release/src-rt-6.x.4708/linux/linux-2.6.36/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 2
  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 (;;);
 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         printk(KERN_EMERG "Sending IPI to other cpus...\n");
 128         msecs = 10000;
 129         while ((cpus_weight(cpus_in_crash) < ncpus) && (--msecs > 0)) {
 130                 cpu_relax();
 131                 mdelay(1);
 132         }
 133
 134         /* Would it be better to replace the trap vector here? */
 135
 136         if (cpus_weight(cpus_in_crash) < ncpus) {
 137                 printk(KERN_EMERG "done waiting: %d cpu(s) not responding\n",
 138                         ncpus - cpus_weight(cpus_in_crash));
 139                 printk(KERN_EMERG "Activate soft-reset to stop other cpu(s)\n");
 140                 cpus_in_sr = CPU_MASK_NONE;
 141                 atomic_set(&enter_on_soft_reset, 0);
 142                 while (cpus_weight(cpus_in_crash) < ncpus)
 143                         cpu_relax();
 144         }
 145         /*
 146          * Make sure all CPUs are entered via soft-reset if the kdump is
 147          * invoked using soft-reset.
 148          */
 149         if (cpu_isset(cpu, cpus_in_sr))
 150                 crash_soft_reset_check(cpu);
 151         /* Leave the IPI callback set */
 152 }
 153
 154 /* wait for all the CPUs to hit real mode but timeout if they don't come in */
 155 #ifdef CONFIG_PPC_STD_MMU_64
 156 static void crash_kexec_wait_realmode(int cpu)
 157 {
 158         unsigned int msecs;
 159         int i;
 160
 161         msecs = 10000;
 162         for (i=0; i < NR_CPUS && msecs > 0; i++) {
 163                 if (i == cpu)
 164                         continue;
 165
 166                 while (paca[i].kexec_state < KEXEC_STATE_REAL_MODE) {
 167                         barrier();
 168                         if (!cpu_possible(i)) {
 169                                 break;
 170                         }
 171                         if (!cpu_online(i)) {
 172                                 break;
 173                         }
 174                         msecs--;
 175                         mdelay(1);
 176                 }
 177         }
 178         mb();
 179 }
 180 #endif
 181
 182 /*
 183  * This function will be called by secondary cpus or by kexec cpu
 184  * if soft-reset is activated to stop some CPUs.
 185  */
 186 void crash_kexec_secondary(struct pt_regs *regs)
 187 {
 188         int cpu = smp_processor_id();
 189         unsigned long flags;
 190         int msecs = 5;
 191
 192         local_irq_save(flags);
 193         /* Wait 5ms if the kexec CPU is not entered yet. */
 194         while (crashing_cpu < 0) {
 195                 if (--msecs < 0) {
 196                         /*
 197                          * Either kdump image is not loaded or
 198                          * kdump process is not started - Probably xmon
 199                          * exited using 'x'(exit and recover) or
 200                          * kexec_should_crash() failed for all running tasks.
 201                          */
 202                         cpu_clear(cpu, cpus_in_sr);
 203                         local_irq_restore(flags);
 204                         return;
 205                 }
 206                 mdelay(1);
 207                 cpu_relax();
 208         }
 209         if (cpu == crashing_cpu) {
 210                 /*
 211                  * Panic CPU will enter this func only via soft-reset.
 212                  * Wait until all secondary CPUs entered and
 213                  * then start kexec boot.
 214                  */
 215                 crash_soft_reset_check(cpu);
 216                 cpu_set(crashing_cpu, cpus_in_crash);
 217                 if (ppc_md.kexec_cpu_down)
 218                         ppc_md.kexec_cpu_down(1, 0);
 219                 machine_kexec(kexec_crash_image);
 220                 /* NOTREACHED */
 221         }
 222         crash_ipi_callback(regs);
 223 }
 224
 225 #else
 226 static void crash_kexec_prepare_cpus(int cpu)
 227 {
 228         /*
 229          * move the secondarys to us so that we can copy
 230          * the new kernel 0-0x100 safely
 231          *
 232          * do this if kexec in setup.c ?
 233          */
 234 #ifdef CONFIG_PPC64
 235         smp_release_cpus();
 236 #else
 237 #endif
 238 }
 239
 240 void crash_kexec_secondary(struct pt_regs *regs)
 241 {
 242         cpus_in_sr = CPU_MASK_NONE;
 243 }
 244 #endif
 245 #ifdef CONFIG_SPU_BASE
 246
 247 #include <asm/spu.h>
 248 #include <asm/spu_priv1.h>
 249
 250 struct crash_spu_info {
 251         struct spu *spu;
 252         u32 saved_spu_runcntl_RW;
 253         u32 saved_spu_status_R;
 254         u32 saved_spu_npc_RW;
 255         u64 saved_mfc_sr1_RW;
 256         u64 saved_mfc_dar;
 257         u64 saved_mfc_dsisr;
 258 };
 259
 260 #define CRASH_NUM_SPUS  16      /* Enough for current hardware */
 261 static struct crash_spu_info crash_spu_info[CRASH_NUM_SPUS];
 262
 263 static void crash_kexec_stop_spus(void)
 264 {
 265         struct spu *spu;
 266         int i;
 267         u64 tmp;
 268
 269         for (i = 0; i < CRASH_NUM_SPUS; i++) {
 270                 if (!crash_spu_info[i].spu)
 271                         continue;
 272
 273                 spu = crash_spu_info[i].spu;
 274
 275                 crash_spu_info[i].saved_spu_runcntl_RW =
 276                         in_be32(&spu->problem->spu_runcntl_RW);
 277                 crash_spu_info[i].saved_spu_status_R =
 278                         in_be32(&spu->problem->spu_status_R);
 279                 crash_spu_info[i].saved_spu_npc_RW =
 280                         in_be32(&spu->problem->spu_npc_RW);
 281
 282                 crash_spu_info[i].saved_mfc_dar    = spu_mfc_dar_get(spu);
 283                 crash_spu_info[i].saved_mfc_dsisr  = spu_mfc_dsisr_get(spu);
 284                 tmp = spu_mfc_sr1_get(spu);
 285                 crash_spu_info[i].saved_mfc_sr1_RW = tmp;
 286
 287                 tmp &= ~MFC_STATE1_MASTER_RUN_CONTROL_MASK;
 288                 spu_mfc_sr1_set(spu, tmp);
 289
 290                 __delay(200);
 291         }
 292 }
 293
 294 void crash_register_spus(struct list_head *list)
 295 {
 296         struct spu *spu;
 297
 298         list_for_each_entry(spu, list, full_list) {
 299                 if (WARN_ON(spu->number >= CRASH_NUM_SPUS))
 300                         continue;
 301
 302                 crash_spu_info[spu->number].spu = spu;
 303         }
 304 }
 305
 306 #else
 307 static inline void crash_kexec_stop_spus(void)
 308 {
 309 }
 310 #endif /* CONFIG_SPU_BASE */
 311
 312 /*
 313  * Register a function to be called on shutdown.  Only use this if you
 314  * can't reset your device in the second kernel.
 315  */
 316 int crash_shutdown_register(crash_shutdown_t handler)
 317 {
 318         unsigned int i, rc;
 319
 320         spin_lock(&crash_handlers_lock);
 321         for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
 322                 if (!crash_shutdown_handles[i]) {
 323                         /* Insert handle at first empty entry */
 324                         crash_shutdown_handles[i] = handler;
 325                         rc = 0;
 326                         break;
 327                 }
 328
 329         if (i == CRASH_HANDLER_MAX) {
 330                 printk(KERN_ERR "Crash shutdown handles full, "
 331                        "not registered.\n");
 332                 rc = 1;
 333         }
 334
 335         spin_unlock(&crash_handlers_lock);
 336         return rc;
 337 }
 338 EXPORT_SYMBOL(crash_shutdown_register);
 339
 340 int crash_shutdown_unregister(crash_shutdown_t handler)
 341 {
 342         unsigned int i, rc;
 343
 344         spin_lock(&crash_handlers_lock);
 345         for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
 346                 if (crash_shutdown_handles[i] == handler)
 347                         break;
 348
 349         if (i == CRASH_HANDLER_MAX) {
 350                 printk(KERN_ERR "Crash shutdown handle not found\n");
 351                 rc = 1;
 352         } else {
 353                 /* Shift handles down */
 354                 for (; crash_shutdown_handles[i]; i++)
 355                         crash_shutdown_handles[i] =
 356                                 crash_shutdown_handles[i+1];
 357                 rc = 0;
 358         }
 359
 360         spin_unlock(&crash_handlers_lock);
 361         return rc;
 362 }
 363 EXPORT_SYMBOL(crash_shutdown_unregister);
 364
 365 static unsigned long crash_shutdown_buf[JMP_BUF_LEN];
 366 static int crash_shutdown_cpu = -1;
 367
 368 static int handle_fault(struct pt_regs *regs)
 369 {
 370         if (crash_shutdown_cpu == smp_processor_id())
 371                 longjmp(crash_shutdown_buf, 1);
 372         return 0;
 373 }
 374
 375 void default_machine_crash_shutdown(struct pt_regs *regs)
 376 {
 377         unsigned int i;
 378         int (*old_handler)(struct pt_regs *regs);
 379
 380
 381         /*
 382          * This function is only called after the system
 383          * has panicked or is otherwise in a critical state.
 384          * The minimum amount of code to allow a kexec'd kernel
 385          * to run successfully needs to happen here.
 386          *
 387          * In practice this means stopping other cpus in
 388          * an SMP system.
 389          * The kernel is broken so disable interrupts.
 390          */
 391         hard_irq_disable();
 392
 393         /*
 394          * Make a note of crashing cpu. Will be used in machine_kexec
 395          * such that another IPI will not be sent.
 396          */
 397         crashing_cpu = smp_processor_id();
 398         crash_save_cpu(regs, crashing_cpu);
 399         crash_kexec_prepare_cpus(crashing_cpu);
 400         cpu_set(crashing_cpu, cpus_in_crash);
 401 #if defined(CONFIG_PPC_STD_MMU_64) && defined(CONFIG_SMP)
 402         crash_kexec_wait_realmode(crashing_cpu);
 403 #endif
 404
 405         for_each_irq(i) {
 406                 struct irq_desc *desc = irq_to_desc(i);
 407
 408                 if (!desc || !desc->chip || !desc->chip->eoi)
 409                         continue;
 410
 411                 if (desc->status & IRQ_INPROGRESS)
 412                         desc->chip->eoi(i);
 413
 414                 if (!(desc->status & IRQ_DISABLED))
 415                         desc->chip->shutdown(i);
 416         }
 417
 418         /*
 419          * Call registered shutdown routines savely.  Swap out
 420          * __debugger_fault_handler, and replace on exit.
 421          */
 422         old_handler = __debugger_fault_handler;
 423         __debugger_fault_handler = handle_fault;
 424         crash_shutdown_cpu = smp_processor_id();
 425         for (i = 0; crash_shutdown_handles[i]; i++) {
 426                 if (setjmp(crash_shutdown_buf) == 0) {
 427                         /*
 428                          * Insert syncs and delay to ensure
 429                          * instructions in the dangerous region don't
 430                          * leak away from this protected region.
 431                          */
 432                         asm volatile("sync; isync");
 433                         /* dangerous region */
 434                         crash_shutdown_handles[i]();
 435                         asm volatile("sync; isync");
 436                 }
 437         }
 438         crash_shutdown_cpu = -1;
 439         __debugger_fault_handler = old_handler;
 440
 441         crash_kexec_stop_spus();
 442
 443         if (ppc_md.kexec_cpu_down)
 444                 ppc_md.kexec_cpu_down(1, 0);
 445 }