crash.c 4.0 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186
  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. #include <linux/init.h>
  10. #include <linux/types.h>
  11. #include <linux/kernel.h>
  12. #include <linux/smp.h>
  13. #include <linux/irq.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. #include <asm/processor.h>
  20. #include <asm/hardirq.h>
  21. #include <asm/nmi.h>
  22. #include <asm/hw_irq.h>
  23. #include <asm/mach_apic.h>
  24. /* This keeps a track of which one is crashing cpu. */
  25. static int crashing_cpu;
  26. static u32 *append_elf_note(u32 *buf, char *name, unsigned type,
  27. void *data, size_t data_len)
  28. {
  29. struct elf_note note;
  30. note.n_namesz = strlen(name) + 1;
  31. note.n_descsz = data_len;
  32. note.n_type = type;
  33. memcpy(buf, &note, sizeof(note));
  34. buf += (sizeof(note) +3)/4;
  35. memcpy(buf, name, note.n_namesz);
  36. buf += (note.n_namesz + 3)/4;
  37. memcpy(buf, data, note.n_descsz);
  38. buf += (note.n_descsz + 3)/4;
  39. return buf;
  40. }
  41. static void final_note(u32 *buf)
  42. {
  43. struct elf_note note;
  44. note.n_namesz = 0;
  45. note.n_descsz = 0;
  46. note.n_type = 0;
  47. memcpy(buf, &note, sizeof(note));
  48. }
  49. static void crash_save_this_cpu(struct pt_regs *regs, int cpu)
  50. {
  51. struct elf_prstatus prstatus;
  52. u32 *buf;
  53. if ((cpu < 0) || (cpu >= NR_CPUS))
  54. return;
  55. /* Using ELF notes here is opportunistic.
  56. * I need a well defined structure format
  57. * for the data I pass, and I need tags
  58. * on the data to indicate what information I have
  59. * squirrelled away. ELF notes happen to provide
  60. * all of that that no need to invent something new.
  61. */
  62. buf = (u32*)per_cpu_ptr(crash_notes, cpu);
  63. if (!buf)
  64. return;
  65. memset(&prstatus, 0, sizeof(prstatus));
  66. prstatus.pr_pid = current->pid;
  67. elf_core_copy_regs(&prstatus.pr_reg, regs);
  68. buf = append_elf_note(buf, "CORE", NT_PRSTATUS, &prstatus,
  69. sizeof(prstatus));
  70. final_note(buf);
  71. }
  72. static void crash_save_self(struct pt_regs *regs)
  73. {
  74. int cpu;
  75. cpu = smp_processor_id();
  76. crash_save_this_cpu(regs, cpu);
  77. }
  78. #ifdef CONFIG_SMP
  79. static atomic_t waiting_for_crash_ipi;
  80. static int crash_nmi_callback(struct pt_regs *regs, int cpu)
  81. {
  82. /*
  83. * Don't do anything if this handler is invoked on crashing cpu.
  84. * Otherwise, system will completely hang. Crashing cpu can get
  85. * an NMI if system was initially booted with nmi_watchdog parameter.
  86. */
  87. if (cpu == crashing_cpu)
  88. return 1;
  89. local_irq_disable();
  90. crash_save_this_cpu(regs, cpu);
  91. disable_local_APIC();
  92. atomic_dec(&waiting_for_crash_ipi);
  93. /* Assume hlt works */
  94. for(;;)
  95. halt();
  96. return 1;
  97. }
  98. static void smp_send_nmi_allbutself(void)
  99. {
  100. send_IPI_allbutself(NMI_VECTOR);
  101. }
  102. /*
  103. * This code is a best effort heuristic to get the
  104. * other cpus to stop executing. So races with
  105. * cpu hotplug shouldn't matter.
  106. */
  107. static void nmi_shootdown_cpus(void)
  108. {
  109. unsigned long msecs;
  110. atomic_set(&waiting_for_crash_ipi, num_online_cpus() - 1);
  111. set_nmi_callback(crash_nmi_callback);
  112. /*
  113. * Ensure the new callback function is set before sending
  114. * out the NMI
  115. */
  116. wmb();
  117. smp_send_nmi_allbutself();
  118. msecs = 1000; /* Wait at most a second for the other cpus to stop */
  119. while ((atomic_read(&waiting_for_crash_ipi) > 0) && msecs) {
  120. mdelay(1);
  121. msecs--;
  122. }
  123. /* Leave the nmi callback set */
  124. disable_local_APIC();
  125. }
  126. #else
  127. static void nmi_shootdown_cpus(void)
  128. {
  129. /* There are no cpus to shootdown */
  130. }
  131. #endif
  132. void machine_crash_shutdown(struct pt_regs *regs)
  133. {
  134. /*
  135. * This function is only called after the system
  136. * has panicked or is otherwise in a critical state.
  137. * The minimum amount of code to allow a kexec'd kernel
  138. * to run successfully needs to happen here.
  139. *
  140. * In practice this means shooting down the other cpus in
  141. * an SMP system.
  142. */
  143. /* The kernel is broken so disable interrupts */
  144. local_irq_disable();
  145. /* Make a note of crashing cpu. Will be used in NMI callback.*/
  146. crashing_cpu = smp_processor_id();
  147. nmi_shootdown_cpus();
  148. if(cpu_has_apic)
  149. disable_local_APIC();
  150. #if defined(CONFIG_X86_IO_APIC)
  151. disable_IO_APIC();
  152. #endif
  153. crash_save_self(regs);
  154. }