2 * Octeon Watchdog driver
4 * Copyright (C) 2007, 2008, 2009, 2010 Cavium Networks
6 * Some parts derived from wdt.c
8 * (c) Copyright 1996-1997 Alan Cox <alan@lxorguk.ukuu.org.uk>,
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License
13 * as published by the Free Software Foundation; either version
14 * 2 of the License, or (at your option) any later version.
16 * Neither Alan Cox nor CymruNet Ltd. admit liability nor provide
17 * warranty for any of this software. This material is provided
18 * "AS-IS" and at no charge.
20 * (c) Copyright 1995 Alan Cox <alan@lxorguk.ukuu.org.uk>
22 * This file is subject to the terms and conditions of the GNU General Public
23 * License. See the file "COPYING" in the main directory of this archive
27 * The OCTEON watchdog has a maximum timeout of 2^32 * io_clock.
28 * For most systems this is less than 10 seconds, so to allow for
29 * software to request longer watchdog heartbeats, we maintain software
30 * counters to count multiples of the base rate. If the system locks
31 * up in such a manner that we can not run the software counters, the
32 * only result is a watchdog reset sooner than was requested. But
33 * that is OK, because in this case userspace would likely not be able
34 * to do anything anyhow.
36 * The hardware watchdog interval we call the period. The OCTEON
37 * watchdog goes through several stages, after the first period an
38 * irq is asserted, then if it is not reset, after the next period NMI
39 * is asserted, then after an additional period a chip wide soft reset.
40 * So for the software counters, we reset watchdog after each period
41 * and decrement the counter. But for the last two periods we need to
42 * let the watchdog progress to the NMI stage so we disable the irq
43 * and let it proceed. Once in the NMI, we print the register state
44 * to the serial port and then wait for the reset.
46 * A watchdog is maintained for each CPU in the system, that way if
47 * one CPU suffers a lockup, we also get a register dump and reset.
48 * The userspace ping resets the watchdog on all CPUs.
50 * Before userspace opens the watchdog device, we still run the
51 * watchdogs to catch any lockups that may be kernel related.
55 #include <linux/miscdevice.h>
56 #include <linux/interrupt.h>
57 #include <linux/watchdog.h>
58 #include <linux/cpumask.h>
59 #include <linux/bitops.h>
60 #include <linux/kernel.h>
61 #include <linux/module.h>
62 #include <linux/string.h>
63 #include <linux/delay.h>
64 #include <linux/cpu.h>
65 #include <linux/smp.h>
67 #include <linux/irq.h>
69 #include <asm/mipsregs.h>
72 #include <asm/octeon/octeon.h>
74 /* The count needed to achieve timeout_sec. */
75 static unsigned int timeout_cnt
;
77 /* The maximum period supported. */
78 static unsigned int max_timeout_sec
;
80 /* The current period. */
81 static unsigned int timeout_sec
;
83 /* Set to non-zero when userspace countdown mode active */
84 static int do_coundown
;
85 static unsigned int countdown_reset
;
86 static unsigned int per_cpu_countdown
[NR_CPUS
];
88 static cpumask_t irq_enabled_cpus
;
90 #define WD_TIMO 60 /* Default heartbeat = 60 seconds */
92 static int heartbeat
= WD_TIMO
;
93 module_param(heartbeat
, int, S_IRUGO
);
94 MODULE_PARM_DESC(heartbeat
,
95 "Watchdog heartbeat in seconds. (0 < heartbeat, default="
96 __MODULE_STRING(WD_TIMO
) ")");
98 static int nowayout
= WATCHDOG_NOWAYOUT
;
99 module_param(nowayout
, int, S_IRUGO
);
100 MODULE_PARM_DESC(nowayout
,
101 "Watchdog cannot be stopped once started (default="
102 __MODULE_STRING(WATCHDOG_NOWAYOUT
) ")");
104 static unsigned long octeon_wdt_is_open
;
105 static char expect_close
;
107 static u32 __initdata nmi_stage1_insns
[64];
108 /* We need one branch and therefore one relocation per target label. */
109 static struct uasm_label __initdata labels
[5];
110 static struct uasm_reloc __initdata relocs
[5];
113 label_enter_bootloader
= 1
116 /* Some CP0 registers */
118 #define C0_CVMMEMCTL 11, 7
119 #define C0_STATUS 12, 0
120 #define C0_EBASE 15, 1
121 #define C0_DESAVE 31, 0
123 void octeon_wdt_nmi_stage2(void);
125 static void __init
octeon_wdt_build_stage1(void)
129 u32
*p
= nmi_stage1_insns
;
130 #ifdef CONFIG_HOTPLUG_CPU
131 struct uasm_label
*l
= labels
;
132 struct uasm_reloc
*r
= relocs
;
136 * For the next few instructions running the debugger may
137 * cause corruption of k0 in the saved registers. Since we're
138 * about to crash, nobody probably cares.
140 * Save K0 into the debug scratch register
142 uasm_i_dmtc0(&p
, K0
, C0_DESAVE
);
144 uasm_i_mfc0(&p
, K0
, C0_STATUS
);
145 #ifdef CONFIG_HOTPLUG_CPU
146 uasm_il_bbit0(&p
, &r
, K0
, ilog2(ST0_NMI
), label_enter_bootloader
);
148 /* Force 64-bit addressing enabled */
149 uasm_i_ori(&p
, K0
, K0
, ST0_UX
| ST0_SX
| ST0_KX
);
150 uasm_i_mtc0(&p
, K0
, C0_STATUS
);
152 #ifdef CONFIG_HOTPLUG_CPU
153 uasm_i_mfc0(&p
, K0
, C0_EBASE
);
154 /* Coreid number in K0 */
155 uasm_i_andi(&p
, K0
, K0
, 0xf);
156 /* 8 * coreid in bits 16-31 */
157 uasm_i_dsll_safe(&p
, K0
, K0
, 3 + 16);
158 uasm_i_ori(&p
, K0
, K0
, 0x8001);
159 uasm_i_dsll_safe(&p
, K0
, K0
, 16);
160 uasm_i_ori(&p
, K0
, K0
, 0x0700);
161 uasm_i_drotr_safe(&p
, K0
, K0
, 32);
163 * Should result in: 0x8001,0700,0000,8*coreid which is
164 * CVMX_CIU_WDOGX(coreid) - 0x0500
166 * Now ld K0, CVMX_CIU_WDOGX(coreid)
168 uasm_i_ld(&p
, K0
, 0x500, K0
);
170 * If bit one set handle the NMI as a watchdog event.
171 * otherwise transfer control to bootloader.
173 uasm_il_bbit0(&p
, &r
, K0
, 1, label_enter_bootloader
);
177 /* Clear Dcache so cvmseg works right. */
178 uasm_i_cache(&p
, 1, 0, 0);
180 /* Use K0 to do a read/modify/write of CVMMEMCTL */
181 uasm_i_dmfc0(&p
, K0
, C0_CVMMEMCTL
);
182 /* Clear out the size of CVMSEG */
183 uasm_i_dins(&p
, K0
, 0, 0, 6);
184 /* Set CVMSEG to its largest value */
185 uasm_i_ori(&p
, K0
, K0
, 0x1c0 | 54);
186 /* Store the CVMMEMCTL value */
187 uasm_i_dmtc0(&p
, K0
, C0_CVMMEMCTL
);
189 /* Load the address of the second stage handler */
190 UASM_i_LA(&p
, K0
, (long)octeon_wdt_nmi_stage2
);
192 uasm_i_dmfc0(&p
, K0
, C0_DESAVE
);
194 #ifdef CONFIG_HOTPLUG_CPU
195 uasm_build_label(&l
, p
, label_enter_bootloader
);
196 /* Jump to the bootloader and restore K0 */
197 UASM_i_LA(&p
, K0
, (long)octeon_bootloader_entry_addr
);
199 uasm_i_dmfc0(&p
, K0
, C0_DESAVE
);
201 uasm_resolve_relocs(relocs
, labels
);
203 len
= (int)(p
- nmi_stage1_insns
);
204 pr_debug("Synthesized NMI stage 1 handler (%d instructions).\n", len
);
206 pr_debug("\t.set push\n");
207 pr_debug("\t.set noreorder\n");
208 for (i
= 0; i
< len
; i
++)
209 pr_debug("\t.word 0x%08x\n", nmi_stage1_insns
[i
]);
210 pr_debug("\t.set pop\n");
213 panic("NMI stage 1 handler exceeds 32 instructions, was %d\n", len
);
216 static int cpu2core(int cpu
)
219 return cpu_logical_map(cpu
);
221 return cvmx_get_core_num();
225 static int core2cpu(int coreid
)
228 return cpu_number_map(coreid
);
235 * Poke the watchdog when an interrupt is received
242 static irqreturn_t
octeon_wdt_poke_irq(int cpl
, void *dev_id
)
244 unsigned int core
= cvmx_get_core_num();
245 int cpu
= core2cpu(core
);
248 if (per_cpu_countdown
[cpu
] > 0) {
249 /* We're alive, poke the watchdog */
250 cvmx_write_csr(CVMX_CIU_PP_POKEX(core
), 1);
251 per_cpu_countdown
[cpu
]--;
253 /* Bad news, you are about to reboot. */
254 disable_irq_nosync(cpl
);
255 cpumask_clear_cpu(cpu
, &irq_enabled_cpus
);
258 /* Not open, just ping away... */
259 cvmx_write_csr(CVMX_CIU_PP_POKEX(core
), 1);
265 extern int prom_putchar(char c
);
268 * Write a string to the uart
270 * @str: String to write
272 static void octeon_wdt_write_string(const char *str
)
274 /* Just loop writing one byte at a time */
276 prom_putchar(*str
++);
280 * Write a hex number out of the uart
282 * @value: Number to display
283 * @digits: Number of digits to print (1 to 16)
285 static void octeon_wdt_write_hex(u64 value
, int digits
)
289 for (d
= 0; d
< digits
; d
++) {
290 v
= (value
>> ((digits
- d
- 1) * 4)) & 0xf;
292 prom_putchar('a' + v
- 10);
294 prom_putchar('0' + v
);
298 const char *reg_name
[] = {
299 "$0", "at", "v0", "v1", "a0", "a1", "a2", "a3",
300 "a4", "a5", "a6", "a7", "t0", "t1", "t2", "t3",
301 "s0", "s1", "s2", "s3", "s4", "s5", "s6", "s7",
302 "t8", "t9", "k0", "k1", "gp", "sp", "s8", "ra"
306 * NMI stage 3 handler. NMIs are handled in the following manner:
307 * 1) The first NMI handler enables CVMSEG and transfers from
308 * the bootbus region into normal memory. It is careful to not
309 * destroy any registers.
310 * 2) The second stage handler uses CVMSEG to save the registers
311 * and create a stack for C code. It then calls the third level
312 * handler with one argument, a pointer to the register values.
313 * 3) The third, and final, level handler is the following C
314 * function that prints out some useful infomration.
316 * @reg: Pointer to register state before the NMI
318 void octeon_wdt_nmi_stage3(u64 reg
[32])
322 unsigned int coreid
= cvmx_get_core_num();
324 * Save status and cause early to get them before any changes
327 u64 cp0_cause
= read_c0_cause();
328 u64 cp0_status
= read_c0_status();
329 u64 cp0_error_epc
= read_c0_errorepc();
330 u64 cp0_epc
= read_c0_epc();
332 /* Delay so output from all cores output is not jumbled together. */
333 __delay(100000000ull * coreid
);
335 octeon_wdt_write_string("\r\n*** NMI Watchdog interrupt on Core 0x");
336 octeon_wdt_write_hex(coreid
, 1);
337 octeon_wdt_write_string(" ***\r\n");
338 for (i
= 0; i
< 32; i
++) {
339 octeon_wdt_write_string("\t");
340 octeon_wdt_write_string(reg_name
[i
]);
341 octeon_wdt_write_string("\t0x");
342 octeon_wdt_write_hex(reg
[i
], 16);
344 octeon_wdt_write_string("\r\n");
346 octeon_wdt_write_string("\terr_epc\t0x");
347 octeon_wdt_write_hex(cp0_error_epc
, 16);
349 octeon_wdt_write_string("\tepc\t0x");
350 octeon_wdt_write_hex(cp0_epc
, 16);
351 octeon_wdt_write_string("\r\n");
353 octeon_wdt_write_string("\tstatus\t0x");
354 octeon_wdt_write_hex(cp0_status
, 16);
355 octeon_wdt_write_string("\tcause\t0x");
356 octeon_wdt_write_hex(cp0_cause
, 16);
357 octeon_wdt_write_string("\r\n");
359 octeon_wdt_write_string("\tsum0\t0x");
360 octeon_wdt_write_hex(cvmx_read_csr(CVMX_CIU_INTX_SUM0(coreid
* 2)), 16);
361 octeon_wdt_write_string("\ten0\t0x");
362 octeon_wdt_write_hex(cvmx_read_csr(CVMX_CIU_INTX_EN0(coreid
* 2)), 16);
363 octeon_wdt_write_string("\r\n");
365 octeon_wdt_write_string("*** Chip soft reset soon ***\r\n");
368 static void octeon_wdt_disable_interrupt(int cpu
)
372 union cvmx_ciu_wdogx ciu_wdog
;
374 core
= cpu2core(cpu
);
376 irq
= OCTEON_IRQ_WDOG0
+ core
;
378 /* Poke the watchdog to clear out its state */
379 cvmx_write_csr(CVMX_CIU_PP_POKEX(core
), 1);
381 /* Disable the hardware. */
383 cvmx_write_csr(CVMX_CIU_WDOGX(core
), ciu_wdog
.u64
);
385 free_irq(irq
, octeon_wdt_poke_irq
);
388 static void octeon_wdt_setup_interrupt(int cpu
)
392 union cvmx_ciu_wdogx ciu_wdog
;
394 core
= cpu2core(cpu
);
396 /* Disable it before doing anything with the interrupts. */
398 cvmx_write_csr(CVMX_CIU_WDOGX(core
), ciu_wdog
.u64
);
400 per_cpu_countdown
[cpu
] = countdown_reset
;
402 irq
= OCTEON_IRQ_WDOG0
+ core
;
404 if (request_irq(irq
, octeon_wdt_poke_irq
,
405 IRQF_NO_THREAD
, "octeon_wdt", octeon_wdt_poke_irq
))
406 panic("octeon_wdt: Couldn't obtain irq %d", irq
);
408 cpumask_set_cpu(cpu
, &irq_enabled_cpus
);
410 /* Poke the watchdog to clear out its state */
411 cvmx_write_csr(CVMX_CIU_PP_POKEX(core
), 1);
413 /* Finally enable the watchdog now that all handlers are installed */
415 ciu_wdog
.s
.len
= timeout_cnt
;
416 ciu_wdog
.s
.mode
= 3; /* 3 = Interrupt + NMI + Soft-Reset */
417 cvmx_write_csr(CVMX_CIU_WDOGX(core
), ciu_wdog
.u64
);
420 static int octeon_wdt_cpu_callback(struct notifier_block
*nfb
,
421 unsigned long action
, void *hcpu
)
423 unsigned int cpu
= (unsigned long)hcpu
;
426 case CPU_DOWN_PREPARE
:
427 octeon_wdt_disable_interrupt(cpu
);
430 case CPU_DOWN_FAILED
:
431 octeon_wdt_setup_interrupt(cpu
);
439 static void octeon_wdt_ping(void)
444 for_each_online_cpu(cpu
) {
445 coreid
= cpu2core(cpu
);
446 cvmx_write_csr(CVMX_CIU_PP_POKEX(coreid
), 1);
447 per_cpu_countdown
[cpu
] = countdown_reset
;
448 if ((countdown_reset
|| !do_coundown
) &&
449 !cpumask_test_cpu(cpu
, &irq_enabled_cpus
)) {
450 /* We have to enable the irq */
451 int irq
= OCTEON_IRQ_WDOG0
+ coreid
;
453 cpumask_set_cpu(cpu
, &irq_enabled_cpus
);
458 static void octeon_wdt_calc_parameters(int t
)
460 unsigned int periods
;
462 timeout_sec
= max_timeout_sec
;
466 * Find the largest interrupt period, that can evenly divide
467 * the requested heartbeat time.
469 while ((t
% timeout_sec
) != 0)
472 periods
= t
/ timeout_sec
;
475 * The last two periods are after the irq is disabled, and
476 * then to the nmi, so we subtract them off.
479 countdown_reset
= periods
> 2 ? periods
- 2 : 0;
481 timeout_cnt
= ((octeon_get_io_clock_rate() >> 8) * timeout_sec
) >> 8;
484 static int octeon_wdt_set_heartbeat(int t
)
488 union cvmx_ciu_wdogx ciu_wdog
;
493 octeon_wdt_calc_parameters(t
);
495 for_each_online_cpu(cpu
) {
496 coreid
= cpu2core(cpu
);
497 cvmx_write_csr(CVMX_CIU_PP_POKEX(coreid
), 1);
499 ciu_wdog
.s
.len
= timeout_cnt
;
500 ciu_wdog
.s
.mode
= 3; /* 3 = Interrupt + NMI + Soft-Reset */
501 cvmx_write_csr(CVMX_CIU_WDOGX(coreid
), ciu_wdog
.u64
);
502 cvmx_write_csr(CVMX_CIU_PP_POKEX(coreid
), 1);
504 octeon_wdt_ping(); /* Get the irqs back on. */
510 * @file: file handle to the watchdog
511 * @buf: buffer to write (unused as data does not matter here
512 * @count: count of bytes
513 * @ppos: pointer to the position to write. No seeks allowed
515 * A write to a watchdog device is defined as a keepalive signal. Any
516 * write of data will do, as we we don't define content meaning.
519 static ssize_t
octeon_wdt_write(struct file
*file
, const char __user
*buf
,
520 size_t count
, loff_t
*ppos
)
526 /* In case it was set long ago */
529 for (i
= 0; i
!= count
; i
++) {
531 if (get_user(c
, buf
+ i
))
544 * @file: file handle to the device
545 * @cmd: watchdog command
546 * @arg: argument pointer
548 * The watchdog API defines a common set of functions for all
549 * watchdogs according to their available features. We only
550 * actually usefully support querying capabilities and setting
554 static long octeon_wdt_ioctl(struct file
*file
, unsigned int cmd
,
557 void __user
*argp
= (void __user
*)arg
;
558 int __user
*p
= argp
;
561 static struct watchdog_info ident
= {
562 .options
= WDIOF_SETTIMEOUT
|
565 .firmware_version
= 1,
566 .identity
= "OCTEON",
570 case WDIOC_GETSUPPORT
:
571 return copy_to_user(argp
, &ident
, sizeof(ident
)) ? -EFAULT
: 0;
572 case WDIOC_GETSTATUS
:
573 case WDIOC_GETBOOTSTATUS
:
574 return put_user(0, p
);
575 case WDIOC_KEEPALIVE
:
578 case WDIOC_SETTIMEOUT
:
579 if (get_user(new_heartbeat
, p
))
581 if (octeon_wdt_set_heartbeat(new_heartbeat
))
584 case WDIOC_GETTIMEOUT
:
585 return put_user(heartbeat
, p
);
593 * @inode: inode of device
594 * @file: file handle to device
596 * The watchdog device has been opened. The watchdog device is single
597 * open and on opening we do a ping to reset the counters.
600 static int octeon_wdt_open(struct inode
*inode
, struct file
*file
)
602 if (test_and_set_bit(0, &octeon_wdt_is_open
))
609 return nonseekable_open(inode
, file
);
613 * octeon_wdt_release:
614 * @inode: inode to board
615 * @file: file handle to board
617 * The watchdog has a configurable API. There is a religious dispute
618 * between people who want their watchdog to be able to shut down and
619 * those who want to be sure if the watchdog manager dies the machine
620 * reboots. In the former case we disable the counters, in the latter
621 * case you have to open it again very soon.
624 static int octeon_wdt_release(struct inode
*inode
, struct file
*file
)
630 pr_crit("octeon_wdt: WDT device closed unexpectedly. WDT will not stop!\n");
632 clear_bit(0, &octeon_wdt_is_open
);
637 static const struct file_operations octeon_wdt_fops
= {
638 .owner
= THIS_MODULE
,
640 .write
= octeon_wdt_write
,
641 .unlocked_ioctl
= octeon_wdt_ioctl
,
642 .open
= octeon_wdt_open
,
643 .release
= octeon_wdt_release
,
646 static struct miscdevice octeon_wdt_miscdev
= {
647 .minor
= WATCHDOG_MINOR
,
649 .fops
= &octeon_wdt_fops
,
652 static struct notifier_block octeon_wdt_cpu_notifier
= {
653 .notifier_call
= octeon_wdt_cpu_callback
,
658 * Module/ driver initialization.
660 * Returns Zero on success
662 static int __init
octeon_wdt_init(void)
670 * Watchdog time expiration length = The 16 bits of LEN
671 * represent the most significant bits of a 24 bit decrementer
672 * that decrements every 256 cycles.
674 * Try for a timeout of 5 sec, if that fails a smaller number
680 timeout_cnt
= ((octeon_get_io_clock_rate() >> 8) * max_timeout_sec
) >> 8;
681 } while (timeout_cnt
> 65535);
683 BUG_ON(timeout_cnt
== 0);
685 octeon_wdt_calc_parameters(heartbeat
);
687 pr_info("octeon_wdt: Initial granularity %d Sec.\n", timeout_sec
);
689 ret
= misc_register(&octeon_wdt_miscdev
);
691 pr_err("octeon_wdt: cannot register miscdev on minor=%d (err=%d)\n",
692 WATCHDOG_MINOR
, ret
);
696 /* Build the NMI handler ... */
697 octeon_wdt_build_stage1();
699 /* ... and install it. */
700 ptr
= (u64
*) nmi_stage1_insns
;
701 for (i
= 0; i
< 16; i
++) {
702 cvmx_write_csr(CVMX_MIO_BOOT_LOC_ADR
, i
* 8);
703 cvmx_write_csr(CVMX_MIO_BOOT_LOC_DAT
, ptr
[i
]);
705 cvmx_write_csr(CVMX_MIO_BOOT_LOC_CFGX(0), 0x81fc0000);
707 cpumask_clear(&irq_enabled_cpus
);
709 for_each_online_cpu(cpu
)
710 octeon_wdt_setup_interrupt(cpu
);
712 register_hotcpu_notifier(&octeon_wdt_cpu_notifier
);
718 * Module / driver shutdown
720 static void __exit
octeon_wdt_cleanup(void)
724 misc_deregister(&octeon_wdt_miscdev
);
726 unregister_hotcpu_notifier(&octeon_wdt_cpu_notifier
);
728 for_each_online_cpu(cpu
) {
729 int core
= cpu2core(cpu
);
730 /* Disable the watchdog */
731 cvmx_write_csr(CVMX_CIU_WDOGX(core
), 0);
732 /* Free the interrupt handler */
733 free_irq(OCTEON_IRQ_WDOG0
+ core
, octeon_wdt_poke_irq
);
736 * Disable the boot-bus memory, the code it points to is soon
739 cvmx_write_csr(CVMX_MIO_BOOT_LOC_CFGX(0), 0);
742 MODULE_LICENSE("GPL");
743 MODULE_AUTHOR("Cavium Networks <support@caviumnetworks.com>");
744 MODULE_DESCRIPTION("Cavium Networks Octeon Watchdog driver.");
745 module_init(octeon_wdt_init
);
746 module_exit(octeon_wdt_cleanup
);