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