arch/blackfin/kernel/nmi.c

   1 /*
   2  * Blackfin nmi_watchdog Driver
   3  *
   4  * Originally based on bfin_wdt.c
   5  * Copyright 2010-2010 Analog Devices Inc.
   6  *              Graff Yang <graf.yang@analog.com>
   7  *
   8  * Enter bugs at http://blackfin.uclinux.org/
   9  *
  10  * Licensed under the GPL-2 or later.
  11  */
  12
  13 #include <linux/bitops.h>
  14 #include <linux/hardirq.h>
  15 #include <linux/sysdev.h>
  16 #include <linux/pm.h>
  17 #include <linux/nmi.h>
  18 #include <linux/smp.h>
  19 #include <linux/timer.h>
  20 #include <asm/blackfin.h>
  21 #include <asm/atomic.h>
  22 #include <asm/cacheflush.h>
  23 #include <asm/bfin_watchdog.h>
  24
  25 #define DRV_NAME "nmi-wdt"
  26
  27 #define NMI_WDT_TIMEOUT 5          /* 5 seconds */
  28 #define NMI_CHECK_TIMEOUT (4 * HZ) /* 4 seconds in jiffies */
  29 static int nmi_wdt_cpu = 1;
  30
  31 static unsigned int timeout = NMI_WDT_TIMEOUT;
  32 static int nmi_active;
  33
  34 static unsigned short wdoga_ctl;
  35 static unsigned int wdoga_cnt;
  36 static struct corelock_slot saved_corelock;
  37 static atomic_t nmi_touched[NR_CPUS];
  38 static struct timer_list ntimer;
  39
  40 enum {
  41         COREA_ENTER_NMI = 0,
  42         COREA_EXIT_NMI,
  43         COREB_EXIT_NMI,
  44
  45         NMI_EVENT_NR,
  46 };
  47 static unsigned long nmi_event __attribute__ ((__section__(".l2.bss")));
  48
  49 /* we are in nmi, non-atomic bit ops is safe */
  50 static inline void set_nmi_event(int event)
  51 {
  52         __set_bit(event, &nmi_event);
  53 }
  54
  55 static inline void wait_nmi_event(int event)
  56 {
  57         while (!test_bit(event, &nmi_event))
  58                 barrier();
  59         __clear_bit(event, &nmi_event);
  60 }
  61
  62 static inline void send_corea_nmi(void)
  63 {
  64         wdoga_ctl = bfin_read_WDOGA_CTL();
  65         wdoga_cnt = bfin_read_WDOGA_CNT();
  66
  67         bfin_write_WDOGA_CTL(WDEN_DISABLE);
  68         bfin_write_WDOGA_CNT(0);
  69         bfin_write_WDOGA_CTL(WDEN_ENABLE | ICTL_NMI);
  70 }
  71
  72 static inline void restore_corea_nmi(void)
  73 {
  74         bfin_write_WDOGA_CTL(WDEN_DISABLE);
  75         bfin_write_WDOGA_CTL(WDOG_EXPIRED | WDEN_DISABLE | ICTL_NONE);
  76
  77         bfin_write_WDOGA_CNT(wdoga_cnt);
  78         bfin_write_WDOGA_CTL(wdoga_ctl);
  79 }
  80
  81 static inline void save_corelock(void)
  82 {
  83         saved_corelock = corelock;
  84         corelock.lock = 0;
  85 }
  86
  87 static inline void restore_corelock(void)
  88 {
  89         corelock = saved_corelock;
  90 }
  91
  92
  93 static inline void nmi_wdt_keepalive(void)
  94 {
  95         bfin_write_WDOGB_STAT(0);
  96 }
  97
  98 static inline void nmi_wdt_stop(void)
  99 {
 100         bfin_write_WDOGB_CTL(WDEN_DISABLE);
 101 }
 102
 103 /* before calling this function, you must stop the WDT */
 104 static inline void nmi_wdt_clear(void)
 105 {
 106         /* clear TRO bit, disable event generation */
 107         bfin_write_WDOGB_CTL(WDOG_EXPIRED | WDEN_DISABLE | ICTL_NONE);
 108 }
 109
 110 static inline void nmi_wdt_start(void)
 111 {
 112         bfin_write_WDOGB_CTL(WDEN_ENABLE | ICTL_NMI);
 113 }
 114
 115 static inline int nmi_wdt_running(void)
 116 {
 117         return ((bfin_read_WDOGB_CTL() & WDEN_MASK) != WDEN_DISABLE);
 118 }
 119
 120 static inline int nmi_wdt_set_timeout(unsigned long t)
 121 {
 122         u32 cnt, max_t, sclk;
 123         int run;
 124
 125         sclk = get_sclk();
 126         max_t = -1 / sclk;
 127         cnt = t * sclk;
 128         if (t > max_t) {
 129                 pr_warning("NMI: timeout value is too large\n");
 130                 return -EINVAL;
 131         }
 132
 133         run = nmi_wdt_running();
 134         nmi_wdt_stop();
 135         bfin_write_WDOGB_CNT(cnt);
 136         if (run)
 137                 nmi_wdt_start();
 138
 139         timeout = t;
 140
 141         return 0;
 142 }
 143
 144 int check_nmi_wdt_touched(void)
 145 {
 146         unsigned int this_cpu = smp_processor_id();
 147         unsigned int cpu;
 148
 149         cpumask_t mask = cpu_online_map;
 150
 151         if (!atomic_read(&nmi_touched[this_cpu]))
 152                 return 0;
 153
 154         atomic_set(&nmi_touched[this_cpu], 0);
 155
 156         cpu_clear(this_cpu, mask);
 157         for_each_cpu_mask(cpu, mask) {
 158                 invalidate_dcache_range((unsigned long)(&nmi_touched[cpu]),
 159                                 (unsigned long)(&nmi_touched[cpu]));
 160                 if (!atomic_read(&nmi_touched[cpu]))
 161                         return 0;
 162                 atomic_set(&nmi_touched[cpu], 0);
 163         }
 164
 165         return 1;
 166 }
 167
 168 static void nmi_wdt_timer(unsigned long data)
 169 {
 170         if (check_nmi_wdt_touched())
 171                 nmi_wdt_keepalive();
 172
 173         mod_timer(&ntimer, jiffies + NMI_CHECK_TIMEOUT);
 174 }
 175
 176 static int __init init_nmi_wdt(void)
 177 {
 178         nmi_wdt_set_timeout(timeout);
 179         nmi_wdt_start();
 180         nmi_active = true;
 181
 182         init_timer(&ntimer);
 183         ntimer.function = nmi_wdt_timer;
 184         ntimer.expires = jiffies + NMI_CHECK_TIMEOUT;
 185         add_timer(&ntimer);
 186
 187         pr_info("nmi_wdt: initialized: timeout=%d sec\n", timeout);
 188         return 0;
 189 }
 190 device_initcall(init_nmi_wdt);
 191
 192 void touch_nmi_watchdog(void)
 193 {
 194         atomic_set(&nmi_touched[smp_processor_id()], 1);
 195 }
 196
 197 /* Suspend/resume support */
 198 #ifdef CONFIG_PM
 199 static int nmi_wdt_suspend(struct sys_device *dev, pm_message_t state)
 200 {
 201         nmi_wdt_stop();
 202         return 0;
 203 }
 204
 205 static int nmi_wdt_resume(struct sys_device *dev)
 206 {
 207         if (nmi_active)
 208                 nmi_wdt_start();
 209         return 0;
 210 }
 211
 212 static struct sysdev_class nmi_sysclass = {
 213         .name           = DRV_NAME,
 214         .resume         = nmi_wdt_resume,
 215         .suspend        = nmi_wdt_suspend,
 216 };
 217
 218 static struct sys_device device_nmi_wdt = {
 219         .id     = 0,
 220         .cls    = &nmi_sysclass,
 221 };
 222
 223 static int __init init_nmi_wdt_sysfs(void)
 224 {
 225         int error;
 226
 227         if (!nmi_active)
 228                 return 0;
 229
 230         error = sysdev_class_register(&nmi_sysclass);
 231         if (!error)
 232                 error = sysdev_register(&device_nmi_wdt);
 233         return error;
 234 }
 235 late_initcall(init_nmi_wdt_sysfs);
 236
 237 #endif  /* CONFIG_PM */
 238
 239
 240 asmlinkage notrace void do_nmi(struct pt_regs *fp)
 241 {
 242         unsigned int cpu = smp_processor_id();
 243         nmi_enter();
 244
 245         cpu_pda[cpu].__nmi_count += 1;
 246
 247         if (cpu == nmi_wdt_cpu) {
 248                 /* CoreB goes here first */
 249
 250                 /* reload the WDOG_STAT */
 251                 nmi_wdt_keepalive();
 252
 253                 /* clear nmi interrupt for CoreB */
 254                 nmi_wdt_stop();
 255                 nmi_wdt_clear();
 256
 257                 /* trigger NMI interrupt of CoreA */
 258                 send_corea_nmi();
 259
 260                 /* waiting CoreB to enter NMI */
 261                 wait_nmi_event(COREA_ENTER_NMI);
 262
 263                 /* recover WDOGA's settings */
 264                 restore_corea_nmi();
 265
 266                 save_corelock();
 267
 268                 /* corelock is save/cleared, CoreA is dummping messages */
 269
 270                 wait_nmi_event(COREA_EXIT_NMI);
 271         } else {
 272                 /* OK, CoreA entered NMI */
 273                 set_nmi_event(COREA_ENTER_NMI);
 274         }
 275
 276         pr_emerg("\nNMI Watchdog detected LOCKUP, dump for CPU %d\n", cpu);
 277         dump_bfin_process(fp);
 278         dump_bfin_mem(fp);
 279         show_regs(fp);
 280         dump_bfin_trace_buffer();
 281         show_stack(current, (unsigned long *)fp);
 282
 283         if (cpu == nmi_wdt_cpu) {
 284                 pr_emerg("This fault is not recoverable, sorry!\n");
 285
 286                 /* CoreA dump finished, restore the corelock */
 287                 restore_corelock();
 288
 289                 set_nmi_event(COREB_EXIT_NMI);
 290         } else {
 291                 /* CoreB dump finished, notice the CoreA we are done */
 292                 set_nmi_event(COREA_EXIT_NMI);
 293
 294                 /* synchronize with CoreA */
 295                 wait_nmi_event(COREB_EXIT_NMI);
 296         }
 297
 298         nmi_exit();
 299 }