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 int hpwdt_nmi_sourcing;
125 static void __iomem *pci_mem_addr; /* the PCI-memory address */
126 static unsigned long __iomem *hpwdt_timer_reg;
127 static unsigned long __iomem *hpwdt_timer_con;
129 static DEFINE_SPINLOCK(rom_lock);
131 static void *cru_rom_addr;
133 static struct cmn_registers cmn_regs;
135 static struct pci_device_id hpwdt_devices[] = {
136 { PCI_DEVICE(PCI_VENDOR_ID_COMPAQ, 0xB203) },
137 { PCI_DEVICE(PCI_VENDOR_ID_HP, 0x3306) },
138 {0}, /* terminate list */
140 MODULE_DEVICE_TABLE(pci, hpwdt_devices);
142 extern asmlinkage void asminline_call(struct cmn_registers *pi86Regs,
143 unsigned long *pRomEntry);
145 #ifndef CONFIG_X86_64
146 /* --32 Bit Bios------------------------------------------------------------ */
148 #define HPWDT_ARCH 32
152 "asminline_call: \n\t"
154 "movl %esp, %ebp \n\t"
160 "movl 8(%ebp),%eax \n\t"
161 "movl 4(%eax),%ebx \n\t"
162 "movl 8(%eax),%ecx \n\t"
163 "movl 12(%eax),%edx \n\t"
164 "movl 16(%eax),%esi \n\t"
165 "movl 20(%eax),%edi \n\t"
166 "movl (%eax),%eax \n\t"
168 "call *12(%ebp) \n\t"
171 "movl 8(%ebp),%eax \n\t"
172 "movl %ebx,4(%eax) \n\t"
173 "movl %ecx,8(%eax) \n\t"
174 "movl %edx,12(%eax) \n\t"
175 "movl %esi,16(%eax) \n\t"
176 "movl %edi,20(%eax) \n\t"
177 "movw %ds,24(%eax) \n\t"
178 "movw %es,26(%eax) \n\t"
180 "movl %ebx,(%eax) \n\t"
182 "movl %ebx,28(%eax) \n\t"
194 * Routine Description:
195 * This function uses the 32-bit BIOS Service Directory record to
196 * search for a $CRU record.
202 static int __devinit cru_detect(unsigned long map_entry,
203 unsigned long map_offset)
206 unsigned long *bios32_entrypoint;
207 unsigned long cru_physical_address;
208 unsigned long cru_length;
209 unsigned long physical_bios_base = 0;
210 unsigned long physical_bios_offset = 0;
211 int retval = -ENODEV;
213 bios32_map = ioremap(map_entry, (2 * PAGE_SIZE));
215 if (bios32_map == NULL)
218 bios32_entrypoint = bios32_map + map_offset;
220 cmn_regs.u1.reax = CRU_BIOS_SIGNATURE_VALUE;
222 asminline_call(&cmn_regs, bios32_entrypoint);
224 if (cmn_regs.u1.ral != 0) {
226 "hpwdt: Call succeeded but with an error: 0x%x\n",
229 physical_bios_base = cmn_regs.u2.rebx;
230 physical_bios_offset = cmn_regs.u4.redx;
231 cru_length = cmn_regs.u3.recx;
232 cru_physical_address =
233 physical_bios_base + physical_bios_offset;
235 /* If the values look OK, then map it in. */
236 if ((physical_bios_base + physical_bios_offset)) {
238 ioremap(cru_physical_address, cru_length);
243 printk(KERN_DEBUG "hpwdt: CRU Base Address: 0x%lx\n",
245 printk(KERN_DEBUG "hpwdt: CRU Offset Address: 0x%lx\n",
246 physical_bios_offset);
247 printk(KERN_DEBUG "hpwdt: CRU Length: 0x%lx\n",
249 printk(KERN_DEBUG "hpwdt: CRU Mapped Address: 0x%x\n",
250 (unsigned int)&cru_rom_addr);
259 static int __devinit bios_checksum(const char __iomem *ptr, int len)
265 * calculate checksum of size bytes. This should add up
266 * to zero if we have a valid header.
268 for (i = 0; i < len; i++)
271 return ((sum == 0) && (len > 0));
277 * Routine Description:
278 * This function finds the 32-bit BIOS Service Directory
284 static int __devinit bios32_present(const char __iomem *p)
286 struct bios32_service_dir *bios_32_ptr;
288 unsigned long map_entry, map_offset;
290 bios_32_ptr = (struct bios32_service_dir *) p;
293 * Search for signature by checking equal to the swizzled value
294 * instead of calling another routine to perform a strcmp.
296 if (bios_32_ptr->signature == PCI_BIOS32_SD_VALUE) {
297 length = bios_32_ptr->length * PCI_BIOS32_PARAGRAPH_LEN;
298 if (bios_checksum(p, length)) {
300 * According to the spec, we're looking for the
301 * first 4KB-aligned address below the entrypoint
302 * listed in the header. The Service Directory code
303 * is guaranteed to occupy no more than 2 4KB pages.
305 map_entry = bios_32_ptr->entry_point & ~(PAGE_SIZE - 1);
306 map_offset = bios_32_ptr->entry_point - map_entry;
308 return cru_detect(map_entry, map_offset);
314 static int __devinit detect_cru_service(void)
320 * Search from 0x0f0000 through 0x0fffff, inclusive.
322 p = ioremap(PCI_ROM_BASE1, ROM_SIZE);
326 for (q = p; q < p + ROM_SIZE; q += 16) {
327 rc = bios32_present(q);
336 /* --64 Bit Bios------------------------------------------------------------ */
338 #define HPWDT_ARCH 64
342 "asminline_call: \n\t"
344 "movq %rsp, %rbp \n\t"
350 "movq %rsi, %r12 \n\t"
351 "movq %rdi, %r9 \n\t"
352 "movl 4(%r9),%ebx \n\t"
353 "movl 8(%r9),%ecx \n\t"
354 "movl 12(%r9),%edx \n\t"
355 "movl 16(%r9),%esi \n\t"
356 "movl 20(%r9),%edi \n\t"
357 "movl (%r9),%eax \n\t"
361 "movl %eax, (%r9) \n\t"
362 "movl %ebx, 4(%r9) \n\t"
363 "movl %ecx, 8(%r9) \n\t"
364 "movl %edx, 12(%r9) \n\t"
365 "movl %esi, 16(%r9) \n\t"
366 "movl %edi, 20(%r9) \n\t"
367 "movq %r12, %rax \n\t"
368 "movl %eax, 28(%r9) \n\t"
381 * Routine Description:
382 * This function checks whether or not a SMBIOS/DMI record is
383 * the 64bit CRU info or not
385 static void __devinit dmi_find_cru(const struct dmi_header *dm, void *dummy)
387 struct smbios_cru64_info *smbios_cru64_ptr;
388 unsigned long cru_physical_address;
390 if (dm->type == SMBIOS_CRU64_INFORMATION) {
391 smbios_cru64_ptr = (struct smbios_cru64_info *) dm;
392 if (smbios_cru64_ptr->signature == CRU_BIOS_SIGNATURE_VALUE) {
393 cru_physical_address =
394 smbios_cru64_ptr->physical_address +
395 smbios_cru64_ptr->double_offset;
396 cru_rom_addr = ioremap(cru_physical_address,
397 smbios_cru64_ptr->double_length);
398 set_memory_x((unsigned long)cru_rom_addr & PAGE_MASK,
399 smbios_cru64_ptr->double_length >> PAGE_SHIFT);
404 static int __devinit detect_cru_service(void)
408 dmi_walk(dmi_find_cru, NULL);
410 /* if cru_rom_addr has been set then we found a CRU service */
411 return ((cru_rom_addr != NULL) ? 0 : -ENODEV);
414 /* ------------------------------------------------------------------------- */
419 * Watchdog operations
421 static void hpwdt_start(void)
423 reload = (soft_margin * 1000) / 128;
424 iowrite16(reload, hpwdt_timer_reg);
425 iowrite16(0x85, hpwdt_timer_con);
428 static void hpwdt_stop(void)
432 data = ioread16(hpwdt_timer_con);
434 iowrite16(data, hpwdt_timer_con);
437 static void hpwdt_ping(void)
439 iowrite16(reload, hpwdt_timer_reg);
442 static int hpwdt_change_timer(int new_margin)
444 /* Arbitrary, can't find the card's limits */
445 if (new_margin < 30 || new_margin > 600) {
447 "hpwdt: New value passed in is invalid: %d seconds.\n",
452 soft_margin = new_margin;
454 "hpwdt: New timer passed in is %d seconds.\n",
456 reload = (soft_margin * 1000) / 128;
464 static int hpwdt_pretimeout(struct notifier_block *nb, unsigned long ulReason,
467 unsigned long rom_pl;
468 static int die_nmi_called;
470 if (ulReason != DIE_NMI && ulReason != DIE_NMI_IPI)
473 if (hpwdt_nmi_sourcing) {
474 spin_lock_irqsave(&rom_lock, rom_pl);
476 asminline_call(&cmn_regs, cru_rom_addr);
478 spin_unlock_irqrestore(&rom_lock, rom_pl);
479 if (cmn_regs.u1.ral == 0) {
480 printk(KERN_WARNING "hpwdt: An NMI occurred, "
481 "but unable to determine source.\n");
485 panic("An NMI occurred, please see the Integrated "
486 "Management Log for details.\n");
493 * /dev/watchdog handling
495 static int hpwdt_open(struct inode *inode, struct file *file)
497 /* /dev/watchdog can only be opened once */
498 if (test_and_set_bit(0, &hpwdt_is_open))
501 /* Start the watchdog */
505 return nonseekable_open(inode, file);
508 static int hpwdt_release(struct inode *inode, struct file *file)
510 /* Stop the watchdog */
511 if (expect_release == 42) {
515 "hpwdt: Unexpected close, not stopping watchdog!\n");
521 /* /dev/watchdog is being closed, make sure it can be re-opened */
522 clear_bit(0, &hpwdt_is_open);
527 static ssize_t hpwdt_write(struct file *file, const char __user *data,
528 size_t len, loff_t *ppos)
530 /* See if we got the magic character 'V' and reload the timer */
535 /* note: just in case someone wrote the magic character
536 * five months ago... */
539 /* scan to see whether or not we got the magic char. */
540 for (i = 0; i != len; i++) {
542 if (get_user(c, data + i))
549 /* someone wrote to us, we should reload the timer */
556 static struct watchdog_info ident = {
557 .options = WDIOF_SETTIMEOUT |
558 WDIOF_KEEPALIVEPING |
560 .identity = "HP iLO2 HW Watchdog Timer",
563 static long hpwdt_ioctl(struct file *file, unsigned int cmd,
566 void __user *argp = (void __user *)arg;
567 int __user *p = argp;
572 case WDIOC_GETSUPPORT:
574 if (copy_to_user(argp, &ident, sizeof(ident)))
578 case WDIOC_GETSTATUS:
579 case WDIOC_GETBOOTSTATUS:
580 ret = put_user(0, p);
583 case WDIOC_KEEPALIVE:
588 case WDIOC_SETTIMEOUT:
589 ret = get_user(new_margin, p);
593 ret = hpwdt_change_timer(new_margin);
599 case WDIOC_GETTIMEOUT:
600 ret = put_user(soft_margin, p);
609 static const struct file_operations hpwdt_fops = {
610 .owner = THIS_MODULE,
612 .write = hpwdt_write,
613 .unlocked_ioctl = hpwdt_ioctl,
615 .release = hpwdt_release,
618 static struct miscdevice hpwdt_miscdev = {
619 .minor = WATCHDOG_MINOR,
624 static struct notifier_block die_notifier = {
625 .notifier_call = hpwdt_pretimeout,
626 .priority = 0x7FFFFFFF,
633 #ifdef ARCH_HAS_NMI_WATCHDOG
634 static void __devinit hpwdt_check_nmi_sourcing(struct pci_dev *dev)
637 * If nmi_watchdog is turned off then we can turn on
638 * our nmi sourcing capability.
640 if (!nmi_watchdog_active())
641 hpwdt_nmi_sourcing = 1;
643 dev_warn(&dev->dev, "NMI sourcing is disabled. To enable this "
644 "functionality you must reboot with nmi_watchdog=0.\n");
647 static void __devinit hpwdt_check_nmi_sourcing(struct pci_dev *dev)
649 dev_warn(&dev->dev, "NMI sourcing is disabled. "
650 "Your kernel does not support a NMI Watchdog.\n");
654 static int __devinit hpwdt_init_one(struct pci_dev *dev,
655 const struct pci_device_id *ent)
660 * Check if we can do NMI sourcing or not
662 hpwdt_check_nmi_sourcing(dev);
665 * First let's find out if we are on an iLO2 server. We will
666 * not run on a legacy ASM box.
667 * So we only support the G5 ProLiant servers and higher.
669 if (dev->subsystem_vendor != PCI_VENDOR_ID_HP) {
671 "This server does not have an iLO2 ASIC.\n");
675 if (pci_enable_device(dev)) {
677 "Not possible to enable PCI Device: 0x%x:0x%x.\n",
678 ent->vendor, ent->device);
682 pci_mem_addr = pci_iomap(dev, 1, 0x80);
685 "Unable to detect the iLO2 server memory.\n");
687 goto error_pci_iomap;
689 hpwdt_timer_reg = pci_mem_addr + 0x70;
690 hpwdt_timer_con = pci_mem_addr + 0x72;
692 /* Make sure that we have a valid soft_margin */
693 if (hpwdt_change_timer(soft_margin))
694 hpwdt_change_timer(DEFAULT_MARGIN);
697 * We need to map the ROM to get the CRU service.
698 * For 32 bit Operating Systems we need to go through the 32 Bit
699 * BIOS Service Directory
700 * For 64 bit Operating Systems we get that service through SMBIOS.
702 retval = detect_cru_service();
705 "Unable to detect the %d Bit CRU Service.\n",
711 * We know this is the only CRU call we need to make so lets keep as
712 * few instructions as possible once the NMI comes in.
714 cmn_regs.u1.rah = 0x0D;
715 cmn_regs.u1.ral = 0x02;
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).\n",
737 HPWDT_VERSION, soft_margin, nowayout,
738 (allow_kdump == 0) ? "OFF" : "ON");
743 unregister_die_notifier(&die_notifier);
746 iounmap(cru_rom_addr);
748 pci_iounmap(dev, pci_mem_addr);
750 pci_disable_device(dev);
754 static void __devexit hpwdt_exit(struct pci_dev *dev)
759 misc_deregister(&hpwdt_miscdev);
760 unregister_die_notifier(&die_notifier);
763 iounmap(cru_rom_addr);
764 pci_iounmap(dev, pci_mem_addr);
765 pci_disable_device(dev);
768 static struct pci_driver hpwdt_driver = {
770 .id_table = hpwdt_devices,
771 .probe = hpwdt_init_one,
772 .remove = __devexit_p(hpwdt_exit),
775 static void __exit hpwdt_cleanup(void)
777 pci_unregister_driver(&hpwdt_driver);
780 static int __init hpwdt_init(void)
782 return pci_register_driver(&hpwdt_driver);
785 MODULE_AUTHOR("Tom Mingarelli");
786 MODULE_DESCRIPTION("hp watchdog driver");
787 MODULE_LICENSE("GPL");
788 MODULE_VERSION(HPWDT_VERSION);
789 MODULE_ALIAS_MISCDEV(WATCHDOG_MINOR);
791 module_param(soft_margin, int, 0);
792 MODULE_PARM_DESC(soft_margin, "Watchdog timeout in seconds");
794 module_param(allow_kdump, int, 0);
795 MODULE_PARM_DESC(allow_kdump, "Start a kernel dump after NMI occurs");
797 module_param(nowayout, int, 0);
798 MODULE_PARM_DESC(nowayout, "Watchdog cannot be stopped once started (default="
799 __MODULE_STRING(WATCHDOG_NOWAYOUT) ")");
801 module_init(hpwdt_init);
802 module_exit(hpwdt_cleanup);