hpwdt_timer_con;
51 static struct pci_device_id hpwdt_devices[] = {
52 { PCI_DEVICE(PCI_VENDOR_ID_COMPAQ, 0xB203) }, /* iLO2 */
53 { PCI_DEVICE(PCI_VENDOR_ID_HP, 0x3306) }, /* iLO3 */
54 {0}, /* terminate list */
56 MODULE_DEVICE_TABLE(pci, hpwdt_devices);
58 #define PCI_BIOS32_SD_VALUE 0x5F32335F /* "_32_" */
59 #define CRU_BIOS_SIGNATURE_VALUE 0x55524324
60 #define PCI_BIOS32_PARAGRAPH_LEN 16
61 #define PCI_ROM_BASE1 0x000F0000
62 #define ROM_SIZE 0x10000
64 struct bios32_service_dir {
74 struct smbios_cru64_info {
83 #define SMBIOS_CRU64_INFORMATION 212
85 struct cmn_registers {
125 } __attribute__((packed));
127 static unsigned int hpwdt_nmi_sourcing;
128 static unsigned int allow_kdump;
129 static unsigned int priority; /* hpwdt at end of die_notify list */
130 static DEFINE_SPINLOCK(rom_lock);
131 static void *cru_rom_addr;
132 static struct cmn_registers cmn_regs;
134 extern asmlinkage void asminline_call(struct cmn_registers *pi86Regs,
135 unsigned long *pRomEntry);
138 /* --32 Bit Bios------------------------------------------------------------ */
140 #define HPWDT_ARCH 32
144 "asminline_call: \n\t"
146 "movl %esp, %ebp \n\t"
152 "movl 8(%ebp),%eax \n\t"
153 "movl 4(%eax),%ebx \n\t"
154 "movl 8(%eax),%ecx \n\t"
155 "movl 12(%eax),%edx \n\t"
156 "movl 16(%eax),%esi \n\t"
157 "movl 20(%eax),%edi \n\t"
158 "movl (%eax),%eax \n\t"
160 "call *12(%ebp) \n\t"
163 "movl 8(%ebp),%eax \n\t"
164 "movl %ebx,4(%eax) \n\t"
165 "movl %ecx,8(%eax) \n\t"
166 "movl %edx,12(%eax) \n\t"
167 "movl %esi,16(%eax) \n\t"
168 "movl %edi,20(%eax) \n\t"
169 "movw %ds,24(%eax) \n\t"
170 "movw %es,26(%eax) \n\t"
172 "movl %ebx,(%eax) \n\t"
174 "movl %ebx,28(%eax) \n\t"
186 * Routine Description:
187 * This function uses the 32-bit BIOS Service Directory record to
188 * search for a $CRU record.
194 static int __devinit cru_detect(unsigned long map_entry,
195 unsigned long map_offset)
198 unsigned long *bios32_entrypoint;
199 unsigned long cru_physical_address;
200 unsigned long cru_length;
201 unsigned long physical_bios_base = 0;
202 unsigned long physical_bios_offset = 0;
203 int retval = -ENODEV;
205 bios32_map = ioremap(map_entry, (2 * PAGE_SIZE));
207 if (bios32_map == NULL)
210 bios32_entrypoint = bios32_map + map_offset;
212 cmn_regs.u1.reax = CRU_BIOS_SIGNATURE_VALUE;
214 asminline_call(&cmn_regs, bios32_entrypoint);
216 if (cmn_regs.u1.ral != 0) {
218 "hpwdt: Call succeeded but with an error: 0x%x\n",
221 physical_bios_base = cmn_regs.u2.rebx;
222 physical_bios_offset = cmn_regs.u4.redx;
223 cru_length = cmn_regs.u3.recx;
224 cru_physical_address =
225 physical_bios_base + physical_bios_offset;
227 /* If the values look OK, then map it in. */
228 if ((physical_bios_base + physical_bios_offset)) {
230 ioremap(cru_physical_address, cru_length);
235 printk(KERN_DEBUG "hpwdt: CRU Base Address: 0x%lx\n",
237 printk(KERN_DEBUG "hpwdt: CRU Offset Address: 0x%lx\n",
238 physical_bios_offset);
239 printk(KERN_DEBUG "hpwdt: CRU Length: 0x%lx\n",
241 printk(KERN_DEBUG "hpwdt: CRU Mapped Address: %p\n",
251 static int __devinit bios_checksum(const char __iomem *ptr, int len)
257 * calculate checksum of size bytes. This should add up
258 * to zero if we have a valid header.
260 for (i = 0; i < len; i++)
263 return ((sum == 0) && (len > 0));
269 * Routine Description:
270 * This function finds the 32-bit BIOS Service Directory
276 static int __devinit bios32_present(const char __iomem *p)
278 struct bios32_service_dir *bios_32_ptr;
280 unsigned long map_entry, map_offset;
282 bios_32_ptr = (struct bios32_service_dir *) p;
285 * Search for signature by checking equal to the swizzled value
286 * instead of calling another routine to perform a strcmp.
288 if (bios_32_ptr->signature == PCI_BIOS32_SD_VALUE) {
289 length = bios_32_ptr->length * PCI_BIOS32_PARAGRAPH_LEN;
290 if (bios_checksum(p, length)) {
292 * According to the spec, we're looking for the
293 * first 4KB-aligned address below the entrypoint
294 * listed in the header. The Service Directory code
295 * is guaranteed to occupy no more than 2 4KB pages.
297 map_entry = bios_32_ptr->entry_point & ~(PAGE_SIZE - 1);
298 map_offset = bios_32_ptr->entry_point - map_entry;
300 return cru_detect(map_entry, map_offset);
306 static int __devinit detect_cru_service(void)
312 * Search from 0x0f0000 through 0x0fffff, inclusive.
314 p = ioremap(PCI_ROM_BASE1, ROM_SIZE);
318 for (q = p; q < p + ROM_SIZE; q += 16) {
319 rc = bios32_present(q);
326 /* ------------------------------------------------------------------------- */
327 #endif /* CONFIG_X86_32 */
329 /* --64 Bit Bios------------------------------------------------------------ */
331 #define HPWDT_ARCH 64
335 "asminline_call: \n\t"
337 "movq %rsp, %rbp \n\t"
343 "movq %rsi, %r12 \n\t"
344 "movq %rdi, %r9 \n\t"
345 "movl 4(%r9),%ebx \n\t"
346 "movl 8(%r9),%ecx \n\t"
347 "movl 12(%r9),%edx \n\t"
348 "movl 16(%r9),%esi \n\t"
349 "movl 20(%r9),%edi \n\t"
350 "movl (%r9),%eax \n\t"
354 "movl %eax, (%r9) \n\t"
355 "movl %ebx, 4(%r9) \n\t"
356 "movl %ecx, 8(%r9) \n\t"
357 "movl %edx, 12(%r9) \n\t"
358 "movl %esi, 16(%r9) \n\t"
359 "movl %edi, 20(%r9) \n\t"
360 "movq %r12, %rax \n\t"
361 "movl %eax, 28(%r9) \n\t"
374 * Routine Description:
375 * This function checks whether or not a SMBIOS/DMI record is
376 * the 64bit CRU info or not
378 static void __devinit dmi_find_cru(const struct dmi_header *dm, void *dummy)
380 struct smbios_cru64_info *smbios_cru64_ptr;
381 unsigned long cru_physical_address;
383 if (dm->type == SMBIOS_CRU64_INFORMATION) {
384 smbios_cru64_ptr = (struct smbios_cru64_info *) dm;
385 if (smbios_cru64_ptr->signature == CRU_BIOS_SIGNATURE_VALUE) {
386 cru_physical_address =
387 smbios_cru64_ptr->physical_address +
388 smbios_cru64_ptr->double_offset;
389 cru_rom_addr = ioremap(cru_physical_address,
390 smbios_cru64_ptr->double_length);
391 set_memory_x((unsigned long)cru_rom_addr & PAGE_MASK,
392 smbios_cru64_ptr->double_length >> PAGE_SHIFT);
397 static int __devinit detect_cru_service(void)
401 dmi_walk(dmi_find_cru, NULL);
403 /* if cru_rom_addr has been set then we found a CRU service */
404 return ((cru_rom_addr != NULL) ? 0 : -ENODEV);
406 /* ------------------------------------------------------------------------- */
407 #endif /* CONFIG_X86_64 */
410 * Watchdog operations
412 static void hpwdt_start(void)
414 reload = SECS_TO_TICKS(soft_margin);
415 iowrite16(reload, hpwdt_timer_reg);
416 iowrite16(0x85, hpwdt_timer_con);
419 static void hpwdt_stop(void)
423 data = ioread16(hpwdt_timer_con);
425 iowrite16(data, hpwdt_timer_con);
428 static void hpwdt_ping(void)
430 iowrite16(reload, hpwdt_timer_reg);
433 static int hpwdt_change_timer(int new_margin)
435 /* Arbitrary, can't find the card's limits */
436 if (new_margin < 5 || new_margin > 600) {
438 "hpwdt: New value passed in is invalid: %d seconds.\n",
443 soft_margin = new_margin;
445 "hpwdt: New timer passed in is %d seconds.\n",
447 reload = SECS_TO_TICKS(soft_margin);
455 static int hpwdt_pretimeout(struct notifier_block *nb, unsigned long ulReason,
458 unsigned long rom_pl;
459 static int die_nmi_called;
461 if (ulReason != DIE_NMI && ulReason != DIE_NMI_IPI)
464 if (hpwdt_nmi_sourcing) {
465 spin_lock_irqsave(&rom_lock, rom_pl);
467 asminline_call(&cmn_regs, cru_rom_addr);
469 spin_unlock_irqrestore(&rom_lock, rom_pl);
470 if (cmn_regs.u1.ral == 0) {
471 printk(KERN_WARNING "hpwdt: An NMI occurred, "
472 "but unable to determine source.\n");
476 panic("An NMI occurred, please see the Integrated "
477 "Management Log for details.\n");
484 * /dev/watchdog handling
486 static int hpwdt_open(struct inode *inode, struct file *file)
488 /* /dev/watchdog can only be opened once */
489 if (test_and_set_bit(0, &hpwdt_is_open))
492 /* Start the watchdog */
496 return nonseekable_open(inode, file);
499 static int hpwdt_release(struct inode *inode, struct file *file)
501 /* Stop the watchdog */
502 if (expect_release == 42) {
506 "hpwdt: Unexpected close, not stopping watchdog!\n");
512 /* /dev/watchdog is being closed, make sure it can be re-opened */
513 clear_bit(0, &hpwdt_is_open);
518 static ssize_t hpwdt_write(struct file *file, const char __user *data,
519 size_t len, loff_t *ppos)
521 /* See if we got the magic character 'V' and reload the timer */
526 /* note: just in case someone wrote the magic character
527 * five months ago... */
530 /* scan to see whether or not we got the magic char. */
531 for (i = 0; i != len; i++) {
533 if (get_user(c, data + i))
540 /* someone wrote to us, we should reload the timer */
547 static const struct watchdog_info ident = {
548 .options = WDIOF_SETTIMEOUT |
549 WDIOF_KEEPALIVEPING |
551 .identity = "HP iLO2+ HW Watchdog Timer",
554 static long hpwdt_ioctl(struct file *file, unsigned int cmd,
557 void __user *argp = (void __user *)arg;
558 int __user *p = argp;
563 case WDIOC_GETSUPPORT:
565 if (copy_to_user(argp, &ident, sizeof(ident)))
569 case WDIOC_GETSTATUS:
570 case WDIOC_GETBOOTSTATUS:
571 ret = put_user(0, p);
574 case WDIOC_KEEPALIVE:
579 case WDIOC_SETTIMEOUT:
580 ret = get_user(new_margin, p);
584 ret = hpwdt_change_timer(new_margin);
590 case WDIOC_GETTIMEOUT:
591 ret = put_user(soft_margin, p);
600 static const struct file_operations hpwdt_fops = {
601 .owner = THIS_MODULE,
603 .write = hpwdt_write,
604 .unlocked_ioctl = hpwdt_ioctl,
606 .release = hpwdt_release,
609 static struct miscdevice hpwdt_miscdev = {
610 .minor = WATCHDOG_MINOR,
615 static struct notifier_block die_notifier = {
616 .notifier_call = hpwdt_pretimeout,
624 #ifdef ARCH_HAS_NMI_WATCHDOG
625 static void __devinit hpwdt_check_nmi_sourcing(struct pci_dev *dev)
628 * If nmi_watchdog is turned off then we can turn on
629 * our nmi sourcing capability.
631 if (!nmi_watchdog_active())
632 hpwdt_nmi_sourcing = 1;
634 dev_warn(&dev->dev, "NMI sourcing is disabled. To enable this "
635 "functionality you must reboot with nmi_watchdog=0 "
636 "and load the hpwdt driver with priority=1.\n");
639 static void __devinit hpwdt_check_nmi_sourcing(struct pci_dev *dev)
641 dev_warn(&dev->dev, "NMI sourcing is disabled. "
642 "Your kernel does not support a NMI Watchdog.\n");
646 static int __devinit hpwdt_init_one(struct pci_dev *dev,
647 const struct pci_device_id *ent)
652 * Check if we can do NMI sourcing or not
654 hpwdt_check_nmi_sourcing(dev);
657 * First let's find out if we are on an iLO2+ server. We will
658 * not run on a legacy ASM box.
659 * So we only support the G5 ProLiant servers and higher.
661 if (dev->subsystem_vendor != PCI_VENDOR_ID_HP) {
663 "This server does not have an iLO2+ ASIC.\n");
667 if (pci_enable_device(dev)) {
669 "Not possible to enable PCI Device: 0x%x:0x%x.\n",
670 ent->vendor, ent->device);
674 pci_mem_addr = pci_iomap(dev, 1, 0x80);
677 "Unable to detect the iLO2+ server memory.\n");
679 goto error_pci_iomap;
681 hpwdt_timer_reg = pci_mem_addr + 0x70;
682 hpwdt_timer_con = pci_mem_addr + 0x72;
684 /* Make sure that we have a valid soft_margin */
685 if (hpwdt_change_timer(soft_margin))
686 hpwdt_change_timer(DEFAULT_MARGIN);
689 * We need to map the ROM to get the CRU service.
690 * For 32 bit Operating Systems we need to go through the 32 Bit
691 * BIOS Service Directory
692 * For 64 bit Operating Systems we get that service through SMBIOS.
694 retval = detect_cru_service();
697 "Unable to detect the %d Bit CRU Service.\n",
703 * We know this is the only CRU call we need to make so lets keep as
704 * few instructions as possible once the NMI comes in.
706 cmn_regs.u1.rah = 0x0D;
707 cmn_regs.u1.ral = 0x02;
710 * If the priority is set to 1, then we will be put first on the
711 * die notify list to handle a critical NMI. The default is to
712 * be last so other users of the NMI signal can function.
715 die_notifier.priority = 0x7FFFFFFF;
717 retval = register_die_notifier(&die_notifier);
720 "Unable to register a die notifier (err=%d).\n",
722 goto error_die_notifier;
725 retval = misc_register(&hpwdt_miscdev);
728 "Unable to register miscdev on minor=%d (err=%d).\n",
729 WATCHDOG_MINOR, retval);
730 goto error_misc_register;
734 "hp Watchdog Timer Driver: %s"
735 ", timer margin: %d seconds (nowayout=%d)"
736 ", allow kernel dump: %s (default = 0/OFF)"
737 ", priority: %s (default = 0/LAST).\n",
738 HPWDT_VERSION, soft_margin, nowayout,
739 (allow_kdump == 0) ? "OFF" : "ON",
740 (priority == 0) ? "LAST" : "FIRST");
745 unregister_die_notifier(&die_notifier);
748 iounmap(cru_rom_addr);
750 pci_iounmap(dev, pci_mem_addr);
752 pci_disable_device(dev);
756 static void __devexit hpwdt_exit(struct pci_dev *dev)
761 misc_deregister(&hpwdt_miscdev);
762 unregister_die_notifier(&die_notifier);
765 iounmap(cru_rom_addr);
766 pci_iounmap(dev, pci_mem_addr);
767 pci_disable_device(dev);
770 static struct pci_driver hpwdt_driver = {
772 .id_table = hpwdt_devices,
773 .probe = hpwdt_init_one,
774 .remove = __devexit_p(hpwdt_exit),
777 static void __exit hpwdt_cleanup(void)
779 pci_unregister_driver(&hpwdt_driver);
782 static int __init hpwdt_init(void)
784 return pci_register_driver(&hpwdt_driver);
787 MODULE_AUTHOR("Tom Mingarelli");
788 MODULE_DESCRIPTION("hp watchdog driver");
789 MODULE_LICENSE("GPL");
790 MODULE_VERSION(HPWDT_VERSION);
791 MODULE_ALIAS_MISCDEV(WATCHDOG_MINOR);
793 module_param(soft_margin, int, 0);
794 MODULE_PARM_DESC(soft_margin, "Watchdog timeout in seconds");
796 module_param(nowayout, int, 0);
797 MODULE_PARM_DESC(nowayout, "Watchdog cannot be stopped once started (default="
798 __MODULE_STRING(WATCHDOG_NOWAYOUT) ")");
800 module_param(allow_kdump, int, 0);
801 MODULE_PARM_DESC(allow_kdump, "Start a kernel dump after NMI occurs");
803 module_param(priority, int, 0);
804 MODULE_PARM_DESC(priority, "The hpwdt driver handles NMIs first or last"
805 " (default = 0/Last)\n");
807 module_init(hpwdt_init);
808 module_exit(hpwdt_cleanup);