arch/x86/kernel/crash.c

   1 /*
   2  * Architecture specific (i386/x86_64) functions for kexec based crash dumps.
   3  *
   4  * Created by: Hariprasad Nellitheertha (hari@in.ibm.com)
   5  *
   6  * Copyright (C) IBM Corporation, 2004. All rights reserved.
   7  *
   8  */
   9
  10 #include <linux/init.h>
  11 #include <linux/types.h>
  12 #include <linux/kernel.h>
  13 #include <linux/smp.h>
  14 #include <linux/reboot.h>
  15 #include <linux/kexec.h>
  16 #include <linux/delay.h>
  17 #include <linux/elf.h>
  18 #include <linux/elfcore.h>
  19
  20 #include <asm/processor.h>
  21 #include <asm/hardirq.h>
  22 #include <asm/nmi.h>
  23 #include <asm/hw_irq.h>
  24 #include <asm/apic.h>
  25 #include <asm/hpet.h>
  26 #include <linux/kdebug.h>
  27 #include <asm/smp.h>
  28
  29 #ifdef CONFIG_X86_32
  30 #include <mach_ipi.h>
  31 #else
  32 #include <asm/mach_apic.h>
  33 #endif
  34
  35 /* This keeps a track of which one is crashing cpu. */
  36 static int crashing_cpu;
  37
  38 #if defined(CONFIG_SMP) && defined(CONFIG_X86_LOCAL_APIC)
  39 static atomic_t waiting_for_crash_ipi;
  40
  41 static int crash_nmi_callback(struct notifier_block *self,
  42                         unsigned long val, void *data)
  43 {
  44         struct pt_regs *regs;
  45 #ifdef CONFIG_X86_32
  46         struct pt_regs fixed_regs;
  47 #endif
  48         int cpu;
  49
  50         if (val != DIE_NMI_IPI)
  51                 return NOTIFY_OK;
  52
  53         regs = ((struct die_args *)data)->regs;
  54         cpu = raw_smp_processor_id();
  55
  56         /* Don't do anything if this handler is invoked on crashing cpu.
  57          * Otherwise, system will completely hang. Crashing cpu can get
  58          * an NMI if system was initially booted with nmi_watchdog parameter.
  59          */
  60         if (cpu == crashing_cpu)
  61                 return NOTIFY_STOP;
  62         local_irq_disable();
  63
  64 #ifdef CONFIG_X86_32
  65         if (!user_mode_vm(regs)) {
  66                 crash_fixup_ss_esp(&fixed_regs, regs);
  67                 regs = &fixed_regs;
  68         }
  69 #endif
  70         crash_save_cpu(regs, cpu);
  71         disable_local_APIC();
  72         atomic_dec(&waiting_for_crash_ipi);
  73         /* Assume hlt works */
  74         halt();
  75         for (;;)
  76                 cpu_relax();
  77
  78         return 1;
  79 }
  80
  81 static void smp_send_nmi_allbutself(void)
  82 {
  83         cpumask_t mask = cpu_online_map;
  84         cpu_clear(safe_smp_processor_id(), mask);
  85         if (!cpus_empty(mask))
  86                 send_IPI_mask(mask, NMI_VECTOR);
  87 }
  88
  89 static struct notifier_block crash_nmi_nb = {
  90         .notifier_call = crash_nmi_callback,
  91 };
  92
  93 static void nmi_shootdown_cpus(void)
  94 {
  95         unsigned long msecs;
  96
  97         atomic_set(&waiting_for_crash_ipi, num_online_cpus() - 1);
  98         /* Would it be better to replace the trap vector here? */
  99         if (register_die_notifier(&crash_nmi_nb))
 100                 return;         /* return what? */
 101         /* Ensure the new callback function is set before sending
 102          * out the NMI
 103          */
 104         wmb();
 105
 106         smp_send_nmi_allbutself();
 107
 108         msecs = 1000; /* Wait at most a second for the other cpus to stop */
 109         while ((atomic_read(&waiting_for_crash_ipi) > 0) && msecs) {
 110                 mdelay(1);
 111                 msecs--;
 112         }
 113
 114         /* Leave the nmi callback set */
 115         disable_local_APIC();
 116 }
 117 #else
 118 static void nmi_shootdown_cpus(void)
 119 {
 120         /* There are no cpus to shootdown */
 121 }
 122 #endif
 123
 124 void machine_crash_shutdown(struct pt_regs *regs)
 125 {
 126         /* This function is only called after the system
 127          * has panicked or is otherwise in a critical state.
 128          * The minimum amount of code to allow a kexec'd kernel
 129          * to run successfully needs to happen here.
 130          *
 131          * In practice this means shooting down the other cpus in
 132          * an SMP system.
 133          */
 134         /* The kernel is broken so disable interrupts */
 135         local_irq_disable();
 136
 137         /* Make a note of crashing cpu. Will be used in NMI callback.*/
 138         crashing_cpu = safe_smp_processor_id();
 139         nmi_shootdown_cpus();
 140         lapic_shutdown();
 141 #if defined(CONFIG_X86_IO_APIC)
 142         disable_IO_APIC();
 143 #endif
 144 #ifdef CONFIG_HPET_TIMER
 145         hpet_disable();
 146 #endif
 147         crash_save_cpu(regs, safe_smp_processor_id());
 148 }