| 1 | |
| 2 |
|
| 3 | #include <linux/cpu.h>
|
| 4 | #include <linux/cpumask.h>
|
| 5 | #include <linux/kernel.h>
|
| 6 | #include <linux/nmi.h>
|
| 7 | #include <linux/percpu-defs.h>
|
| 8 |
|
| 9 | static cpumask_t __read_mostly watchdog_cpus;
|
| 10 |
|
| 11 | static unsigned int watchdog_next_cpu(unsigned int cpu)
|
| 12 | {
|
| 13 | unsigned int next_cpu;
|
| 14 |
|
| 15 | next_cpu = cpumask_next_wrap(cpu, &watchdog_cpus);
|
| 16 | if (next_cpu == cpu)
|
| 17 | return nr_cpu_ids;
|
| 18 |
|
| 19 | return next_cpu;
|
| 20 | }
|
| 21 |
|
| 22 | int __init watchdog_hardlockup_probe(void)
|
| 23 | {
|
| 24 | watchdog_hardlockup_miss_thresh = 3;
|
| 25 | return 0;
|
| 26 | }
|
| 27 |
|
| 28 | void watchdog_hardlockup_enable(unsigned int cpu)
|
| 29 | {
|
| 30 | unsigned int next_cpu;
|
| 31 |
|
| 32 | |
| 33 | * The new CPU will be marked online before the hrtimer interrupt
|
| 34 | * gets a chance to run on it. If another CPU tests for a
|
| 35 | * hardlockup on the new CPU before it has run its the hrtimer
|
| 36 | * interrupt, it will get a false positive. Touch the watchdog on
|
| 37 | * the new CPU to delay the check for at least 3 sampling periods
|
| 38 | * to guarantee one hrtimer has run on the new CPU.
|
| 39 |
|
| 40 | watchdog_hardlockup_touch_cpu(cpu);
|
| 41 |
|
| 42 | |
| 43 | * We are going to check the next CPU. Our watchdog_hrtimer
|
| 44 | * need not be zero if the CPU has already been online earlier.
|
| 45 | * Touch the watchdog on the next CPU to avoid false positive
|
| 46 | * if we try to check it in less then 3 interrupts.
|
| 47 |
|
| 48 | next_cpu = watchdog_next_cpu(cpu);
|
| 49 | if (next_cpu < nr_cpu_ids)
|
| 50 | watchdog_hardlockup_touch_cpu(next_cpu);
|
| 51 |
|
| 52 | |
| 53 | * Makes sure that watchdog is touched on this CPU before
|
| 54 | * other CPUs could see it in watchdog_cpus. The counter
|
| 55 | * part is in watchdog_buddy_check_hardlockup().
|
| 56 |
|
| 57 | smp_wmb();
|
| 58 |
|
| 59 | cpumask_set_cpu(cpu, &watchdog_cpus);
|
| 60 | }
|
| 61 |
|
| 62 | void watchdog_hardlockup_disable(unsigned int cpu)
|
| 63 | {
|
| 64 | unsigned int next_cpu = watchdog_next_cpu(cpu);
|
| 65 |
|
| 66 | |
| 67 | * Offlining this CPU will cause the CPU before this one to start
|
| 68 | * checking the one after this one. If this CPU just finished checking
|
| 69 | * the next CPU and updating hrtimer_interrupts_saved, and then the
|
| 70 | * previous CPU checks it within one sample period, it will trigger a
|
| 71 | * false positive. Touch the watchdog on the next CPU to prevent it.
|
| 72 |
|
| 73 | if (next_cpu < nr_cpu_ids)
|
| 74 | watchdog_hardlockup_touch_cpu(next_cpu);
|
| 75 |
|
| 76 | |
| 77 | * Makes sure that watchdog is touched on the next CPU before
|
| 78 | * this CPU disappear in watchdog_cpus. The counter part is in
|
| 79 | * watchdog_buddy_check_hardlockup().
|
| 80 |
|
| 81 | smp_wmb();
|
| 82 |
|
| 83 | cpumask_clear_cpu(cpu, &watchdog_cpus);
|
| 84 | }
|
| 85 |
|
| 86 | void watchdog_buddy_check_hardlockup(int hrtimer_interrupts)
|
| 87 | {
|
| 88 | unsigned int next_cpu;
|
| 89 |
|
| 90 |
|
| 91 | next_cpu = watchdog_next_cpu(smp_processor_id());
|
| 92 | if (next_cpu >= nr_cpu_ids)
|
| 93 | return;
|
| 94 |
|
| 95 | |
| 96 | * Make sure that the watchdog was touched on next CPU when
|
| 97 | * watchdog_next_cpu() returned another one because of
|
| 98 | * a change in watchdog_hardlockup_enable()/disable().
|
| 99 |
|
| 100 | smp_rmb();
|
| 101 |
|
| 102 | watchdog_hardlockup_check(next_cpu, NULL);
|
| 103 | }
|