Linux Audio

Check our new training course

Loading...
v6.13.7
  1// SPDX-License-Identifier: GPL-2.0+
  2/*
  3 * Octeon Watchdog driver
  4 *
  5 * Copyright (C) 2007-2017 Cavium, Inc.
  6 *
  7 * Converted to use WATCHDOG_CORE by Aaro Koskinen <aaro.koskinen@iki.fi>.
  8 *
  9 * Some parts derived from wdt.c
 10 *
 11 *	(c) Copyright 1996-1997 Alan Cox <alan@lxorguk.ukuu.org.uk>,
 12 *						All Rights Reserved.
 13 *
 
 
 
 
 
 14 *	Neither Alan Cox nor CymruNet Ltd. admit liability nor provide
 15 *	warranty for any of this software. This material is provided
 16 *	"AS-IS" and at no charge.
 17 *
 18 *	(c) Copyright 1995    Alan Cox <alan@lxorguk.ukuu.org.uk>
 19 *
 
 
 
 
 
 20 * The OCTEON watchdog has a maximum timeout of 2^32 * io_clock.
 21 * For most systems this is less than 10 seconds, so to allow for
 22 * software to request longer watchdog heartbeats, we maintain software
 23 * counters to count multiples of the base rate.  If the system locks
 24 * up in such a manner that we can not run the software counters, the
 25 * only result is a watchdog reset sooner than was requested.  But
 26 * that is OK, because in this case userspace would likely not be able
 27 * to do anything anyhow.
 28 *
 29 * The hardware watchdog interval we call the period.  The OCTEON
 30 * watchdog goes through several stages, after the first period an
 31 * irq is asserted, then if it is not reset, after the next period NMI
 32 * is asserted, then after an additional period a chip wide soft reset.
 33 * So for the software counters, we reset watchdog after each period
 34 * and decrement the counter.  But for the last two periods we need to
 35 * let the watchdog progress to the NMI stage so we disable the irq
 36 * and let it proceed.  Once in the NMI, we print the register state
 37 * to the serial port and then wait for the reset.
 38 *
 39 * A watchdog is maintained for each CPU in the system, that way if
 40 * one CPU suffers a lockup, we also get a register dump and reset.
 41 * The userspace ping resets the watchdog on all CPUs.
 42 *
 43 * Before userspace opens the watchdog device, we still run the
 44 * watchdogs to catch any lockups that may be kernel related.
 45 *
 46 */
 47
 48#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
 49
 50#include <linux/interrupt.h>
 51#include <linux/watchdog.h>
 52#include <linux/cpumask.h>
 
 
 53#include <linux/module.h>
 
 54#include <linux/delay.h>
 55#include <linux/cpu.h>
 
 
 56#include <linux/irq.h>
 57#include <linux/irqdomain.h>
 58
 59#include <asm/mipsregs.h>
 60#include <asm/uasm.h>
 61
 62#include <asm/octeon/octeon.h>
 63#include <asm/octeon/cvmx-boot-vector.h>
 64#include <asm/octeon/cvmx-ciu2-defs.h>
 65#include <asm/octeon/cvmx-rst-defs.h>
 66
 67/* Watchdog interrupt major block number (8 MSBs of intsn) */
 68#define WD_BLOCK_NUMBER		0x01
 69
 70static int divisor;
 71
 72/* The count needed to achieve timeout_sec. */
 73static unsigned int timeout_cnt;
 74
 75/* The maximum period supported. */
 76static unsigned int max_timeout_sec;
 77
 78/* The current period.  */
 79static unsigned int timeout_sec;
 80
 81/* Set to non-zero when userspace countdown mode active */
 82static bool do_countdown;
 83static unsigned int countdown_reset;
 84static unsigned int per_cpu_countdown[NR_CPUS];
 85
 86static cpumask_t irq_enabled_cpus;
 87
 88#define WD_TIMO 60			/* Default heartbeat = 60 seconds */
 89
 90#define CVMX_GSERX_SCRATCH(offset) (CVMX_ADD_IO_SEG(0x0001180090000020ull) + ((offset) & 15) * 0x1000000ull)
 91
 92static int heartbeat = WD_TIMO;
 93module_param(heartbeat, int, 0444);
 94MODULE_PARM_DESC(heartbeat,
 95	"Watchdog heartbeat in seconds. (0 < heartbeat, default="
 96				__MODULE_STRING(WD_TIMO) ")");
 97
 98static bool nowayout = WATCHDOG_NOWAYOUT;
 99module_param(nowayout, bool, 0444);
100MODULE_PARM_DESC(nowayout,
101	"Watchdog cannot be stopped once started (default="
102				__MODULE_STRING(WATCHDOG_NOWAYOUT) ")");
103
104static int disable;
105module_param(disable, int, 0444);
106MODULE_PARM_DESC(disable,
107	"Disable the watchdog entirely (default=0)");
 
 
 
 
108
109static struct cvmx_boot_vector_element *octeon_wdt_bootvector;
 
 
 
 
 
110
111void octeon_wdt_nmi_stage2(void);
112
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
113static int cpu2core(int cpu)
114{
115#ifdef CONFIG_SMP
116	return cpu_logical_map(cpu) & 0x3f;
117#else
118	return cvmx_get_core_num();
119#endif
120}
121
 
 
 
 
 
 
 
 
 
122/**
123 * octeon_wdt_poke_irq - Poke the watchdog when an interrupt is received
124 *
125 * @cpl:
126 * @dev_id:
127 *
128 * Returns
129 */
130static irqreturn_t octeon_wdt_poke_irq(int cpl, void *dev_id)
131{
132	int cpu = raw_smp_processor_id();
133	unsigned int core = cpu2core(cpu);
134	int node = cpu_to_node(cpu);
135
136	if (do_countdown) {
137		if (per_cpu_countdown[cpu] > 0) {
138			/* We're alive, poke the watchdog */
139			cvmx_write_csr_node(node, CVMX_CIU_PP_POKEX(core), 1);
140			per_cpu_countdown[cpu]--;
141		} else {
142			/* Bad news, you are about to reboot. */
143			disable_irq_nosync(cpl);
144			cpumask_clear_cpu(cpu, &irq_enabled_cpus);
145		}
146	} else {
147		/* Not open, just ping away... */
148		cvmx_write_csr_node(node, CVMX_CIU_PP_POKEX(core), 1);
149	}
150	return IRQ_HANDLED;
151}
152
153/* From setup.c */
154extern int prom_putchar(char c);
155
156/**
157 * octeon_wdt_write_string - Write a string to the uart
158 *
159 * @str:        String to write
160 */
161static void octeon_wdt_write_string(const char *str)
162{
163	/* Just loop writing one byte at a time */
164	while (*str)
165		prom_putchar(*str++);
166}
167
168/**
169 * octeon_wdt_write_hex() - Write a hex number out of the uart
170 *
171 * @value:      Number to display
172 * @digits:     Number of digits to print (1 to 16)
173 */
174static void octeon_wdt_write_hex(u64 value, int digits)
175{
176	int d;
177	int v;
178
179	for (d = 0; d < digits; d++) {
180		v = (value >> ((digits - d - 1) * 4)) & 0xf;
181		if (v >= 10)
182			prom_putchar('a' + v - 10);
183		else
184			prom_putchar('0' + v);
185	}
186}
187
188static const char reg_name[][3] = {
189	"$0", "at", "v0", "v1", "a0", "a1", "a2", "a3",
190	"a4", "a5", "a6", "a7", "t0", "t1", "t2", "t3",
191	"s0", "s1", "s2", "s3", "s4", "s5", "s6", "s7",
192	"t8", "t9", "k0", "k1", "gp", "sp", "s8", "ra"
193};
194
195/**
196 * octeon_wdt_nmi_stage3:
197 *
198 * NMI stage 3 handler. NMIs are handled in the following manner:
199 * 1) The first NMI handler enables CVMSEG and transfers from
200 * the bootbus region into normal memory. It is careful to not
201 * destroy any registers.
202 * 2) The second stage handler uses CVMSEG to save the registers
203 * and create a stack for C code. It then calls the third level
204 * handler with one argument, a pointer to the register values.
205 * 3) The third, and final, level handler is the following C
206 * function that prints out some useful infomration.
207 *
208 * @reg:    Pointer to register state before the NMI
209 */
210void octeon_wdt_nmi_stage3(u64 reg[32])
211{
212	u64 i;
213
214	unsigned int coreid = cvmx_get_core_num();
215	/*
216	 * Save status and cause early to get them before any changes
217	 * might happen.
218	 */
219	u64 cp0_cause = read_c0_cause();
220	u64 cp0_status = read_c0_status();
221	u64 cp0_error_epc = read_c0_errorepc();
222	u64 cp0_epc = read_c0_epc();
223
224	/* Delay so output from all cores output is not jumbled together. */
225	udelay(85000 * coreid);
226
227	octeon_wdt_write_string("\r\n*** NMI Watchdog interrupt on Core 0x");
228	octeon_wdt_write_hex(coreid, 2);
229	octeon_wdt_write_string(" ***\r\n");
230	for (i = 0; i < 32; i++) {
231		octeon_wdt_write_string("\t");
232		octeon_wdt_write_string(reg_name[i]);
233		octeon_wdt_write_string("\t0x");
234		octeon_wdt_write_hex(reg[i], 16);
235		if (i & 1)
236			octeon_wdt_write_string("\r\n");
237	}
238	octeon_wdt_write_string("\terr_epc\t0x");
239	octeon_wdt_write_hex(cp0_error_epc, 16);
240
241	octeon_wdt_write_string("\tepc\t0x");
242	octeon_wdt_write_hex(cp0_epc, 16);
243	octeon_wdt_write_string("\r\n");
244
245	octeon_wdt_write_string("\tstatus\t0x");
246	octeon_wdt_write_hex(cp0_status, 16);
247	octeon_wdt_write_string("\tcause\t0x");
248	octeon_wdt_write_hex(cp0_cause, 16);
249	octeon_wdt_write_string("\r\n");
250
251	/* The CIU register is different for each Octeon model. */
252	if (OCTEON_IS_MODEL(OCTEON_CN68XX)) {
253		octeon_wdt_write_string("\tsrc_wd\t0x");
254		octeon_wdt_write_hex(cvmx_read_csr(CVMX_CIU2_SRC_PPX_IP2_WDOG(coreid)), 16);
255		octeon_wdt_write_string("\ten_wd\t0x");
256		octeon_wdt_write_hex(cvmx_read_csr(CVMX_CIU2_EN_PPX_IP2_WDOG(coreid)), 16);
257		octeon_wdt_write_string("\r\n");
258		octeon_wdt_write_string("\tsrc_rml\t0x");
259		octeon_wdt_write_hex(cvmx_read_csr(CVMX_CIU2_SRC_PPX_IP2_RML(coreid)), 16);
260		octeon_wdt_write_string("\ten_rml\t0x");
261		octeon_wdt_write_hex(cvmx_read_csr(CVMX_CIU2_EN_PPX_IP2_RML(coreid)), 16);
262		octeon_wdt_write_string("\r\n");
263		octeon_wdt_write_string("\tsum\t0x");
264		octeon_wdt_write_hex(cvmx_read_csr(CVMX_CIU2_SUM_PPX_IP2(coreid)), 16);
265		octeon_wdt_write_string("\r\n");
266	} else if (!octeon_has_feature(OCTEON_FEATURE_CIU3)) {
267		octeon_wdt_write_string("\tsum0\t0x");
268		octeon_wdt_write_hex(cvmx_read_csr(CVMX_CIU_INTX_SUM0(coreid * 2)), 16);
269		octeon_wdt_write_string("\ten0\t0x");
270		octeon_wdt_write_hex(cvmx_read_csr(CVMX_CIU_INTX_EN0(coreid * 2)), 16);
271		octeon_wdt_write_string("\r\n");
272	}
273
274	octeon_wdt_write_string("*** Chip soft reset soon ***\r\n");
275
276	/*
277	 * G-30204: We must trigger a soft reset before watchdog
278	 * does an incomplete job of doing it.
279	 */
280	if (OCTEON_IS_OCTEON3() && !OCTEON_IS_MODEL(OCTEON_CN70XX)) {
281		u64 scr;
282		unsigned int node = cvmx_get_node_num();
283		unsigned int lcore = cvmx_get_local_core_num();
284		union cvmx_ciu_wdogx ciu_wdog;
285
286		/*
287		 * Wait for other cores to print out information, but
288		 * not too long.  Do the soft reset before watchdog
289		 * can trigger it.
290		 */
291		do {
292			ciu_wdog.u64 = cvmx_read_csr_node(node, CVMX_CIU_WDOGX(lcore));
293		} while (ciu_wdog.s.cnt > 0x10000);
294
295		scr = cvmx_read_csr_node(0, CVMX_GSERX_SCRATCH(0));
296		scr |= 1 << 11; /* Indicate watchdog in bit 11 */
297		cvmx_write_csr_node(0, CVMX_GSERX_SCRATCH(0), scr);
298		cvmx_write_csr_node(0, CVMX_RST_SOFT_RST, 1);
299	}
300}
301
302static int octeon_wdt_cpu_to_irq(int cpu)
303{
304	unsigned int coreid;
305	int node;
306	int irq;
307
308	coreid = cpu2core(cpu);
309	node = cpu_to_node(cpu);
310
311	if (octeon_has_feature(OCTEON_FEATURE_CIU3)) {
312		struct irq_domain *domain;
313		int hwirq;
314
315		domain = octeon_irq_get_block_domain(node,
316						     WD_BLOCK_NUMBER);
317		hwirq = WD_BLOCK_NUMBER << 12 | 0x200 | coreid;
318		irq = irq_find_mapping(domain, hwirq);
319	} else {
320		irq = OCTEON_IRQ_WDOG0 + coreid;
321	}
322	return irq;
323}
324
325static int octeon_wdt_cpu_pre_down(unsigned int cpu)
326{
327	unsigned int core;
328	int node;
329	union cvmx_ciu_wdogx ciu_wdog;
330
331	core = cpu2core(cpu);
332
333	node = cpu_to_node(cpu);
334
335	/* Poke the watchdog to clear out its state */
336	cvmx_write_csr_node(node, CVMX_CIU_PP_POKEX(core), 1);
337
338	/* Disable the hardware. */
339	ciu_wdog.u64 = 0;
340	cvmx_write_csr_node(node, CVMX_CIU_WDOGX(core), ciu_wdog.u64);
341
342	free_irq(octeon_wdt_cpu_to_irq(cpu), octeon_wdt_poke_irq);
343	return 0;
344}
345
346static int octeon_wdt_cpu_online(unsigned int cpu)
347{
348	unsigned int core;
349	unsigned int irq;
350	union cvmx_ciu_wdogx ciu_wdog;
351	int node;
352	struct irq_domain *domain;
353	int hwirq;
354
355	core = cpu2core(cpu);
356	node = cpu_to_node(cpu);
357
358	octeon_wdt_bootvector[core].target_ptr = (u64)octeon_wdt_nmi_stage2;
359
360	/* Disable it before doing anything with the interrupts. */
361	ciu_wdog.u64 = 0;
362	cvmx_write_csr_node(node, CVMX_CIU_WDOGX(core), ciu_wdog.u64);
363
364	per_cpu_countdown[cpu] = countdown_reset;
365
366	if (octeon_has_feature(OCTEON_FEATURE_CIU3)) {
367		/* Must get the domain for the watchdog block */
368		domain = octeon_irq_get_block_domain(node, WD_BLOCK_NUMBER);
369
370		/* Get a irq for the wd intsn (hardware interrupt) */
371		hwirq = WD_BLOCK_NUMBER << 12 | 0x200 | core;
372		irq = irq_create_mapping(domain, hwirq);
373		irqd_set_trigger_type(irq_get_irq_data(irq),
374				      IRQ_TYPE_EDGE_RISING);
375	} else
376		irq = OCTEON_IRQ_WDOG0 + core;
377
378	if (request_irq(irq, octeon_wdt_poke_irq,
379			IRQF_NO_THREAD, "octeon_wdt", octeon_wdt_poke_irq))
380		panic("octeon_wdt: Couldn't obtain irq %d", irq);
381
382	/* Must set the irq affinity here */
383	if (octeon_has_feature(OCTEON_FEATURE_CIU3)) {
384		irq_set_affinity(irq, cpumask_of(cpu));
385	}
386
387	cpumask_set_cpu(cpu, &irq_enabled_cpus);
388
389	/* Poke the watchdog to clear out its state */
390	cvmx_write_csr_node(node, CVMX_CIU_PP_POKEX(core), 1);
391
392	/* Finally enable the watchdog now that all handlers are installed */
393	ciu_wdog.u64 = 0;
394	ciu_wdog.s.len = timeout_cnt;
395	ciu_wdog.s.mode = 3;	/* 3 = Interrupt + NMI + Soft-Reset */
396	cvmx_write_csr_node(node, CVMX_CIU_WDOGX(core), ciu_wdog.u64);
397
398	return 0;
399}
400
401static int octeon_wdt_ping(struct watchdog_device __always_unused *wdog)
402{
403	int cpu;
404	int coreid;
405	int node;
406
407	if (disable)
408		return 0;
409
410	for_each_online_cpu(cpu) {
411		coreid = cpu2core(cpu);
412		node = cpu_to_node(cpu);
413		cvmx_write_csr_node(node, CVMX_CIU_PP_POKEX(coreid), 1);
414		per_cpu_countdown[cpu] = countdown_reset;
415		if ((countdown_reset || !do_countdown) &&
416		    !cpumask_test_cpu(cpu, &irq_enabled_cpus)) {
417			/* We have to enable the irq */
418			enable_irq(octeon_wdt_cpu_to_irq(cpu));
 
 
419			cpumask_set_cpu(cpu, &irq_enabled_cpus);
420		}
421	}
422	return 0;
423}
424
425static void octeon_wdt_calc_parameters(int t)
426{
427	unsigned int periods;
428
429	timeout_sec = max_timeout_sec;
430
431
432	/*
433	 * Find the largest interrupt period, that can evenly divide
434	 * the requested heartbeat time.
435	 */
436	while ((t % timeout_sec) != 0)
437		timeout_sec--;
438
439	periods = t / timeout_sec;
440
441	/*
442	 * The last two periods are after the irq is disabled, and
443	 * then to the nmi, so we subtract them off.
444	 */
445
446	countdown_reset = periods > 2 ? periods - 2 : 0;
447	heartbeat = t;
448	timeout_cnt = ((octeon_get_io_clock_rate() / divisor) * timeout_sec) >> 8;
449}
450
451static int octeon_wdt_set_timeout(struct watchdog_device *wdog,
452				  unsigned int t)
453{
454	int cpu;
455	int coreid;
456	union cvmx_ciu_wdogx ciu_wdog;
457	int node;
458
459	if (t <= 0)
460		return -1;
461
462	octeon_wdt_calc_parameters(t);
463
464	if (disable)
465		return 0;
466
467	for_each_online_cpu(cpu) {
468		coreid = cpu2core(cpu);
469		node = cpu_to_node(cpu);
470		cvmx_write_csr_node(node, CVMX_CIU_PP_POKEX(coreid), 1);
471		ciu_wdog.u64 = 0;
472		ciu_wdog.s.len = timeout_cnt;
473		ciu_wdog.s.mode = 3;	/* 3 = Interrupt + NMI + Soft-Reset */
474		cvmx_write_csr_node(node, CVMX_CIU_WDOGX(coreid), ciu_wdog.u64);
475		cvmx_write_csr_node(node, CVMX_CIU_PP_POKEX(coreid), 1);
476	}
477	octeon_wdt_ping(wdog); /* Get the irqs back on. */
478	return 0;
479}
480
481static int octeon_wdt_start(struct watchdog_device *wdog)
482{
483	octeon_wdt_ping(wdog);
484	do_countdown = 1;
485	return 0;
486}
487
488static int octeon_wdt_stop(struct watchdog_device *wdog)
489{
490	do_countdown = 0;
491	octeon_wdt_ping(wdog);
492	return 0;
493}
494
495static const struct watchdog_info octeon_wdt_info = {
496	.options = WDIOF_SETTIMEOUT | WDIOF_MAGICCLOSE | WDIOF_KEEPALIVEPING,
497	.identity = "OCTEON",
498};
499
500static const struct watchdog_ops octeon_wdt_ops = {
501	.owner		= THIS_MODULE,
502	.start		= octeon_wdt_start,
503	.stop		= octeon_wdt_stop,
504	.ping		= octeon_wdt_ping,
505	.set_timeout	= octeon_wdt_set_timeout,
506};
507
508static struct watchdog_device octeon_wdt = {
509	.info	= &octeon_wdt_info,
510	.ops	= &octeon_wdt_ops,
511};
512
513static enum cpuhp_state octeon_wdt_online;
514/**
515 * octeon_wdt_init - Module/ driver initialization.
516 *
517 * Returns Zero on success
518 */
519static int __init octeon_wdt_init(void)
520{
 
521	int ret;
522
523	octeon_wdt_bootvector = cvmx_boot_vector_get();
524	if (!octeon_wdt_bootvector) {
525		pr_err("Error: Cannot allocate boot vector.\n");
526		return -ENOMEM;
527	}
528
529	if (OCTEON_IS_MODEL(OCTEON_CN68XX))
530		divisor = 0x200;
531	else if (OCTEON_IS_MODEL(OCTEON_CN78XX))
532		divisor = 0x400;
533	else
534		divisor = 0x100;
535
536	/*
537	 * Watchdog time expiration length = The 16 bits of LEN
538	 * represent the most significant bits of a 24 bit decrementer
539	 * that decrements every divisor cycle.
540	 *
541	 * Try for a timeout of 5 sec, if that fails a smaller number
542	 * of even seconds,
543	 */
544	max_timeout_sec = 6;
545	do {
546		max_timeout_sec--;
547		timeout_cnt = ((octeon_get_io_clock_rate() / divisor) * max_timeout_sec) >> 8;
 
548	} while (timeout_cnt > 65535);
549
550	BUG_ON(timeout_cnt == 0);
551
552	octeon_wdt_calc_parameters(heartbeat);
553
554	pr_info("Initial granularity %d Sec\n", timeout_sec);
555
556	octeon_wdt.timeout	= timeout_sec;
557	octeon_wdt.max_timeout	= UINT_MAX;
558
559	watchdog_set_nowayout(&octeon_wdt, nowayout);
560
561	ret = watchdog_register_device(&octeon_wdt);
562	if (ret)
 
563		return ret;
 
 
 
 
564
565	if (disable) {
566		pr_notice("disabled\n");
567		return 0;
 
 
568	}
 
569
570	cpumask_clear(&irq_enabled_cpus);
571
572	ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "watchdog/octeon:online",
573				octeon_wdt_cpu_online, octeon_wdt_cpu_pre_down);
574	if (ret < 0)
575		goto err;
576	octeon_wdt_online = ret;
577	return 0;
578err:
579	cvmx_write_csr(CVMX_MIO_BOOT_LOC_CFGX(0), 0);
580	watchdog_unregister_device(&octeon_wdt);
581	return ret;
582}
583
584/**
585 * octeon_wdt_cleanup - Module / driver shutdown
586 */
587static void __exit octeon_wdt_cleanup(void)
588{
589	watchdog_unregister_device(&octeon_wdt);
590
591	if (disable)
592		return;
593
594	cpuhp_remove_state(octeon_wdt_online);
595
596	/*
597	 * Disable the boot-bus memory, the code it points to is soon
598	 * to go missing.
599	 */
600	cvmx_write_csr(CVMX_MIO_BOOT_LOC_CFGX(0), 0);
601}
602
603MODULE_LICENSE("GPL");
604MODULE_AUTHOR("Cavium Inc. <support@cavium.com>");
605MODULE_DESCRIPTION("Cavium Inc. OCTEON Watchdog driver.");
606module_init(octeon_wdt_init);
607module_exit(octeon_wdt_cleanup);
v4.10.11
 
  1/*
  2 * Octeon Watchdog driver
  3 *
  4 * Copyright (C) 2007, 2008, 2009, 2010 Cavium Networks
  5 *
  6 * Converted to use WATCHDOG_CORE by Aaro Koskinen <aaro.koskinen@iki.fi>.
  7 *
  8 * Some parts derived from wdt.c
  9 *
 10 *	(c) Copyright 1996-1997 Alan Cox <alan@lxorguk.ukuu.org.uk>,
 11 *						All Rights Reserved.
 12 *
 13 *	This program is free software; you can redistribute it and/or
 14 *	modify it under the terms of the GNU General Public License
 15 *	as published by the Free Software Foundation; either version
 16 *	2 of the License, or (at your option) any later version.
 17 *
 18 *	Neither Alan Cox nor CymruNet Ltd. admit liability nor provide
 19 *	warranty for any of this software. This material is provided
 20 *	"AS-IS" and at no charge.
 21 *
 22 *	(c) Copyright 1995    Alan Cox <alan@lxorguk.ukuu.org.uk>
 23 *
 24 * This file is subject to the terms and conditions of the GNU General Public
 25 * License.  See the file "COPYING" in the main directory of this archive
 26 * for more details.
 27 *
 28 *
 29 * The OCTEON watchdog has a maximum timeout of 2^32 * io_clock.
 30 * For most systems this is less than 10 seconds, so to allow for
 31 * software to request longer watchdog heartbeats, we maintain software
 32 * counters to count multiples of the base rate.  If the system locks
 33 * up in such a manner that we can not run the software counters, the
 34 * only result is a watchdog reset sooner than was requested.  But
 35 * that is OK, because in this case userspace would likely not be able
 36 * to do anything anyhow.
 37 *
 38 * The hardware watchdog interval we call the period.  The OCTEON
 39 * watchdog goes through several stages, after the first period an
 40 * irq is asserted, then if it is not reset, after the next period NMI
 41 * is asserted, then after an additional period a chip wide soft reset.
 42 * So for the software counters, we reset watchdog after each period
 43 * and decrement the counter.  But for the last two periods we need to
 44 * let the watchdog progress to the NMI stage so we disable the irq
 45 * and let it proceed.  Once in the NMI, we print the register state
 46 * to the serial port and then wait for the reset.
 47 *
 48 * A watchdog is maintained for each CPU in the system, that way if
 49 * one CPU suffers a lockup, we also get a register dump and reset.
 50 * The userspace ping resets the watchdog on all CPUs.
 51 *
 52 * Before userspace opens the watchdog device, we still run the
 53 * watchdogs to catch any lockups that may be kernel related.
 54 *
 55 */
 56
 57#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
 58
 59#include <linux/interrupt.h>
 60#include <linux/watchdog.h>
 61#include <linux/cpumask.h>
 62#include <linux/bitops.h>
 63#include <linux/kernel.h>
 64#include <linux/module.h>
 65#include <linux/string.h>
 66#include <linux/delay.h>
 67#include <linux/cpu.h>
 68#include <linux/smp.h>
 69#include <linux/fs.h>
 70#include <linux/irq.h>
 
 71
 72#include <asm/mipsregs.h>
 73#include <asm/uasm.h>
 74
 75#include <asm/octeon/octeon.h>
 
 
 
 
 
 
 
 
 76
 77/* The count needed to achieve timeout_sec. */
 78static unsigned int timeout_cnt;
 79
 80/* The maximum period supported. */
 81static unsigned int max_timeout_sec;
 82
 83/* The current period.  */
 84static unsigned int timeout_sec;
 85
 86/* Set to non-zero when userspace countdown mode active */
 87static int do_coundown;
 88static unsigned int countdown_reset;
 89static unsigned int per_cpu_countdown[NR_CPUS];
 90
 91static cpumask_t irq_enabled_cpus;
 92
 93#define WD_TIMO 60			/* Default heartbeat = 60 seconds */
 94
 
 
 95static int heartbeat = WD_TIMO;
 96module_param(heartbeat, int, S_IRUGO);
 97MODULE_PARM_DESC(heartbeat,
 98	"Watchdog heartbeat in seconds. (0 < heartbeat, default="
 99				__MODULE_STRING(WD_TIMO) ")");
100
101static bool nowayout = WATCHDOG_NOWAYOUT;
102module_param(nowayout, bool, S_IRUGO);
103MODULE_PARM_DESC(nowayout,
104	"Watchdog cannot be stopped once started (default="
105				__MODULE_STRING(WATCHDOG_NOWAYOUT) ")");
106
107static u32 nmi_stage1_insns[64] __initdata;
108/* We need one branch and therefore one relocation per target label. */
109static struct uasm_label labels[5] __initdata;
110static struct uasm_reloc relocs[5] __initdata;
111
112enum lable_id {
113	label_enter_bootloader = 1
114};
115
116/* Some CP0 registers */
117#define K0		26
118#define C0_CVMMEMCTL 11, 7
119#define C0_STATUS 12, 0
120#define C0_EBASE 15, 1
121#define C0_DESAVE 31, 0
122
123void octeon_wdt_nmi_stage2(void);
124
125static void __init octeon_wdt_build_stage1(void)
126{
127	int i;
128	int len;
129	u32 *p = nmi_stage1_insns;
130#ifdef CONFIG_HOTPLUG_CPU
131	struct uasm_label *l = labels;
132	struct uasm_reloc *r = relocs;
133#endif
134
135	/*
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.
139	 *
140	 * Save K0 into the debug scratch register
141	 */
142	uasm_i_dmtc0(&p, K0, C0_DESAVE);
143
144	uasm_i_mfc0(&p, K0, C0_STATUS);
145#ifdef CONFIG_HOTPLUG_CPU
146	if (octeon_bootloader_entry_addr)
147		uasm_il_bbit0(&p, &r, K0, ilog2(ST0_NMI),
148			      label_enter_bootloader);
149#endif
150	/* Force 64-bit addressing enabled */
151	uasm_i_ori(&p, K0, K0, ST0_UX | ST0_SX | ST0_KX);
152	uasm_i_mtc0(&p, K0, C0_STATUS);
153
154#ifdef CONFIG_HOTPLUG_CPU
155	if (octeon_bootloader_entry_addr) {
156		uasm_i_mfc0(&p, K0, C0_EBASE);
157		/* Coreid number in K0 */
158		uasm_i_andi(&p, K0, K0, 0xf);
159		/* 8 * coreid in bits 16-31 */
160		uasm_i_dsll_safe(&p, K0, K0, 3 + 16);
161		uasm_i_ori(&p, K0, K0, 0x8001);
162		uasm_i_dsll_safe(&p, K0, K0, 16);
163		uasm_i_ori(&p, K0, K0, 0x0700);
164		uasm_i_drotr_safe(&p, K0, K0, 32);
165		/*
166		 * Should result in: 0x8001,0700,0000,8*coreid which is
167		 * CVMX_CIU_WDOGX(coreid) - 0x0500
168		 *
169		 * Now ld K0, CVMX_CIU_WDOGX(coreid)
170		 */
171		uasm_i_ld(&p, K0, 0x500, K0);
172		/*
173		 * If bit one set handle the NMI as a watchdog event.
174		 * otherwise transfer control to bootloader.
175		 */
176		uasm_il_bbit0(&p, &r, K0, 1, label_enter_bootloader);
177		uasm_i_nop(&p);
178	}
179#endif
180
181	/* Clear Dcache so cvmseg works right. */
182	uasm_i_cache(&p, 1, 0, 0);
183
184	/* Use K0 to do a read/modify/write of CVMMEMCTL */
185	uasm_i_dmfc0(&p, K0, C0_CVMMEMCTL);
186	/* Clear out the size of CVMSEG	*/
187	uasm_i_dins(&p, K0, 0, 0, 6);
188	/* Set CVMSEG to its largest value */
189	uasm_i_ori(&p, K0, K0, 0x1c0 | 54);
190	/* Store the CVMMEMCTL value */
191	uasm_i_dmtc0(&p, K0, C0_CVMMEMCTL);
192
193	/* Load the address of the second stage handler */
194	UASM_i_LA(&p, K0, (long)octeon_wdt_nmi_stage2);
195	uasm_i_jr(&p, K0);
196	uasm_i_dmfc0(&p, K0, C0_DESAVE);
197
198#ifdef CONFIG_HOTPLUG_CPU
199	if (octeon_bootloader_entry_addr) {
200		uasm_build_label(&l, p, label_enter_bootloader);
201		/* Jump to the bootloader and restore K0 */
202		UASM_i_LA(&p, K0, (long)octeon_bootloader_entry_addr);
203		uasm_i_jr(&p, K0);
204		uasm_i_dmfc0(&p, K0, C0_DESAVE);
205	}
206#endif
207	uasm_resolve_relocs(relocs, labels);
208
209	len = (int)(p - nmi_stage1_insns);
210	pr_debug("Synthesized NMI stage 1 handler (%d instructions)\n", len);
211
212	pr_debug("\t.set push\n");
213	pr_debug("\t.set noreorder\n");
214	for (i = 0; i < len; i++)
215		pr_debug("\t.word 0x%08x\n", nmi_stage1_insns[i]);
216	pr_debug("\t.set pop\n");
217
218	if (len > 32)
219		panic("NMI stage 1 handler exceeds 32 instructions, was %d\n",
220		      len);
221}
222
223static int cpu2core(int cpu)
224{
225#ifdef CONFIG_SMP
226	return cpu_logical_map(cpu);
227#else
228	return cvmx_get_core_num();
229#endif
230}
231
232static int core2cpu(int coreid)
233{
234#ifdef CONFIG_SMP
235	return cpu_number_map(coreid);
236#else
237	return 0;
238#endif
239}
240
241/**
242 * Poke the watchdog when an interrupt is received
243 *
244 * @cpl:
245 * @dev_id:
246 *
247 * Returns
248 */
249static irqreturn_t octeon_wdt_poke_irq(int cpl, void *dev_id)
250{
251	unsigned int core = cvmx_get_core_num();
252	int cpu = core2cpu(core);
 
253
254	if (do_coundown) {
255		if (per_cpu_countdown[cpu] > 0) {
256			/* We're alive, poke the watchdog */
257			cvmx_write_csr(CVMX_CIU_PP_POKEX(core), 1);
258			per_cpu_countdown[cpu]--;
259		} else {
260			/* Bad news, you are about to reboot. */
261			disable_irq_nosync(cpl);
262			cpumask_clear_cpu(cpu, &irq_enabled_cpus);
263		}
264	} else {
265		/* Not open, just ping away... */
266		cvmx_write_csr(CVMX_CIU_PP_POKEX(core), 1);
267	}
268	return IRQ_HANDLED;
269}
270
271/* From setup.c */
272extern int prom_putchar(char c);
273
274/**
275 * Write a string to the uart
276 *
277 * @str:        String to write
278 */
279static void octeon_wdt_write_string(const char *str)
280{
281	/* Just loop writing one byte at a time */
282	while (*str)
283		prom_putchar(*str++);
284}
285
286/**
287 * Write a hex number out of the uart
288 *
289 * @value:      Number to display
290 * @digits:     Number of digits to print (1 to 16)
291 */
292static void octeon_wdt_write_hex(u64 value, int digits)
293{
294	int d;
295	int v;
296
297	for (d = 0; d < digits; d++) {
298		v = (value >> ((digits - d - 1) * 4)) & 0xf;
299		if (v >= 10)
300			prom_putchar('a' + v - 10);
301		else
302			prom_putchar('0' + v);
303	}
304}
305
306static const char reg_name[][3] = {
307	"$0", "at", "v0", "v1", "a0", "a1", "a2", "a3",
308	"a4", "a5", "a6", "a7", "t0", "t1", "t2", "t3",
309	"s0", "s1", "s2", "s3", "s4", "s5", "s6", "s7",
310	"t8", "t9", "k0", "k1", "gp", "sp", "s8", "ra"
311};
312
313/**
 
 
314 * NMI stage 3 handler. NMIs are handled in the following manner:
315 * 1) The first NMI handler enables CVMSEG and transfers from
316 * the bootbus region into normal memory. It is careful to not
317 * destroy any registers.
318 * 2) The second stage handler uses CVMSEG to save the registers
319 * and create a stack for C code. It then calls the third level
320 * handler with one argument, a pointer to the register values.
321 * 3) The third, and final, level handler is the following C
322 * function that prints out some useful infomration.
323 *
324 * @reg:    Pointer to register state before the NMI
325 */
326void octeon_wdt_nmi_stage3(u64 reg[32])
327{
328	u64 i;
329
330	unsigned int coreid = cvmx_get_core_num();
331	/*
332	 * Save status and cause early to get them before any changes
333	 * might happen.
334	 */
335	u64 cp0_cause = read_c0_cause();
336	u64 cp0_status = read_c0_status();
337	u64 cp0_error_epc = read_c0_errorepc();
338	u64 cp0_epc = read_c0_epc();
339
340	/* Delay so output from all cores output is not jumbled together. */
341	__delay(100000000ull * coreid);
342
343	octeon_wdt_write_string("\r\n*** NMI Watchdog interrupt on Core 0x");
344	octeon_wdt_write_hex(coreid, 1);
345	octeon_wdt_write_string(" ***\r\n");
346	for (i = 0; i < 32; i++) {
347		octeon_wdt_write_string("\t");
348		octeon_wdt_write_string(reg_name[i]);
349		octeon_wdt_write_string("\t0x");
350		octeon_wdt_write_hex(reg[i], 16);
351		if (i & 1)
352			octeon_wdt_write_string("\r\n");
353	}
354	octeon_wdt_write_string("\terr_epc\t0x");
355	octeon_wdt_write_hex(cp0_error_epc, 16);
356
357	octeon_wdt_write_string("\tepc\t0x");
358	octeon_wdt_write_hex(cp0_epc, 16);
359	octeon_wdt_write_string("\r\n");
360
361	octeon_wdt_write_string("\tstatus\t0x");
362	octeon_wdt_write_hex(cp0_status, 16);
363	octeon_wdt_write_string("\tcause\t0x");
364	octeon_wdt_write_hex(cp0_cause, 16);
365	octeon_wdt_write_string("\r\n");
366
367	octeon_wdt_write_string("\tsum0\t0x");
368	octeon_wdt_write_hex(cvmx_read_csr(CVMX_CIU_INTX_SUM0(coreid * 2)), 16);
369	octeon_wdt_write_string("\ten0\t0x");
370	octeon_wdt_write_hex(cvmx_read_csr(CVMX_CIU_INTX_EN0(coreid * 2)), 16);
371	octeon_wdt_write_string("\r\n");
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
372
373	octeon_wdt_write_string("*** Chip soft reset soon ***\r\n");
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
374}
375
376static int octeon_wdt_cpu_pre_down(unsigned int cpu)
377{
378	unsigned int core;
379	unsigned int irq;
380	union cvmx_ciu_wdogx ciu_wdog;
381
382	core = cpu2core(cpu);
383
384	irq = OCTEON_IRQ_WDOG0 + core;
385
386	/* Poke the watchdog to clear out its state */
387	cvmx_write_csr(CVMX_CIU_PP_POKEX(core), 1);
388
389	/* Disable the hardware. */
390	ciu_wdog.u64 = 0;
391	cvmx_write_csr(CVMX_CIU_WDOGX(core), ciu_wdog.u64);
392
393	free_irq(irq, octeon_wdt_poke_irq);
394	return 0;
395}
396
397static int octeon_wdt_cpu_online(unsigned int cpu)
398{
399	unsigned int core;
400	unsigned int irq;
401	union cvmx_ciu_wdogx ciu_wdog;
 
 
 
402
403	core = cpu2core(cpu);
 
 
 
404
405	/* Disable it before doing anything with the interrupts. */
406	ciu_wdog.u64 = 0;
407	cvmx_write_csr(CVMX_CIU_WDOGX(core), ciu_wdog.u64);
408
409	per_cpu_countdown[cpu] = countdown_reset;
410
411	irq = OCTEON_IRQ_WDOG0 + core;
 
 
 
 
 
 
 
 
 
 
412
413	if (request_irq(irq, octeon_wdt_poke_irq,
414			IRQF_NO_THREAD, "octeon_wdt", octeon_wdt_poke_irq))
415		panic("octeon_wdt: Couldn't obtain irq %d", irq);
416
 
 
 
 
 
417	cpumask_set_cpu(cpu, &irq_enabled_cpus);
418
419	/* Poke the watchdog to clear out its state */
420	cvmx_write_csr(CVMX_CIU_PP_POKEX(core), 1);
421
422	/* Finally enable the watchdog now that all handlers are installed */
423	ciu_wdog.u64 = 0;
424	ciu_wdog.s.len = timeout_cnt;
425	ciu_wdog.s.mode = 3;	/* 3 = Interrupt + NMI + Soft-Reset */
426	cvmx_write_csr(CVMX_CIU_WDOGX(core), ciu_wdog.u64);
427
428	return 0;
429}
430
431static int octeon_wdt_ping(struct watchdog_device __always_unused *wdog)
432{
433	int cpu;
434	int coreid;
 
 
 
 
435
436	for_each_online_cpu(cpu) {
437		coreid = cpu2core(cpu);
438		cvmx_write_csr(CVMX_CIU_PP_POKEX(coreid), 1);
 
439		per_cpu_countdown[cpu] = countdown_reset;
440		if ((countdown_reset || !do_coundown) &&
441		    !cpumask_test_cpu(cpu, &irq_enabled_cpus)) {
442			/* We have to enable the irq */
443			int irq = OCTEON_IRQ_WDOG0 + coreid;
444
445			enable_irq(irq);
446			cpumask_set_cpu(cpu, &irq_enabled_cpus);
447		}
448	}
449	return 0;
450}
451
452static void octeon_wdt_calc_parameters(int t)
453{
454	unsigned int periods;
455
456	timeout_sec = max_timeout_sec;
457
458
459	/*
460	 * Find the largest interrupt period, that can evenly divide
461	 * the requested heartbeat time.
462	 */
463	while ((t % timeout_sec) != 0)
464		timeout_sec--;
465
466	periods = t / timeout_sec;
467
468	/*
469	 * The last two periods are after the irq is disabled, and
470	 * then to the nmi, so we subtract them off.
471	 */
472
473	countdown_reset = periods > 2 ? periods - 2 : 0;
474	heartbeat = t;
475	timeout_cnt = ((octeon_get_io_clock_rate() >> 8) * timeout_sec) >> 8;
476}
477
478static int octeon_wdt_set_timeout(struct watchdog_device *wdog,
479				  unsigned int t)
480{
481	int cpu;
482	int coreid;
483	union cvmx_ciu_wdogx ciu_wdog;
 
484
485	if (t <= 0)
486		return -1;
487
488	octeon_wdt_calc_parameters(t);
489
 
 
 
490	for_each_online_cpu(cpu) {
491		coreid = cpu2core(cpu);
492		cvmx_write_csr(CVMX_CIU_PP_POKEX(coreid), 1);
 
493		ciu_wdog.u64 = 0;
494		ciu_wdog.s.len = timeout_cnt;
495		ciu_wdog.s.mode = 3;	/* 3 = Interrupt + NMI + Soft-Reset */
496		cvmx_write_csr(CVMX_CIU_WDOGX(coreid), ciu_wdog.u64);
497		cvmx_write_csr(CVMX_CIU_PP_POKEX(coreid), 1);
498	}
499	octeon_wdt_ping(wdog); /* Get the irqs back on. */
500	return 0;
501}
502
503static int octeon_wdt_start(struct watchdog_device *wdog)
504{
505	octeon_wdt_ping(wdog);
506	do_coundown = 1;
507	return 0;
508}
509
510static int octeon_wdt_stop(struct watchdog_device *wdog)
511{
512	do_coundown = 0;
513	octeon_wdt_ping(wdog);
514	return 0;
515}
516
517static const struct watchdog_info octeon_wdt_info = {
518	.options = WDIOF_SETTIMEOUT | WDIOF_MAGICCLOSE | WDIOF_KEEPALIVEPING,
519	.identity = "OCTEON",
520};
521
522static const struct watchdog_ops octeon_wdt_ops = {
523	.owner		= THIS_MODULE,
524	.start		= octeon_wdt_start,
525	.stop		= octeon_wdt_stop,
526	.ping		= octeon_wdt_ping,
527	.set_timeout	= octeon_wdt_set_timeout,
528};
529
530static struct watchdog_device octeon_wdt = {
531	.info	= &octeon_wdt_info,
532	.ops	= &octeon_wdt_ops,
533};
534
535static enum cpuhp_state octeon_wdt_online;
536/**
537 * Module/ driver initialization.
538 *
539 * Returns Zero on success
540 */
541static int __init octeon_wdt_init(void)
542{
543	int i;
544	int ret;
545	u64 *ptr;
 
 
 
 
 
 
 
 
 
 
 
 
546
547	/*
548	 * Watchdog time expiration length = The 16 bits of LEN
549	 * represent the most significant bits of a 24 bit decrementer
550	 * that decrements every 256 cycles.
551	 *
552	 * Try for a timeout of 5 sec, if that fails a smaller number
553	 * of even seconds,
554	 */
555	max_timeout_sec = 6;
556	do {
557		max_timeout_sec--;
558		timeout_cnt = ((octeon_get_io_clock_rate() >> 8) *
559			      max_timeout_sec) >> 8;
560	} while (timeout_cnt > 65535);
561
562	BUG_ON(timeout_cnt == 0);
563
564	octeon_wdt_calc_parameters(heartbeat);
565
566	pr_info("Initial granularity %d Sec\n", timeout_sec);
567
568	octeon_wdt.timeout	= timeout_sec;
569	octeon_wdt.max_timeout	= UINT_MAX;
570
571	watchdog_set_nowayout(&octeon_wdt, nowayout);
572
573	ret = watchdog_register_device(&octeon_wdt);
574	if (ret) {
575		pr_err("watchdog_register_device() failed: %d\n", ret);
576		return ret;
577	}
578
579	/* Build the NMI handler ... */
580	octeon_wdt_build_stage1();
581
582	/* ... and install it. */
583	ptr = (u64 *) nmi_stage1_insns;
584	for (i = 0; i < 16; i++) {
585		cvmx_write_csr(CVMX_MIO_BOOT_LOC_ADR, i * 8);
586		cvmx_write_csr(CVMX_MIO_BOOT_LOC_DAT, ptr[i]);
587	}
588	cvmx_write_csr(CVMX_MIO_BOOT_LOC_CFGX(0), 0x81fc0000);
589
590	cpumask_clear(&irq_enabled_cpus);
591
592	ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "watchdog/octeon:online",
593				octeon_wdt_cpu_online, octeon_wdt_cpu_pre_down);
594	if (ret < 0)
595		goto err;
596	octeon_wdt_online = ret;
597	return 0;
598err:
599	cvmx_write_csr(CVMX_MIO_BOOT_LOC_CFGX(0), 0);
600	watchdog_unregister_device(&octeon_wdt);
601	return ret;
602}
603
604/**
605 * Module / driver shutdown
606 */
607static void __exit octeon_wdt_cleanup(void)
608{
609	watchdog_unregister_device(&octeon_wdt);
 
 
 
 
610	cpuhp_remove_state(octeon_wdt_online);
611
612	/*
613	 * Disable the boot-bus memory, the code it points to is soon
614	 * to go missing.
615	 */
616	cvmx_write_csr(CVMX_MIO_BOOT_LOC_CFGX(0), 0);
617}
618
619MODULE_LICENSE("GPL");
620MODULE_AUTHOR("Cavium Networks <support@caviumnetworks.com>");
621MODULE_DESCRIPTION("Cavium Networks Octeon Watchdog driver.");
622module_init(octeon_wdt_init);
623module_exit(octeon_wdt_cleanup);