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 #include <linux/irq.h>
  28
  29 #include <asm/processor.h>
  30 #include <asm/machdep.h>
  31 #include <asm/kexec.h>
  32 #include <asm/kdump.h>
  33 #include <asm/lmb.h>
  34 #include <asm/firmware.h>
  35 #include <asm/smp.h>
  36
  37 #ifdef DEBUG
  38 #include <asm/udbg.h>
  39 #define DBG(fmt...) udbg_printf(fmt)
  40 #else
  41 #define DBG(fmt...)
  42 #endif
  43
  44 /* This keeps a track of which one is crashing cpu. */
  45 int crashing_cpu = -1;
  46 static cpumask_t cpus_in_crash = CPU_MASK_NONE;
  47
  48 static u32 *append_elf_note(u32 *buf, char *name, unsigned type, void *data,
  49                                                                size_t data_len)
  50 {
  51         struct elf_note note;
  52
  53         note.n_namesz = strlen(name) + 1;
  54         note.n_descsz = data_len;
  55         note.n_type   = type;
  56         memcpy(buf, &note, sizeof(note));
  57         buf += (sizeof(note) +3)/4;
  58         memcpy(buf, name, note.n_namesz);
  59         buf += (note.n_namesz + 3)/4;
  60         memcpy(buf, data, note.n_descsz);
  61         buf += (note.n_descsz + 3)/4;
  62
  63         return buf;
  64 }
  65
  66 static void final_note(u32 *buf)
  67 {
  68         struct elf_note note;
  69
  70         note.n_namesz = 0;
  71         note.n_descsz = 0;
  72         note.n_type   = 0;
  73         memcpy(buf, &note, sizeof(note));
  74 }
  75
  76 static void crash_save_this_cpu(struct pt_regs *regs, int cpu)
  77 {
  78         struct elf_prstatus prstatus;
  79         u32 *buf;
  80
  81         if ((cpu < 0) || (cpu >= NR_CPUS))
  82                 return;
  83
  84         /* Using ELF notes here is opportunistic.
  85          * I need a well defined structure format
  86          * for the data I pass, and I need tags
  87          * on the data to indicate what information I have
  88          * squirrelled away.  ELF notes happen to provide
  89          * all of that that no need to invent something new.
  90          */
  91         buf = (u32*)per_cpu_ptr(crash_notes, cpu);
  92         if (!buf)
  93                 return;
  94
  95         memset(&prstatus, 0, sizeof(prstatus));
  96         prstatus.pr_pid = current->pid;
  97         elf_core_copy_regs(&prstatus.pr_reg, regs);
  98         buf = append_elf_note(buf, "CORE", NT_PRSTATUS, &prstatus,
  99                         sizeof(prstatus));
 100         final_note(buf);
 101 }
 102
 103 #ifdef CONFIG_SMP
 104 static atomic_t enter_on_soft_reset = ATOMIC_INIT(0);
 105
 106 void crash_ipi_callback(struct pt_regs *regs)
 107 {
 108         int cpu = smp_processor_id();
 109
 110         if (!cpu_online(cpu))
 111                 return;
 112
 113         local_irq_disable();
 114         if (!cpu_isset(cpu, cpus_in_crash))
 115                 crash_save_this_cpu(regs, cpu);
 116         cpu_set(cpu, cpus_in_crash);
 117
 118         /*
 119          * Entered via soft-reset - could be the kdump
 120          * process is invoked using soft-reset or user activated
 121          * it if some CPU did not respond to an IPI.
 122          * For soft-reset, the secondary CPU can enter this func
 123          * twice. 1 - using IPI, and 2. soft-reset.
 124          * Tell the kexec CPU that entered via soft-reset and ready
 125          * to go down.
 126          */
 127         if (cpu_isset(cpu, cpus_in_sr)) {
 128                 cpu_clear(cpu, cpus_in_sr);
 129                 atomic_inc(&enter_on_soft_reset);
 130         }
 131
 132         /*
 133          * Starting the kdump boot.
 134          * This barrier is needed to make sure that all CPUs are stopped.
 135          * If not, soft-reset will be invoked to bring other CPUs.
 136          */
 137         while (!cpu_isset(crashing_cpu, cpus_in_crash))
 138                 cpu_relax();
 139
 140         if (ppc_md.kexec_cpu_down)
 141                 ppc_md.kexec_cpu_down(1, 1);
 142         kexec_smp_wait();
 143         /* NOTREACHED */
 144 }
 145
 146 /*
 147  * Wait until all CPUs are entered via soft-reset.
 148  */
 149 static void crash_soft_reset_check(int cpu)
 150 {
 151         unsigned int ncpus = num_online_cpus() - 1;/* Excluding the panic cpu */
 152
 153         cpu_clear(cpu, cpus_in_sr);
 154         while (atomic_read(&enter_on_soft_reset) != ncpus)
 155                 cpu_relax();
 156 }
 157
 158
 159 static void crash_kexec_prepare_cpus(int cpu)
 160 {
 161         unsigned int msecs;
 162
 163         unsigned int ncpus = num_online_cpus() - 1;/* Excluding the panic cpu */
 164
 165         crash_send_ipi(crash_ipi_callback);
 166         smp_wmb();
 167
 168         /*
 169          * FIXME: Until we will have the way to stop other CPUSs reliabally,
 170          * the crash CPU will send an IPI and wait for other CPUs to
 171          * respond.
 172          * Delay of at least 10 seconds.
 173          */
 174         printk(KERN_EMERG "Sending IPI to other cpus...\n");
 175         msecs = 10000;
 176         while ((cpus_weight(cpus_in_crash) < ncpus) && (--msecs > 0)) {
 177                 cpu_relax();
 178                 mdelay(1);
 179         }
 180
 181         /* Would it be better to replace the trap vector here? */
 182
 183         /*
 184          * FIXME: In case if we do not get all CPUs, one possibility: ask the
 185          * user to do soft reset such that we get all.
 186          * Soft-reset will be used until better mechanism is implemented.
 187          */
 188         if (cpus_weight(cpus_in_crash) < ncpus) {
 189                 printk(KERN_EMERG "done waiting: %d cpu(s) not responding\n",
 190                         ncpus - cpus_weight(cpus_in_crash));
 191                 printk(KERN_EMERG "Activate soft-reset to stop other cpu(s)\n");
 192                 cpus_in_sr = CPU_MASK_NONE;
 193                 atomic_set(&enter_on_soft_reset, 0);
 194                 while (cpus_weight(cpus_in_crash) < ncpus)
 195                         cpu_relax();
 196         }
 197         /*
 198          * Make sure all CPUs are entered via soft-reset if the kdump is
 199          * invoked using soft-reset.
 200          */
 201         if (cpu_isset(cpu, cpus_in_sr))
 202                 crash_soft_reset_check(cpu);
 203         /* Leave the IPI callback set */
 204 }
 205
 206 /*
 207  * This function will be called by secondary cpus or by kexec cpu
 208  * if soft-reset is activated to stop some CPUs.
 209  */
 210 void crash_kexec_secondary(struct pt_regs *regs)
 211 {
 212         int cpu = smp_processor_id();
 213         unsigned long flags;
 214         int msecs = 5;
 215
 216         local_irq_save(flags);
 217         /* Wait 5ms if the kexec CPU is not entered yet. */
 218         while (crashing_cpu < 0) {
 219                 if (--msecs < 0) {
 220                         /*
 221                          * Either kdump image is not loaded or
 222                          * kdump process is not started - Probably xmon
 223                          * exited using 'x'(exit and recover) or
 224                          * kexec_should_crash() failed for all running tasks.
 225                          */
 226                         cpu_clear(cpu, cpus_in_sr);
 227                         local_irq_restore(flags);
 228                         return;
 229                 }
 230                 mdelay(1);
 231                 cpu_relax();
 232         }
 233         if (cpu == crashing_cpu) {
 234                 /*
 235                  * Panic CPU will enter this func only via soft-reset.
 236                  * Wait until all secondary CPUs entered and
 237                  * then start kexec boot.
 238                  */
 239                 crash_soft_reset_check(cpu);
 240                 cpu_set(crashing_cpu, cpus_in_crash);
 241                 if (ppc_md.kexec_cpu_down)
 242                         ppc_md.kexec_cpu_down(1, 0);
 243                 machine_kexec(kexec_crash_image);
 244                 /* NOTREACHED */
 245         }
 246         crash_ipi_callback(regs);
 247 }
 248
 249 #else
 250 static void crash_kexec_prepare_cpus(int cpu)
 251 {
 252         /*
 253          * move the secondarys to us so that we can copy
 254          * the new kernel 0-0x100 safely
 255          *
 256          * do this if kexec in setup.c ?
 257          */
 258         smp_release_cpus();
 259 }
 260
 261 void crash_kexec_secondary(struct pt_regs *regs)
 262 {
 263         cpus_in_sr = CPU_MASK_NONE;
 264 }
 265 #endif
 266
 267 void default_machine_crash_shutdown(struct pt_regs *regs)
 268 {
 269         unsigned int irq;
 270
 271         /*
 272          * This function is only called after the system
 273          * has panicked or is otherwise in a critical state.
 274          * The minimum amount of code to allow a kexec'd kernel
 275          * to run successfully needs to happen here.
 276          *
 277          * In practice this means stopping other cpus in
 278          * an SMP system.
 279          * The kernel is broken so disable interrupts.
 280          */
 281         local_irq_disable();
 282
 283         for_each_irq(irq) {
 284                 struct irq_desc *desc = irq_desc + irq;
 285
 286                 if (desc->status & IRQ_INPROGRESS)
 287                         desc->chip->end(irq);
 288
 289                 if (!(desc->status & IRQ_DISABLED))
 290                         desc->chip->disable(irq);
 291         }
 292
 293         /*
 294          * Make a note of crashing cpu. Will be used in machine_kexec
 295          * such that another IPI will not be sent.
 296          */
 297         crashing_cpu = smp_processor_id();
 298         crash_save_this_cpu(regs, crashing_cpu);
 299         crash_kexec_prepare_cpus(crashing_cpu);
 300         cpu_set(crashing_cpu, cpus_in_crash);
 301         if (ppc_md.kexec_cpu_down)
 302                 ppc_md.kexec_cpu_down(1, 0);
 303 }