Blackfin: implement nmi_watchdog for SMP on BF561

Signed-off-by: Graf Yang <graf.yang@analog.com>
Signed-off-by: Mike Frysinger <vapier@gentoo.org>

arch/blackfin/Kconfig.debug

@@ -238,6 +238,15 @@ config EARLY_PRINTK
 	  all of this lives in the init section and is thrown away after the
 	  kernel boots completely.
 
+config NMI_WATCHDOG
+	bool "Enable NMI watchdog to help debugging lockup on SMP"
+	default n
+	depends on (SMP && !BFIN_SCRATCH_REG_RETN)
+	help
+	  If any CPU in the system does not execute the period local timer
+	  interrupt for more than 5 seconds, then the NMI handler dumps debug
+	  information. This information can be used to debug the lockup.
+
 config CPLB_INFO
 	bool "Display the CPLB information"
 	help

arch/blackfin/include/asm/irq.h

@@ -38,4 +38,8 @@
 
 #include <asm-generic/irq.h>
 
+#ifdef CONFIG_NMI_WATCHDOG
+# define ARCH_HAS_NMI_WATCHDOG
+#endif
+
 #endif				/* _BFIN_IRQ_H_ */

arch/blackfin/include/asm/nmi.h

@@ -0,0 +1,12 @@
+/*
+ * Copyright 2010 Analog Devices Inc.
+ *
+ * Licensed under the GPL-2
+ */
+
+#ifndef _BFIN_NMI_H_
+#define _BFIN_NMI_H_
+
+#include <linux/nmi.h>
+
+#endif

arch/blackfin/include/asm/smp.h

@@ -22,6 +22,7 @@ extern char coreb_trampoline_start, coreb_trampoline_end;
 struct corelock_slot {
 	int lock;
 };
+extern struct corelock_slot corelock;
 
 void smp_icache_flush_range_others(unsigned long start,
 				   unsigned long end);

arch/blackfin/kernel/Makefile

@@ -25,6 +25,7 @@ obj-$(CONFIG_CPLB_INFO)              += cplbinfo.o
 obj-$(CONFIG_MODULES)                += module.o
 obj-$(CONFIG_KGDB)                   += kgdb.o
 obj-$(CONFIG_KGDB_TESTS)             += kgdb_test.o
+obj-$(CONFIG_NMI_WATCHDOG)           += nmi.o
 obj-$(CONFIG_EARLY_PRINTK)           += early_printk.o
 obj-$(CONFIG_EARLY_PRINTK)           += shadow_console.o
 obj-$(CONFIG_STACKTRACE)             += stacktrace.o

arch/blackfin/kernel/nmi.c

@@ -0,0 +1,313 @@
+/*
+ * Blackfin nmi_watchdog Driver
+ *
+ * Originally based on bfin_wdt.c
+ * Copyright 2010-2010 Analog Devices Inc.
+ *		Graff Yang <graf.yang@analog.com>
+ *
+ * Enter bugs at http://blackfin.uclinux.org/
+ *
+ * Licensed under the GPL-2 or later.
+ */
+
+#include <linux/bitops.h>
+#include <linux/hardirq.h>
+#include <linux/sysdev.h>
+#include <linux/pm.h>
+#include <linux/nmi.h>
+#include <linux/smp.h>
+#include <linux/timer.h>
+#include <asm/blackfin.h>
+#include <asm/atomic.h>
+#include <asm/cacheflush.h>
+
+/* Bit in WDOG_CTL that indicates watchdog has expired (WDR0) */
+#define WDOG_EXPIRED 0x8000
+
+/* Masks for WDEV field in WDOG_CTL register */
+#define ICTL_RESET   0x0
+#define ICTL_NMI     0x2
+#define ICTL_GPI     0x4
+#define ICTL_NONE    0x6
+#define ICTL_MASK    0x6
+
+/* Masks for WDEN field in WDOG_CTL register */
+#define WDEN_MASK    0x0FF0
+#define WDEN_ENABLE  0x0000
+#define WDEN_DISABLE 0x0AD0
+
+#define DRV_NAME "nmi-wdt"
+
+#define NMI_WDT_TIMEOUT 5          /* 5 seconds */
+#define NMI_CHECK_TIMEOUT (4 * HZ) /* 4 seconds in jiffies */
+static int nmi_wdt_cpu = 1;
+
+static unsigned int timeout = NMI_WDT_TIMEOUT;
+static int nmi_active;
+
+static unsigned short wdoga_ctl;
+static unsigned int wdoga_cnt;
+static struct corelock_slot saved_corelock;
+static atomic_t nmi_touched[NR_CPUS];
+static struct timer_list ntimer;
+
+enum {
+	COREA_ENTER_NMI = 0,
+	COREA_EXIT_NMI,
+	COREB_EXIT_NMI,
+
+	NMI_EVENT_NR,
+};
+static unsigned long nmi_event __attribute__ ((__section__(".l2.bss")));
+
+/* we are in nmi, non-atomic bit ops is safe */
+static inline void set_nmi_event(int event)
+{
+	__set_bit(event, &nmi_event);
+}
+
+static inline void wait_nmi_event(int event)
+{
+	while (!test_bit(event, &nmi_event))
+		barrier();
+	__clear_bit(event, &nmi_event);
+}
+
+static inline void send_corea_nmi(void)
+{
+	wdoga_ctl = bfin_read_WDOGA_CTL();
+	wdoga_cnt = bfin_read_WDOGA_CNT();
+
+	bfin_write_WDOGA_CTL(WDEN_DISABLE);
+	bfin_write_WDOGA_CNT(0);
+	bfin_write_WDOGA_CTL(WDEN_ENABLE | ICTL_NMI);
+}
+
+static inline void restore_corea_nmi(void)
+{
+	bfin_write_WDOGA_CTL(WDEN_DISABLE);
+	bfin_write_WDOGA_CTL(WDOG_EXPIRED | WDEN_DISABLE | ICTL_NONE);
+
+	bfin_write_WDOGA_CNT(wdoga_cnt);
+	bfin_write_WDOGA_CTL(wdoga_ctl);
+}
+
+static inline void save_corelock(void)
+{
+	saved_corelock = corelock;
+	corelock.lock = 0;
+}
+
+static inline void restore_corelock(void)
+{
+	corelock = saved_corelock;
+}
+
+
+static inline void nmi_wdt_keepalive(void)
+{
+	bfin_write_WDOGB_STAT(0);
+}
+
+static inline void nmi_wdt_stop(void)
+{
+	bfin_write_WDOGB_CTL(WDEN_DISABLE);
+}
+
+/* before calling this function, you must stop the WDT */
+static inline void nmi_wdt_clear(void)
+{
+	/* clear TRO bit, disable event generation */
+	bfin_write_WDOGB_CTL(WDOG_EXPIRED | WDEN_DISABLE | ICTL_NONE);
+}
+
+static inline void nmi_wdt_start(void)
+{
+	bfin_write_WDOGB_CTL(WDEN_ENABLE | ICTL_NMI);
+}
+
+static inline int nmi_wdt_running(void)
+{
+	return ((bfin_read_WDOGB_CTL() & WDEN_MASK) != WDEN_DISABLE);
+}
+
+static inline int nmi_wdt_set_timeout(unsigned long t)
+{
+	u32 cnt, max_t, sclk;
+	int run;
+
+	sclk = get_sclk();
+	max_t = -1 / sclk;
+	cnt = t * sclk;
+	if (t > max_t) {
+		pr_warning("NMI: timeout value is too large\n");
+		return -EINVAL;
+	}
+
+	run = nmi_wdt_running();
+	nmi_wdt_stop();
+	bfin_write_WDOGB_CNT(cnt);
+	if (run)
+		nmi_wdt_start();
+
+	timeout = t;
+
+	return 0;
+}
+
+int check_nmi_wdt_touched(void)
+{
+	unsigned int this_cpu = smp_processor_id();
+	unsigned int cpu;
+
+	cpumask_t mask = cpu_online_map;
+
+	if (!atomic_read(&nmi_touched[this_cpu]))
+		return 0;
+
+	atomic_set(&nmi_touched[this_cpu], 0);
+
+	cpu_clear(this_cpu, mask);
+	for_each_cpu_mask(cpu, mask) {
+		invalidate_dcache_range((unsigned long)(&nmi_touched[cpu]),
+				(unsigned long)(&nmi_touched[cpu]));
+		if (!atomic_read(&nmi_touched[cpu]))
+			return 0;
+		atomic_set(&nmi_touched[cpu], 0);
+	}
+
+	return 1;
+}
+
+static void nmi_wdt_timer(unsigned long data)
+{
+	if (check_nmi_wdt_touched())
+		nmi_wdt_keepalive();
+
+	mod_timer(&ntimer, jiffies + NMI_CHECK_TIMEOUT);
+}
+
+static int __init init_nmi_wdt(void)
+{
+	nmi_wdt_set_timeout(timeout);
+	nmi_wdt_start();
+	nmi_active = true;
+
+	init_timer(&ntimer);
+	ntimer.function = nmi_wdt_timer;
+	ntimer.expires = jiffies + NMI_CHECK_TIMEOUT;
+	add_timer(&ntimer);
+
+	pr_info("nmi_wdt: initialized: timeout=%d sec\n", timeout);
+	return 0;
+}
+device_initcall(init_nmi_wdt);
+
+void touch_nmi_watchdog(void)
+{
+	atomic_set(&nmi_touched[smp_processor_id()], 1);
+}
+
+/* Suspend/resume support */
+#ifdef CONFIG_PM
+static int nmi_wdt_suspend(struct sys_device *dev, pm_message_t state)
+{
+	nmi_wdt_stop();
+	return 0;
+}
+
+static int nmi_wdt_resume(struct sys_device *dev)
+{
+	if (nmi_active)
+		nmi_wdt_start();
+	return 0;
+}
+
+static struct sysdev_class nmi_sysclass = {
+	.name		= DRV_NAME,
+	.resume		= nmi_wdt_resume,
+	.suspend	= nmi_wdt_suspend,
+};
+
+static struct sys_device device_nmi_wdt = {
+	.id	= 0,
+	.cls	= &nmi_sysclass,
+};
+
+static int __init init_nmi_wdt_sysfs(void)
+{
+	int error;
+
+	if (!nmi_active)
+		return 0;
+
+	error = sysdev_class_register(&nmi_sysclass);
+	if (!error)
+		error = sysdev_register(&device_nmi_wdt);
+	return error;
+}
+late_initcall(init_nmi_wdt_sysfs);
+
+#endif	/* CONFIG_PM */
+
+
+asmlinkage notrace void do_nmi(struct pt_regs *fp)
+{
+	unsigned int cpu = smp_processor_id();
+	nmi_enter();
+
+	cpu_pda[cpu].__nmi_count += 1;
+
+	if (cpu == nmi_wdt_cpu) {
+		/* CoreB goes here first */
+
+		/* reload the WDOG_STAT */
+		nmi_wdt_keepalive();
+
+		/* clear nmi interrupt for CoreB */
+		nmi_wdt_stop();
+		nmi_wdt_clear();
+
+		/* trigger NMI interrupt of CoreA */
+		send_corea_nmi();
+
+		/* waiting CoreB to enter NMI */
+		wait_nmi_event(COREA_ENTER_NMI);
+
+		/* recover WDOGA's settings */
+		restore_corea_nmi();
+
+		save_corelock();
+
+		/* corelock is save/cleared, CoreA is dummping messages */
+
+		wait_nmi_event(COREA_EXIT_NMI);
+	} else {
+		/* OK, CoreA entered NMI */
+		set_nmi_event(COREA_ENTER_NMI);
+	}
+
+	pr_emerg("\nNMI Watchdog detected LOCKUP, dump for CPU %d\n", cpu);
+	dump_bfin_process(fp);
+	dump_bfin_mem(fp);
+	show_regs(fp);
+	dump_bfin_trace_buffer();
+	show_stack(current, (unsigned long *)fp);
+
+	if (cpu == nmi_wdt_cpu) {
+		pr_emerg("This fault is not recoverable, sorry!\n");
+
+		/* CoreA dump finished, restore the corelock */
+		restore_corelock();
+
+		set_nmi_event(COREB_EXIT_NMI);
+	} else {
+		/* CoreB dump finished, notice the CoreA we are done */
+		set_nmi_event(COREA_EXIT_NMI);
+
+		/* synchronize with CoreA */
+		wait_nmi_event(COREB_EXIT_NMI);
+	}
+
+	nmi_exit();
+}

arch/blackfin/kernel/time-ts.c

@@ -21,6 +21,7 @@
 #include <asm/blackfin.h>
 #include <asm/time.h>
 #include <asm/gptimers.h>
+#include <asm/nmi.h>
 
 /* Accelerators for sched_clock()
  * convert from cycles(64bits) => nanoseconds (64bits)
@@ -309,6 +310,9 @@ irqreturn_t bfin_coretmr_interrupt(int irq, void *dev_id)
 
 	smp_mb();
 	evt->event_handler(evt);
+
+	touch_nmi_watchdog();
+
 	return IRQ_HANDLED;
 }