arch/i386/kernel/crash.c

   1 /*
   2  * Architecture specific (i386) 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 <mach_ipi.h>
  26
  27
  28 /* This keeps a track of which one is crashing cpu. */
  29 static int crashing_cpu;
  30
  31 static u32 *append_elf_note(u32 *buf, char *name, unsigned type, void *data,
  32                                                                size_t data_len)
  33 {
  34         struct elf_note note;
  35
  36         note.n_namesz = strlen(name) + 1;
  37         note.n_descsz = data_len;
  38         note.n_type   = type;
  39         memcpy(buf, &note, sizeof(note));
  40         buf += (sizeof(note) +3)/4;
  41         memcpy(buf, name, note.n_namesz);
  42         buf += (note.n_namesz + 3)/4;
  43         memcpy(buf, data, note.n_descsz);
  44         buf += (note.n_descsz + 3)/4;
  45
  46         return buf;
  47 }
  48
  49 static void final_note(u32 *buf)
  50 {
  51         struct elf_note note;
  52
  53         note.n_namesz = 0;
  54         note.n_descsz = 0;
  55         note.n_type   = 0;
  56         memcpy(buf, &note, sizeof(note));
  57 }
  58
  59 static void crash_save_this_cpu(struct pt_regs *regs, int cpu)
  60 {
  61         struct elf_prstatus prstatus;
  62         u32 *buf;
  63
  64         if ((cpu < 0) || (cpu >= NR_CPUS))
  65                 return;
  66
  67         /* Using ELF notes here is opportunistic.
  68          * I need a well defined structure format
  69          * for the data I pass, and I need tags
  70          * on the data to indicate what information I have
  71          * squirrelled away.  ELF notes happen to provide
  72          * all of that that no need to invent something new.
  73          */
  74         buf = (u32*)per_cpu_ptr(crash_notes, cpu);
  75         if (!buf)
  76                 return;
  77         memset(&prstatus, 0, sizeof(prstatus));
  78         prstatus.pr_pid = current->pid;
  79         elf_core_copy_regs(&prstatus.pr_reg, regs);
  80         buf = append_elf_note(buf, "CORE", NT_PRSTATUS, &prstatus,
  81                                 sizeof(prstatus));
  82         final_note(buf);
  83 }
  84
  85 static void crash_save_self(struct pt_regs *regs)
  86 {
  87         int cpu;
  88
  89         cpu = smp_processor_id();
  90         crash_save_this_cpu(regs, cpu);
  91 }
  92
  93 #ifdef CONFIG_SMP
  94 static atomic_t waiting_for_crash_ipi;
  95
  96 static int crash_nmi_callback(struct pt_regs *regs, int cpu)
  97 {
  98         struct pt_regs fixed_regs;
  99
 100         /* Don't do anything if this handler is invoked on crashing cpu.
 101          * Otherwise, system will completely hang. Crashing cpu can get
 102          * an NMI if system was initially booted with nmi_watchdog parameter.
 103          */
 104         if (cpu == crashing_cpu)
 105                 return 1;
 106         local_irq_disable();
 107
 108         if (!user_mode_vm(regs)) {
 109                 crash_fixup_ss_esp(&fixed_regs, regs);
 110                 regs = &fixed_regs;
 111         }
 112         crash_save_this_cpu(regs, cpu);
 113         disable_local_APIC();
 114         atomic_dec(&waiting_for_crash_ipi);
 115         /* Assume hlt works */
 116         halt();
 117         for(;;);
 118
 119         return 1;
 120 }
 121
 122 /*
 123  * By using the NMI code instead of a vector we just sneak thru the
 124  * word generator coming out with just what we want.  AND it does
 125  * not matter if clustered_apic_mode is set or not.
 126  */
 127 static void smp_send_nmi_allbutself(void)
 128 {
 129         send_IPI_allbutself(APIC_DM_NMI);
 130 }
 131
 132 static void nmi_shootdown_cpus(void)
 133 {
 134         unsigned long msecs;
 135
 136         atomic_set(&waiting_for_crash_ipi, num_online_cpus() - 1);
 137         /* Would it be better to replace the trap vector here? */
 138         set_nmi_callback(crash_nmi_callback);
 139         /* Ensure the new callback function is set before sending
 140          * out the NMI
 141          */
 142         wmb();
 143
 144         smp_send_nmi_allbutself();
 145
 146         msecs = 1000; /* Wait at most a second for the other cpus to stop */
 147         while ((atomic_read(&waiting_for_crash_ipi) > 0) && msecs) {
 148                 mdelay(1);
 149                 msecs--;
 150         }
 151
 152         /* Leave the nmi callback set */
 153         disable_local_APIC();
 154 }
 155 #else
 156 static void nmi_shootdown_cpus(void)
 157 {
 158         /* There are no cpus to shootdown */
 159 }
 160 #endif
 161
 162 void machine_crash_shutdown(struct pt_regs *regs)
 163 {
 164         /* This function is only called after the system
 165          * has paniced or is otherwise in a critical state.
 166          * The minimum amount of code to allow a kexec'd kernel
 167          * to run successfully needs to happen here.
 168          *
 169          * In practice this means shooting down the other cpus in
 170          * an SMP system.
 171          */
 172         /* The kernel is broken so disable interrupts */
 173         local_irq_disable();
 174
 175         /* Make a note of crashing cpu. Will be used in NMI callback.*/
 176         crashing_cpu = smp_processor_id();
 177         nmi_shootdown_cpus();
 178         lapic_shutdown();
 179 #if defined(CONFIG_X86_IO_APIC)
 180         disable_IO_APIC();
 181 #endif
 182         crash_save_self(regs);
 183 }