arch/x86_64/kernel/crash.c

   1 /*
   2  * Architecture specific (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/irq.h>
  15 #include <linux/reboot.h>
  16 #include <linux/kexec.h>
  17 #include <linux/delay.h>
  18 #include <linux/elf.h>
  19 #include <linux/elfcore.h>
  20
  21 #include <asm/processor.h>
  22 #include <asm/hardirq.h>
  23 #include <asm/nmi.h>
  24 #include <asm/hw_irq.h>
  25 #include <asm/mach_apic.h>
  26
  27 /* This keeps a track of which one is crashing cpu. */
  28 static int crashing_cpu;
  29
  30 static u32 *append_elf_note(u32 *buf, char *name, unsigned type,
  31                                                 void *data, size_t data_len)
  32 {
  33         struct elf_note note;
  34
  35         note.n_namesz = strlen(name) + 1;
  36         note.n_descsz = data_len;
  37         note.n_type   = type;
  38         memcpy(buf, &note, sizeof(note));
  39         buf += (sizeof(note) +3)/4;
  40         memcpy(buf, name, note.n_namesz);
  41         buf += (note.n_namesz + 3)/4;
  42         memcpy(buf, data, note.n_descsz);
  43         buf += (note.n_descsz + 3)/4;
  44
  45         return buf;
  46 }
  47
  48 static void final_note(u32 *buf)
  49 {
  50         struct elf_note note;
  51
  52         note.n_namesz = 0;
  53         note.n_descsz = 0;
  54         note.n_type   = 0;
  55         memcpy(buf, &note, sizeof(note));
  56 }
  57
  58 static void crash_save_this_cpu(struct pt_regs *regs, int cpu)
  59 {
  60         struct elf_prstatus prstatus;
  61         u32 *buf;
  62
  63         if ((cpu < 0) || (cpu >= NR_CPUS))
  64                 return;
  65
  66         /* Using ELF notes here is opportunistic.
  67          * I need a well defined structure format
  68          * for the data I pass, and I need tags
  69          * on the data to indicate what information I have
  70          * squirrelled away.  ELF notes happen to provide
  71          * all of that that no need to invent something new.
  72          */
  73
  74         buf = (u32*)per_cpu_ptr(crash_notes, cpu);
  75
  76         if (!buf)
  77                 return;
  78
  79         memset(&prstatus, 0, sizeof(prstatus));
  80         prstatus.pr_pid = current->pid;
  81         elf_core_copy_regs(&prstatus.pr_reg, regs);
  82         buf = append_elf_note(buf, "CORE", NT_PRSTATUS, &prstatus,
  83                                         sizeof(prstatus));
  84         final_note(buf);
  85 }
  86
  87 static void crash_save_self(struct pt_regs *regs)
  88 {
  89         int cpu;
  90
  91         cpu = smp_processor_id();
  92         crash_save_this_cpu(regs, cpu);
  93 }
  94
  95 #ifdef CONFIG_SMP
  96 static atomic_t waiting_for_crash_ipi;
  97
  98 static int crash_nmi_callback(struct pt_regs *regs, int cpu)
  99 {
 100         /*
 101          * Don't do anything if this handler is invoked on crashing cpu.
 102          * Otherwise, system will completely hang. Crashing cpu can get
 103          * an NMI if system was initially booted with nmi_watchdog parameter.
 104          */
 105         if (cpu == crashing_cpu)
 106                 return 1;
 107         local_irq_disable();
 108
 109         crash_save_this_cpu(regs, cpu);
 110         disable_local_APIC();
 111         atomic_dec(&waiting_for_crash_ipi);
 112         /* Assume hlt works */
 113         for(;;)
 114                 asm("hlt");
 115
 116         return 1;
 117 }
 118
 119 static void smp_send_nmi_allbutself(void)
 120 {
 121         send_IPI_allbutself(APIC_DM_NMI);
 122 }
 123
 124 /*
 125  * This code is a best effort heuristic to get the
 126  * other cpus to stop executing. So races with
 127  * cpu hotplug shouldn't matter.
 128  */
 129
 130 static void nmi_shootdown_cpus(void)
 131 {
 132         unsigned long msecs;
 133
 134         atomic_set(&waiting_for_crash_ipi, num_online_cpus() - 1);
 135         set_nmi_callback(crash_nmi_callback);
 136
 137         /*
 138          * Ensure the new callback function is set before sending
 139          * out the NMI
 140          */
 141         wmb();
 142
 143         smp_send_nmi_allbutself();
 144
 145         msecs = 1000; /* Wait at most a second for the other cpus to stop */
 146         while ((atomic_read(&waiting_for_crash_ipi) > 0) && msecs) {
 147                 mdelay(1);
 148                 msecs--;
 149         }
 150         /* Leave the nmi callback set */
 151         disable_local_APIC();
 152 }
 153 #else
 154 static void nmi_shootdown_cpus(void)
 155 {
 156         /* There are no cpus to shootdown */
 157 }
 158 #endif
 159
 160 void machine_crash_shutdown(struct pt_regs *regs)
 161 {
 162         /*
 163          * This function is only called after the system
 164          * has paniced or is otherwise in a critical state.
 165          * The minimum amount of code to allow a kexec'd kernel
 166          * to run successfully needs to happen here.
 167          *
 168          * In practice this means shooting down the other cpus in
 169          * an SMP system.
 170          */
 171         /* The kernel is broken so disable interrupts */
 172         local_irq_disable();
 173
 174         /* Make a note of crashing cpu. Will be used in NMI callback.*/
 175         crashing_cpu = smp_processor_id();
 176         nmi_shootdown_cpus();
 177
 178         if(cpu_has_apic)
 179                  disable_local_APIC();
 180
 181 #if defined(CONFIG_X86_IO_APIC)
 182         disable_IO_APIC();
 183 #endif
 184
 185         crash_save_self(regs);
 186 }