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 note_buf_t crash_notes[NR_CPUS];
  29 /* This keeps a track of which one is crashing cpu. */
  30 static int crashing_cpu;
  31
  32 static u32 *append_elf_note(u32 *buf, char *name, unsigned type, void *data,
  33                                                                size_t data_len)
  34 {
  35         struct elf_note note;
  36
  37         note.n_namesz = strlen(name) + 1;
  38         note.n_descsz = data_len;
  39         note.n_type   = type;
  40         memcpy(buf, &note, sizeof(note));
  41         buf += (sizeof(note) +3)/4;
  42         memcpy(buf, name, note.n_namesz);
  43         buf += (note.n_namesz + 3)/4;
  44         memcpy(buf, data, note.n_descsz);
  45         buf += (note.n_descsz + 3)/4;
  46
  47         return buf;
  48 }
  49
  50 static void final_note(u32 *buf)
  51 {
  52         struct elf_note note;
  53
  54         note.n_namesz = 0;
  55         note.n_descsz = 0;
  56         note.n_type   = 0;
  57         memcpy(buf, &note, sizeof(note));
  58 }
  59
  60 static void crash_save_this_cpu(struct pt_regs *regs, int cpu)
  61 {
  62         struct elf_prstatus prstatus;
  63         u32 *buf;
  64
  65         if ((cpu < 0) || (cpu >= NR_CPUS))
  66                 return;
  67
  68         /* Using ELF notes here is opportunistic.
  69          * I need a well defined structure format
  70          * for the data I pass, and I need tags
  71          * on the data to indicate what information I have
  72          * squirrelled away.  ELF notes happen to provide
  73          * all of that that no need to invent something new.
  74          */
  75         buf = &crash_notes[cpu][0];
  76         memset(&prstatus, 0, sizeof(prstatus));
  77         prstatus.pr_pid = current->pid;
  78         elf_core_copy_regs(&prstatus.pr_reg, regs);
  79         buf = append_elf_note(buf, "CORE", NT_PRSTATUS, &prstatus,
  80                                 sizeof(prstatus));
  81         final_note(buf);
  82 }
  83
  84 static void crash_get_current_regs(struct pt_regs *regs)
  85 {
  86         __asm__ __volatile__("movl %%ebx,%0" : "=m"(regs->ebx));
  87         __asm__ __volatile__("movl %%ecx,%0" : "=m"(regs->ecx));
  88         __asm__ __volatile__("movl %%edx,%0" : "=m"(regs->edx));
  89         __asm__ __volatile__("movl %%esi,%0" : "=m"(regs->esi));
  90         __asm__ __volatile__("movl %%edi,%0" : "=m"(regs->edi));
  91         __asm__ __volatile__("movl %%ebp,%0" : "=m"(regs->ebp));
  92         __asm__ __volatile__("movl %%eax,%0" : "=m"(regs->eax));
  93         __asm__ __volatile__("movl %%esp,%0" : "=m"(regs->esp));
  94         __asm__ __volatile__("movw %%ss, %%ax;" :"=a"(regs->xss));
  95         __asm__ __volatile__("movw %%cs, %%ax;" :"=a"(regs->xcs));
  96         __asm__ __volatile__("movw %%ds, %%ax;" :"=a"(regs->xds));
  97         __asm__ __volatile__("movw %%es, %%ax;" :"=a"(regs->xes));
  98         __asm__ __volatile__("pushfl; popl %0" :"=m"(regs->eflags));
  99
 100         regs->eip = (unsigned long)current_text_addr();
 101 }
 102
 103 /* CPU does not save ss and esp on stack if execution is already
 104  * running in kernel mode at the time of NMI occurrence. This code
 105  * fixes it.
 106  */
 107 static void crash_setup_regs(struct pt_regs *newregs, struct pt_regs *oldregs)
 108 {
 109         memcpy(newregs, oldregs, sizeof(*newregs));
 110         newregs->esp = (unsigned long)&(oldregs->esp);
 111         __asm__ __volatile__("xorl %eax, %eax;");
 112         __asm__ __volatile__ ("movw %%ss, %%ax;" :"=a"(newregs->xss));
 113 }
 114
 115 /* We may have saved_regs from where the error came from
 116  * or it is NULL if via a direct panic().
 117  */
 118 static void crash_save_self(struct pt_regs *saved_regs)
 119 {
 120         struct pt_regs regs;
 121         int cpu;
 122
 123         cpu = smp_processor_id();
 124         if (saved_regs)
 125                 crash_setup_regs(&regs, saved_regs);
 126         else
 127                 crash_get_current_regs(&regs);
 128         crash_save_this_cpu(&regs, cpu);
 129 }
 130
 131 #ifdef CONFIG_SMP
 132 static atomic_t waiting_for_crash_ipi;
 133
 134 static int crash_nmi_callback(struct pt_regs *regs, int cpu)
 135 {
 136         struct pt_regs fixed_regs;
 137
 138         /* Don't do anything if this handler is invoked on crashing cpu.
 139          * Otherwise, system will completely hang. Crashing cpu can get
 140          * an NMI if system was initially booted with nmi_watchdog parameter.
 141          */
 142         if (cpu == crashing_cpu)
 143                 return 1;
 144         local_irq_disable();
 145
 146         if (!user_mode(regs)) {
 147                 crash_setup_regs(&fixed_regs, regs);
 148                 regs = &fixed_regs;
 149         }
 150         crash_save_this_cpu(regs, cpu);
 151         disable_local_APIC();
 152         atomic_dec(&waiting_for_crash_ipi);
 153         /* Assume hlt works */
 154         halt();
 155         for(;;);
 156
 157         return 1;
 158 }
 159
 160 /*
 161  * By using the NMI code instead of a vector we just sneak thru the
 162  * word generator coming out with just what we want.  AND it does
 163  * not matter if clustered_apic_mode is set or not.
 164  */
 165 static void smp_send_nmi_allbutself(void)
 166 {
 167         send_IPI_allbutself(APIC_DM_NMI);
 168 }
 169
 170 static void nmi_shootdown_cpus(void)
 171 {
 172         unsigned long msecs;
 173
 174         atomic_set(&waiting_for_crash_ipi, num_online_cpus() - 1);
 175         /* Would it be better to replace the trap vector here? */
 176         set_nmi_callback(crash_nmi_callback);
 177         /* Ensure the new callback function is set before sending
 178          * out the NMI
 179          */
 180         wmb();
 181
 182         smp_send_nmi_allbutself();
 183
 184         msecs = 1000; /* Wait at most a second for the other cpus to stop */
 185         while ((atomic_read(&waiting_for_crash_ipi) > 0) && msecs) {
 186                 mdelay(1);
 187                 msecs--;
 188         }
 189
 190         /* Leave the nmi callback set */
 191         disable_local_APIC();
 192 }
 193 #else
 194 static void nmi_shootdown_cpus(void)
 195 {
 196         /* There are no cpus to shootdown */
 197 }
 198 #endif
 199
 200 void machine_crash_shutdown(struct pt_regs *regs)
 201 {
 202         /* This function is only called after the system
 203          * has paniced or is otherwise in a critical state.
 204          * The minimum amount of code to allow a kexec'd kernel
 205          * to run successfully needs to happen here.
 206          *
 207          * In practice this means shooting down the other cpus in
 208          * an SMP system.
 209          */
 210         /* The kernel is broken so disable interrupts */
 211         local_irq_disable();
 212
 213         /* Make a note of crashing cpu. Will be used in NMI callback.*/
 214         crashing_cpu = smp_processor_id();
 215         nmi_shootdown_cpus();
 216         lapic_shutdown();
 217 #if defined(CONFIG_X86_IO_APIC)
 218         disable_IO_APIC();
 219 #endif
 220         crash_save_self(regs);
 221 }