arch/powerpc/kernel/crash.c

   1 /*
   2  * Architecture specific (PPC64) functions for kexec based crash dumps.
   3  *
   4  * Copyright (C) 2005, IBM Corp.
   5  *
   6  * Created by: Haren Myneni
   7  *
   8  * This source code is licensed under the GNU General Public License,
   9  * Version 2.  See the file COPYING for more details.
  10  *
  11  */
  12
  13 #undef DEBUG
  14
  15 #include <linux/kernel.h>
  16 #include <linux/smp.h>
  17 #include <linux/reboot.h>
  18 #include <linux/kexec.h>
  19 #include <linux/bootmem.h>
  20 #include <linux/crash_dump.h>
  21 #include <linux/delay.h>
  22 #include <linux/elf.h>
  23 #include <linux/elfcore.h>
  24 #include <linux/init.h>
  25 #include <linux/irq.h>
  26 #include <linux/types.h>
  27
  28 #include <asm/processor.h>
  29 #include <asm/machdep.h>
  30 #include <asm/kdump.h>
  31 #include <asm/lmb.h>
  32 #include <asm/firmware.h>
  33 #include <asm/smp.h>
  34
  35 #ifdef DEBUG
  36 #include <asm/udbg.h>
  37 #define DBG(fmt...) udbg_printf(fmt)
  38 #else
  39 #define DBG(fmt...)
  40 #endif
  41
  42 /* This keeps a track of which one is crashing cpu. */
  43 int crashing_cpu = -1;
  44
  45 static u32 *append_elf_note(u32 *buf, char *name, unsigned type, void *data,
  46                                                                size_t data_len)
  47 {
  48         struct elf_note note;
  49
  50         note.n_namesz = strlen(name) + 1;
  51         note.n_descsz = data_len;
  52         note.n_type   = type;
  53         memcpy(buf, &note, sizeof(note));
  54         buf += (sizeof(note) +3)/4;
  55         memcpy(buf, name, note.n_namesz);
  56         buf += (note.n_namesz + 3)/4;
  57         memcpy(buf, data, note.n_descsz);
  58         buf += (note.n_descsz + 3)/4;
  59
  60         return buf;
  61 }
  62
  63 static void final_note(u32 *buf)
  64 {
  65         struct elf_note note;
  66
  67         note.n_namesz = 0;
  68         note.n_descsz = 0;
  69         note.n_type   = 0;
  70         memcpy(buf, &note, sizeof(note));
  71 }
  72
  73 static void crash_save_this_cpu(struct pt_regs *regs, int cpu)
  74 {
  75         struct elf_prstatus prstatus;
  76         u32 *buf;
  77
  78         if ((cpu < 0) || (cpu >= NR_CPUS))
  79                 return;
  80
  81         /* Using ELF notes here is opportunistic.
  82          * I need a well defined structure format
  83          * for the data I pass, and I need tags
  84          * on the data to indicate what information I have
  85          * squirrelled away.  ELF notes happen to provide
  86          * all of that that no need to invent something new.
  87          */
  88         buf = (u32*)per_cpu_ptr(crash_notes, cpu);
  89         if (!buf)
  90                 return;
  91
  92         memset(&prstatus, 0, sizeof(prstatus));
  93         prstatus.pr_pid = current->pid;
  94         elf_core_copy_regs(&prstatus.pr_reg, regs);
  95         buf = append_elf_note(buf, "CORE", NT_PRSTATUS, &prstatus,
  96                         sizeof(prstatus));
  97         final_note(buf);
  98 }
  99
 100 #ifdef CONFIG_SMP
 101 static atomic_t waiting_for_crash_ipi;
 102
 103 void crash_ipi_callback(struct pt_regs *regs)
 104 {
 105         int cpu = smp_processor_id();
 106
 107         if (cpu == crashing_cpu)
 108                 return;
 109
 110         if (!cpu_online(cpu))
 111                 return;
 112
 113         if (ppc_md.kexec_cpu_down)
 114                 ppc_md.kexec_cpu_down(1, 1);
 115
 116         local_irq_disable();
 117
 118         crash_save_this_cpu(regs, cpu);
 119         atomic_dec(&waiting_for_crash_ipi);
 120         kexec_smp_wait();
 121         /* NOTREACHED */
 122 }
 123
 124 static void crash_kexec_prepare_cpus(void)
 125 {
 126         unsigned int msecs;
 127
 128         atomic_set(&waiting_for_crash_ipi, num_online_cpus() - 1);
 129
 130         crash_send_ipi(crash_ipi_callback);
 131         smp_wmb();
 132
 133         /*
 134          * FIXME: Until we will have the way to stop other CPUSs reliabally,
 135          * the crash CPU will send an IPI and wait for other CPUs to
 136          * respond. If not, proceed the kexec boot even though we failed to
 137          * capture other CPU states.
 138          * Delay of at least 10 seconds.
 139          */
 140         printk(KERN_ALERT "Sending IPI to other cpus...\n");
 141         msecs = 10000;
 142         while ((atomic_read(&waiting_for_crash_ipi) > 0) && (--msecs > 0)) {
 143                 barrier();
 144                 mdelay(1);
 145         }
 146
 147         /* Would it be better to replace the trap vector here? */
 148
 149         /*
 150          * FIXME: In case if we do not get all CPUs, one possibility: ask the
 151          * user to do soft reset such that we get all.
 152          * IPI handler is already set by the panic cpu initially. Therefore,
 153          * all cpus could invoke this handler from die() and the panic CPU
 154          * will call machine_kexec() directly from this handler to do
 155          * kexec boot.
 156          */
 157         if (atomic_read(&waiting_for_crash_ipi))
 158                 printk(KERN_ALERT "done waiting: %d cpus not responding\n",
 159                         atomic_read(&waiting_for_crash_ipi));
 160         /* Leave the IPI callback set */
 161 }
 162 #else
 163 static void crash_kexec_prepare_cpus(void)
 164 {
 165         /*
 166          * move the secondarys to us so that we can copy
 167          * the new kernel 0-0x100 safely
 168          *
 169          * do this if kexec in setup.c ?
 170          */
 171         smp_release_cpus();
 172 }
 173
 174 #endif
 175
 176 void default_machine_crash_shutdown(struct pt_regs *regs)
 177 {
 178         unsigned int irq;
 179
 180         /*
 181          * This function is only called after the system
 182          * has panicked or is otherwise in a critical state.
 183          * The minimum amount of code to allow a kexec'd kernel
 184          * to run successfully needs to happen here.
 185          *
 186          * In practice this means stopping other cpus in
 187          * an SMP system.
 188          * The kernel is broken so disable interrupts.
 189          */
 190         local_irq_disable();
 191
 192         for_each_irq(irq) {
 193                 struct irq_desc *desc = irq_descp(irq);
 194
 195                 if (desc->status & IRQ_INPROGRESS)
 196                         desc->chip->end(irq);
 197
 198                 if (!(desc->status & IRQ_DISABLED))
 199                         desc->chip->disable(irq);
 200         }
 201
 202         if (ppc_md.kexec_cpu_down)
 203                 ppc_md.kexec_cpu_down(1, 0);
 204
 205         /*
 206          * Make a note of crashing cpu. Will be used in machine_kexec
 207          * such that another IPI will not be sent.
 208          */
 209         crashing_cpu = smp_processor_id();
 210         crash_kexec_prepare_cpus();
 211         crash_save_this_cpu(regs, crashing_cpu);
 212 }