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 #include <linux/kernel.h>
  14 #include <linux/smp.h>
  15 #include <linux/reboot.h>
  16 #include <linux/kexec.h>
  17 #include <linux/export.h>
  18 #include <linux/crash_dump.h>
  19 #include <linux/delay.h>
  20 #include <linux/irq.h>
  21 #include <linux/types.h>
  22
  23 #include <asm/processor.h>
  24 #include <asm/machdep.h>
  25 #include <asm/kexec.h>
  26 #include <asm/kdump.h>
  27 #include <asm/prom.h>
  28 #include <asm/smp.h>
  29 #include <asm/setjmp.h>
  30 #include <asm/debug.h>
  31
  32 /*
  33  * The primary CPU waits a while for all secondary CPUs to enter. This is to
  34  * avoid sending an IPI if the secondary CPUs are entering
  35  * crash_kexec_secondary on their own (eg via a system reset).
  36  *
  37  * The secondary timeout has to be longer than the primary. Both timeouts are
  38  * in milliseconds.
  39  */
  40 #define PRIMARY_TIMEOUT         500
  41 #define SECONDARY_TIMEOUT       1000
  42
  43 #define IPI_TIMEOUT             10000
  44 #define REAL_MODE_TIMEOUT       10000
  45
  46 static int time_to_dump;
  47 /*
  48  * crash_wake_offline should be set to 1 by platforms that intend to wake
  49  * up offline cpus prior to jumping to a kdump kernel. Currently powernv
  50  * sets it to 1, since we want to avoid things from happening when an
  51  * offline CPU wakes up due to something like an HMI (malfunction error),
  52  * which propagates to all threads.
  53  */
  54 int crash_wake_offline;
  55
  56 #define CRASH_HANDLER_MAX 3
  57 /* List of shutdown handles */
  58 static crash_shutdown_t crash_shutdown_handles[CRASH_HANDLER_MAX];
  59 static DEFINE_SPINLOCK(crash_handlers_lock);
  60
  61 static unsigned long crash_shutdown_buf[JMP_BUF_LEN];
  62 static int crash_shutdown_cpu = -1;
  63
  64 static int handle_fault(struct pt_regs *regs)
  65 {
  66         if (crash_shutdown_cpu == smp_processor_id())
  67                 longjmp(crash_shutdown_buf, 1);
  68         return 0;
  69 }
  70
  71 #ifdef CONFIG_SMP
  72
  73 static atomic_t cpus_in_crash;
  74 void crash_ipi_callback(struct pt_regs *regs)
  75 {
  76         static cpumask_t cpus_state_saved = CPU_MASK_NONE;
  77
  78         int cpu = smp_processor_id();
  79
  80         hard_irq_disable();
  81         if (!cpumask_test_cpu(cpu, &cpus_state_saved)) {
  82                 crash_save_cpu(regs, cpu);
  83                 cpumask_set_cpu(cpu, &cpus_state_saved);
  84         }
  85
  86         atomic_inc(&cpus_in_crash);
  87         smp_mb__after_atomic();
  88
  89         /*
  90          * Starting the kdump boot.
  91          * This barrier is needed to make sure that all CPUs are stopped.
  92          */
  93         while (!time_to_dump)
  94                 cpu_relax();
  95
  96         if (ppc_md.kexec_cpu_down)
  97                 ppc_md.kexec_cpu_down(1, 1);
  98
  99 #ifdef CONFIG_PPC64
 100         kexec_smp_wait();
 101 #else
 102         for (;;);       /* FIXME */
 103 #endif
 104
 105         /* NOTREACHED */
 106 }
 107
 108 static void crash_kexec_prepare_cpus(int cpu)
 109 {
 110         unsigned int msecs;
 111         unsigned int ncpus = num_online_cpus() - 1;/* Excluding the panic cpu */
 112         int tries = 0;
 113         int (*old_handler)(struct pt_regs *regs);
 114
 115         printk(KERN_EMERG "Sending IPI to other CPUs\n");
 116
 117         if (crash_wake_offline)
 118                 ncpus = num_present_cpus() - 1;
 119
 120         crash_send_ipi(crash_ipi_callback);
 121         smp_wmb();
 122
 123 again:
 124         /*
 125          * FIXME: Until we will have the way to stop other CPUs reliably,
 126          * the crash CPU will send an IPI and wait for other CPUs to
 127          * respond.
 128          */
 129         msecs = IPI_TIMEOUT;
 130         while ((atomic_read(&cpus_in_crash) < ncpus) && (--msecs > 0))
 131                 mdelay(1);
 132
 133         /* Would it be better to replace the trap vector here? */
 134
 135         if (atomic_read(&cpus_in_crash) >= ncpus) {
 136                 printk(KERN_EMERG "IPI complete\n");
 137                 return;
 138         }
 139
 140         printk(KERN_EMERG "ERROR: %d cpu(s) not responding\n",
 141                 ncpus - atomic_read(&cpus_in_crash));
 142
 143         /*
 144          * If we have a panic timeout set then we can't wait indefinitely
 145          * for someone to activate system reset. We also give up on the
 146          * second time through if system reset fail to work.
 147          */
 148         if ((panic_timeout > 0) || (tries > 0))
 149                 return;
 150
 151         /*
 152          * A system reset will cause all CPUs to take an 0x100 exception.
 153          * The primary CPU returns here via setjmp, and the secondary
 154          * CPUs reexecute the crash_kexec_secondary path.
 155          */
 156         old_handler = __debugger;
 157         __debugger = handle_fault;
 158         crash_shutdown_cpu = smp_processor_id();
 159
 160         if (setjmp(crash_shutdown_buf) == 0) {
 161                 printk(KERN_EMERG "Activate system reset (dumprestart) "
 162                                   "to stop other cpu(s)\n");
 163
 164                 /*
 165                  * A system reset will force all CPUs to execute the
 166                  * crash code again. We need to reset cpus_in_crash so we
 167                  * wait for everyone to do this.
 168                  */
 169                 atomic_set(&cpus_in_crash, 0);
 170                 smp_mb();
 171
 172                 while (atomic_read(&cpus_in_crash) < ncpus)
 173                         cpu_relax();
 174         }
 175
 176         crash_shutdown_cpu = -1;
 177         __debugger = old_handler;
 178
 179         tries++;
 180         goto again;
 181 }
 182
 183 /*
 184  * This function will be called by secondary cpus.
 185  */
 186 void crash_kexec_secondary(struct pt_regs *regs)
 187 {
 188         unsigned long flags;
 189         int msecs = SECONDARY_TIMEOUT;
 190
 191         local_irq_save(flags);
 192
 193         /* Wait for the primary crash CPU to signal its progress */
 194         while (crashing_cpu < 0) {
 195                 if (--msecs < 0) {
 196                         /* No response, kdump image may not have been loaded */
 197                         local_irq_restore(flags);
 198                         return;
 199                 }
 200
 201                 mdelay(1);
 202         }
 203
 204         crash_ipi_callback(regs);
 205 }
 206
 207 #else   /* ! CONFIG_SMP */
 208
 209 static void crash_kexec_prepare_cpus(int cpu)
 210 {
 211         /*
 212          * move the secondaries to us so that we can copy
 213          * the new kernel 0-0x100 safely
 214          *
 215          * do this if kexec in setup.c ?
 216          */
 217 #ifdef CONFIG_PPC64
 218         smp_release_cpus();
 219 #else
 220         /* FIXME */
 221 #endif
 222 }
 223
 224 void crash_kexec_secondary(struct pt_regs *regs)
 225 {
 226 }
 227 #endif  /* CONFIG_SMP */
 228
 229 /* wait for all the CPUs to hit real mode but timeout if they don't come in */
 230 #if defined(CONFIG_SMP) && defined(CONFIG_PPC64)
 231 static void __maybe_unused crash_kexec_wait_realmode(int cpu)
 232 {
 233         unsigned int msecs;
 234         int i;
 235
 236         msecs = REAL_MODE_TIMEOUT;
 237         for (i=0; i < nr_cpu_ids && msecs > 0; i++) {
 238                 if (i == cpu)
 239                         continue;
 240
 241                 while (paca[i].kexec_state < KEXEC_STATE_REAL_MODE) {
 242                         barrier();
 243                         if (!cpu_possible(i) || !cpu_online(i) || (msecs <= 0))
 244                                 break;
 245                         msecs--;
 246                         mdelay(1);
 247                 }
 248         }
 249         mb();
 250 }
 251 #else
 252 static inline void crash_kexec_wait_realmode(int cpu) {}
 253 #endif  /* CONFIG_SMP && CONFIG_PPC64 */
 254
 255 /*
 256  * Register a function to be called on shutdown.  Only use this if you
 257  * can't reset your device in the second kernel.
 258  */
 259 int crash_shutdown_register(crash_shutdown_t handler)
 260 {
 261         unsigned int i, rc;
 262
 263         spin_lock(&crash_handlers_lock);
 264         for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
 265                 if (!crash_shutdown_handles[i]) {
 266                         /* Insert handle at first empty entry */
 267                         crash_shutdown_handles[i] = handler;
 268                         rc = 0;
 269                         break;
 270                 }
 271
 272         if (i == CRASH_HANDLER_MAX) {
 273                 printk(KERN_ERR "Crash shutdown handles full, "
 274                        "not registered.\n");
 275                 rc = 1;
 276         }
 277
 278         spin_unlock(&crash_handlers_lock);
 279         return rc;
 280 }
 281 EXPORT_SYMBOL(crash_shutdown_register);
 282
 283 int crash_shutdown_unregister(crash_shutdown_t handler)
 284 {
 285         unsigned int i, rc;
 286
 287         spin_lock(&crash_handlers_lock);
 288         for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
 289                 if (crash_shutdown_handles[i] == handler)
 290                         break;
 291
 292         if (i == CRASH_HANDLER_MAX) {
 293                 printk(KERN_ERR "Crash shutdown handle not found\n");
 294                 rc = 1;
 295         } else {
 296                 /* Shift handles down */
 297                 for (; i < (CRASH_HANDLER_MAX - 1); i++)
 298                         crash_shutdown_handles[i] =
 299                                 crash_shutdown_handles[i+1];
 300                 /*
 301                  * Reset last entry to NULL now that it has been shifted down,
 302                  * this will allow new handles to be added here.
 303                  */
 304                 crash_shutdown_handles[i] = NULL;
 305                 rc = 0;
 306         }
 307
 308         spin_unlock(&crash_handlers_lock);
 309         return rc;
 310 }
 311 EXPORT_SYMBOL(crash_shutdown_unregister);
 312
 313 void default_machine_crash_shutdown(struct pt_regs *regs)
 314 {
 315         unsigned int i;
 316         int (*old_handler)(struct pt_regs *regs);
 317
 318         /*
 319          * This function is only called after the system
 320          * has panicked or is otherwise in a critical state.
 321          * The minimum amount of code to allow a kexec'd kernel
 322          * to run successfully needs to happen here.
 323          *
 324          * In practice this means stopping other cpus in
 325          * an SMP system.
 326          * The kernel is broken so disable interrupts.
 327          */
 328         hard_irq_disable();
 329
 330         /*
 331          * Make a note of crashing cpu. Will be used in machine_kexec
 332          * such that another IPI will not be sent.
 333          */
 334         crashing_cpu = smp_processor_id();
 335
 336         /*
 337          * If we came in via system reset, wait a while for the secondary
 338          * CPUs to enter.
 339          */
 340         if (TRAP(regs) == 0x100)
 341                 mdelay(PRIMARY_TIMEOUT);
 342
 343         crash_kexec_prepare_cpus(crashing_cpu);
 344
 345         crash_save_cpu(regs, crashing_cpu);
 346
 347         time_to_dump = 1;
 348
 349         crash_kexec_wait_realmode(crashing_cpu);
 350
 351         machine_kexec_mask_interrupts();
 352
 353         /*
 354          * Call registered shutdown routines safely.  Swap out
 355          * __debugger_fault_handler, and replace on exit.
 356          */
 357         old_handler = __debugger_fault_handler;
 358         __debugger_fault_handler = handle_fault;
 359         crash_shutdown_cpu = smp_processor_id();
 360         for (i = 0; i < CRASH_HANDLER_MAX && crash_shutdown_handles[i]; i++) {
 361                 if (setjmp(crash_shutdown_buf) == 0) {
 362                         /*
 363                          * Insert syncs and delay to ensure
 364                          * instructions in the dangerous region don't
 365                          * leak away from this protected region.
 366                          */
 367                         asm volatile("sync; isync");
 368                         /* dangerous region */
 369                         crash_shutdown_handles[i]();
 370                         asm volatile("sync; isync");
 371                 }
 372         }
 373         crash_shutdown_cpu = -1;
 374         __debugger_fault_handler = old_handler;
 375
 376         if (ppc_md.kexec_cpu_down)
 377                 ppc_md.kexec_cpu_down(1, 0);
 378 }