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/memblock.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/prom.h>
  34 #include <asm/firmware.h>
  35 #include <asm/smp.h>
  36 #include <asm/system.h>
  37 #include <asm/setjmp.h>
  38
  39 #ifdef DEBUG
  40 #include <asm/udbg.h>
  41 #define DBG(fmt...) udbg_printf(fmt)
  42 #else
  43 #define DBG(fmt...)
  44 #endif
  45
  46 /* This keeps a track of which one is crashing cpu. */
  47 int crashing_cpu = -1;
  48 static cpumask_t cpus_in_crash = CPU_MASK_NONE;
  49 cpumask_t cpus_in_sr = CPU_MASK_NONE;
  50
  51 #define CRASH_HANDLER_MAX 3
  52 /* NULL terminated list of shutdown handles */
  53 static crash_shutdown_t crash_shutdown_handles[CRASH_HANDLER_MAX+1];
  54 static DEFINE_SPINLOCK(crash_handlers_lock);
  55
  56 #ifdef CONFIG_SMP
  57 static atomic_t enter_on_soft_reset = ATOMIC_INIT(0);
  58
  59 void crash_ipi_callback(struct pt_regs *regs)
  60 {
  61         int cpu = smp_processor_id();
  62
  63         if (!cpu_online(cpu))
  64                 return;
  65
  66         hard_irq_disable();
  67         if (!cpu_isset(cpu, cpus_in_crash))
  68                 crash_save_cpu(regs, cpu);
  69         cpu_set(cpu, cpus_in_crash);
  70
  71         /*
  72          * Entered via soft-reset - could be the kdump
  73          * process is invoked using soft-reset or user activated
  74          * it if some CPU did not respond to an IPI.
  75          * For soft-reset, the secondary CPU can enter this func
  76          * twice. 1 - using IPI, and 2. soft-reset.
  77          * Tell the kexec CPU that entered via soft-reset and ready
  78          * to go down.
  79          */
  80         if (cpu_isset(cpu, cpus_in_sr)) {
  81                 cpu_clear(cpu, cpus_in_sr);
  82                 atomic_inc(&enter_on_soft_reset);
  83         }
  84
  85         /*
  86          * Starting the kdump boot.
  87          * This barrier is needed to make sure that all CPUs are stopped.
  88          * If not, soft-reset will be invoked to bring other CPUs.
  89          */
  90         while (!cpu_isset(crashing_cpu, cpus_in_crash))
  91                 cpu_relax();
  92
  93         if (ppc_md.kexec_cpu_down)
  94                 ppc_md.kexec_cpu_down(1, 1);
  95
  96 #ifdef CONFIG_PPC64
  97         kexec_smp_wait();
  98 #else
  99         for (;;);       /* FIXME */
 100 #endif
 101
 102         /* NOTREACHED */
 103 }
 104
 105 /*
 106  * Wait until all CPUs are entered via soft-reset.
 107  */
 108 static void crash_soft_reset_check(int cpu)
 109 {
 110         unsigned int ncpus = num_online_cpus() - 1;/* Excluding the panic cpu */
 111
 112         cpu_clear(cpu, cpus_in_sr);
 113         while (atomic_read(&enter_on_soft_reset) != ncpus)
 114                 cpu_relax();
 115 }
 116
 117
 118 static void crash_kexec_prepare_cpus(int cpu)
 119 {
 120         unsigned int msecs;
 121
 122         unsigned int ncpus = num_online_cpus() - 1;/* Excluding the panic cpu */
 123
 124         crash_send_ipi(crash_ipi_callback);
 125         smp_wmb();
 126
 127         /*
 128          * FIXME: Until we will have the way to stop other CPUs reliably,
 129          * the crash CPU will send an IPI and wait for other CPUs to
 130          * respond.
 131          * Delay of at least 10 seconds.
 132          */
 133         printk(KERN_EMERG "Sending IPI to other cpus...\n");
 134         msecs = 10000;
 135         while ((cpus_weight(cpus_in_crash) < ncpus) && (--msecs > 0)) {
 136                 cpu_relax();
 137                 mdelay(1);
 138         }
 139
 140         /* Would it be better to replace the trap vector here? */
 141
 142         /*
 143          * FIXME: In case if we do not get all CPUs, one possibility: ask the
 144          * user to do soft reset such that we get all.
 145          * Soft-reset will be used until better mechanism is implemented.
 146          */
 147         if (cpus_weight(cpus_in_crash) < ncpus) {
 148                 printk(KERN_EMERG "done waiting: %d cpu(s) not responding\n",
 149                         ncpus - cpus_weight(cpus_in_crash));
 150                 printk(KERN_EMERG "Activate soft-reset to stop other cpu(s)\n");
 151                 cpus_in_sr = CPU_MASK_NONE;
 152                 atomic_set(&enter_on_soft_reset, 0);
 153                 while (cpus_weight(cpus_in_crash) < ncpus)
 154                         cpu_relax();
 155         }
 156         /*
 157          * Make sure all CPUs are entered via soft-reset if the kdump is
 158          * invoked using soft-reset.
 159          */
 160         if (cpu_isset(cpu, cpus_in_sr))
 161                 crash_soft_reset_check(cpu);
 162         /* Leave the IPI callback set */
 163 }
 164
 165 /* wait for all the CPUs to hit real mode but timeout if they don't come in */
 166 #ifdef CONFIG_PPC_STD_MMU_64
 167 static void crash_kexec_wait_realmode(int cpu)
 168 {
 169         unsigned int msecs;
 170         int i;
 171
 172         msecs = 10000;
 173         for (i=0; i < NR_CPUS && msecs > 0; i++) {
 174                 if (i == cpu)
 175                         continue;
 176
 177                 while (paca[i].kexec_state < KEXEC_STATE_REAL_MODE) {
 178                         barrier();
 179                         if (!cpu_possible(i)) {
 180                                 break;
 181                         }
 182                         if (!cpu_online(i)) {
 183                                 break;
 184                         }
 185                         msecs--;
 186                         mdelay(1);
 187                 }
 188         }
 189         mb();
 190 }
 191 #endif  /* CONFIG_PPC_STD_MMU_64 */
 192
 193 /*
 194  * This function will be called by secondary cpus or by kexec cpu
 195  * if soft-reset is activated to stop some CPUs.
 196  */
 197 void crash_kexec_secondary(struct pt_regs *regs)
 198 {
 199         int cpu = smp_processor_id();
 200         unsigned long flags;
 201         int msecs = 5;
 202
 203         local_irq_save(flags);
 204         /* Wait 5ms if the kexec CPU is not entered yet. */
 205         while (crashing_cpu < 0) {
 206                 if (--msecs < 0) {
 207                         /*
 208                          * Either kdump image is not loaded or
 209                          * kdump process is not started - Probably xmon
 210                          * exited using 'x'(exit and recover) or
 211                          * kexec_should_crash() failed for all running tasks.
 212                          */
 213                         cpu_clear(cpu, cpus_in_sr);
 214                         local_irq_restore(flags);
 215                         return;
 216                 }
 217                 mdelay(1);
 218                 cpu_relax();
 219         }
 220         if (cpu == crashing_cpu) {
 221                 /*
 222                  * Panic CPU will enter this func only via soft-reset.
 223                  * Wait until all secondary CPUs entered and
 224                  * then start kexec boot.
 225                  */
 226                 crash_soft_reset_check(cpu);
 227                 cpu_set(crashing_cpu, cpus_in_crash);
 228                 if (ppc_md.kexec_cpu_down)
 229                         ppc_md.kexec_cpu_down(1, 0);
 230                 machine_kexec(kexec_crash_image);
 231                 /* NOTREACHED */
 232         }
 233         crash_ipi_callback(regs);
 234 }
 235
 236 #else   /* ! CONFIG_SMP */
 237 static inline void crash_kexec_wait_realmode(int cpu) {}
 238
 239 static void crash_kexec_prepare_cpus(int cpu)
 240 {
 241         /*
 242          * move the secondarys to us so that we can copy
 243          * the new kernel 0-0x100 safely
 244          *
 245          * do this if kexec in setup.c ?
 246          */
 247 #ifdef CONFIG_PPC64
 248         smp_release_cpus();
 249 #else
 250         /* FIXME */
 251 #endif
 252 }
 253
 254 void crash_kexec_secondary(struct pt_regs *regs)
 255 {
 256         cpus_in_sr = CPU_MASK_NONE;
 257 }
 258 #endif  /* CONFIG_SMP */
 259
 260 /*
 261  * Register a function to be called on shutdown.  Only use this if you
 262  * can't reset your device in the second kernel.
 263  */
 264 int crash_shutdown_register(crash_shutdown_t handler)
 265 {
 266         unsigned int i, rc;
 267
 268         spin_lock(&crash_handlers_lock);
 269         for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
 270                 if (!crash_shutdown_handles[i]) {
 271                         /* Insert handle at first empty entry */
 272                         crash_shutdown_handles[i] = handler;
 273                         rc = 0;
 274                         break;
 275                 }
 276
 277         if (i == CRASH_HANDLER_MAX) {
 278                 printk(KERN_ERR "Crash shutdown handles full, "
 279                        "not registered.\n");
 280                 rc = 1;
 281         }
 282
 283         spin_unlock(&crash_handlers_lock);
 284         return rc;
 285 }
 286 EXPORT_SYMBOL(crash_shutdown_register);
 287
 288 int crash_shutdown_unregister(crash_shutdown_t handler)
 289 {
 290         unsigned int i, rc;
 291
 292         spin_lock(&crash_handlers_lock);
 293         for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
 294                 if (crash_shutdown_handles[i] == handler)
 295                         break;
 296
 297         if (i == CRASH_HANDLER_MAX) {
 298                 printk(KERN_ERR "Crash shutdown handle not found\n");
 299                 rc = 1;
 300         } else {
 301                 /* Shift handles down */
 302                 for (; crash_shutdown_handles[i]; i++)
 303                         crash_shutdown_handles[i] =
 304                                 crash_shutdown_handles[i+1];
 305                 rc = 0;
 306         }
 307
 308         spin_unlock(&crash_handlers_lock);
 309         return rc;
 310 }
 311 EXPORT_SYMBOL(crash_shutdown_unregister);
 312
 313 static unsigned long crash_shutdown_buf[JMP_BUF_LEN];
 314 static int crash_shutdown_cpu = -1;
 315
 316 static int handle_fault(struct pt_regs *regs)
 317 {
 318         if (crash_shutdown_cpu == smp_processor_id())
 319                 longjmp(crash_shutdown_buf, 1);
 320         return 0;
 321 }
 322
 323 void default_machine_crash_shutdown(struct pt_regs *regs)
 324 {
 325         unsigned int i;
 326         int (*old_handler)(struct pt_regs *regs);
 327
 328
 329         /*
 330          * This function is only called after the system
 331          * has panicked or is otherwise in a critical state.
 332          * The minimum amount of code to allow a kexec'd kernel
 333          * to run successfully needs to happen here.
 334          *
 335          * In practice this means stopping other cpus in
 336          * an SMP system.
 337          * The kernel is broken so disable interrupts.
 338          */
 339         hard_irq_disable();
 340
 341         /*
 342          * Make a note of crashing cpu. Will be used in machine_kexec
 343          * such that another IPI will not be sent.
 344          */
 345         crashing_cpu = smp_processor_id();
 346         crash_save_cpu(regs, crashing_cpu);
 347         crash_kexec_prepare_cpus(crashing_cpu);
 348         cpu_set(crashing_cpu, cpus_in_crash);
 349         crash_kexec_wait_realmode(crashing_cpu);
 350
 351         machine_kexec_mask_interrupts();
 352
 353         /*
 354          * Call registered shutdown routines savely.  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; 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 }