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 (;;);       /* 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 CPUSs reliabally,
 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
 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
 237 static void crash_kexec_prepare_cpus(int cpu)
 238 {
 239         /*
 240          * move the secondarys to us so that we can copy
 241          * the new kernel 0-0x100 safely
 242          *
 243          * do this if kexec in setup.c ?
 244          */
 245 #ifdef CONFIG_PPC64
 246         smp_release_cpus();
 247 #else
 248         /* FIXME */
 249 #endif
 250 }
 251
 252 void crash_kexec_secondary(struct pt_regs *regs)
 253 {
 254         cpus_in_sr = CPU_MASK_NONE;
 255 }
 256 #endif
 257 #ifdef CONFIG_SPU_BASE
 258
 259 #include <asm/spu.h>
 260 #include <asm/spu_priv1.h>
 261
 262 struct crash_spu_info {
 263         struct spu *spu;
 264         u32 saved_spu_runcntl_RW;
 265         u32 saved_spu_status_R;
 266         u32 saved_spu_npc_RW;
 267         u64 saved_mfc_sr1_RW;
 268         u64 saved_mfc_dar;
 269         u64 saved_mfc_dsisr;
 270 };
 271
 272 #define CRASH_NUM_SPUS  16      /* Enough for current hardware */
 273 static struct crash_spu_info crash_spu_info[CRASH_NUM_SPUS];
 274
 275 static void crash_kexec_stop_spus(void)
 276 {
 277         struct spu *spu;
 278         int i;
 279         u64 tmp;
 280
 281         for (i = 0; i < CRASH_NUM_SPUS; i++) {
 282                 if (!crash_spu_info[i].spu)
 283                         continue;
 284
 285                 spu = crash_spu_info[i].spu;
 286
 287                 crash_spu_info[i].saved_spu_runcntl_RW =
 288                         in_be32(&spu->problem->spu_runcntl_RW);
 289                 crash_spu_info[i].saved_spu_status_R =
 290                         in_be32(&spu->problem->spu_status_R);
 291                 crash_spu_info[i].saved_spu_npc_RW =
 292                         in_be32(&spu->problem->spu_npc_RW);
 293
 294                 crash_spu_info[i].saved_mfc_dar    = spu_mfc_dar_get(spu);
 295                 crash_spu_info[i].saved_mfc_dsisr  = spu_mfc_dsisr_get(spu);
 296                 tmp = spu_mfc_sr1_get(spu);
 297                 crash_spu_info[i].saved_mfc_sr1_RW = tmp;
 298
 299                 tmp &= ~MFC_STATE1_MASTER_RUN_CONTROL_MASK;
 300                 spu_mfc_sr1_set(spu, tmp);
 301
 302                 __delay(200);
 303         }
 304 }
 305
 306 void crash_register_spus(struct list_head *list)
 307 {
 308         struct spu *spu;
 309
 310         list_for_each_entry(spu, list, full_list) {
 311                 if (WARN_ON(spu->number >= CRASH_NUM_SPUS))
 312                         continue;
 313
 314                 crash_spu_info[spu->number].spu = spu;
 315         }
 316 }
 317
 318 #else
 319 static inline void crash_kexec_stop_spus(void)
 320 {
 321 }
 322 #endif /* CONFIG_SPU_BASE */
 323
 324 /*
 325  * Register a function to be called on shutdown.  Only use this if you
 326  * can't reset your device in the second kernel.
 327  */
 328 int crash_shutdown_register(crash_shutdown_t handler)
 329 {
 330         unsigned int i, rc;
 331
 332         spin_lock(&crash_handlers_lock);
 333         for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
 334                 if (!crash_shutdown_handles[i]) {
 335                         /* Insert handle at first empty entry */
 336                         crash_shutdown_handles[i] = handler;
 337                         rc = 0;
 338                         break;
 339                 }
 340
 341         if (i == CRASH_HANDLER_MAX) {
 342                 printk(KERN_ERR "Crash shutdown handles full, "
 343                        "not registered.\n");
 344                 rc = 1;
 345         }
 346
 347         spin_unlock(&crash_handlers_lock);
 348         return rc;
 349 }
 350 EXPORT_SYMBOL(crash_shutdown_register);
 351
 352 int crash_shutdown_unregister(crash_shutdown_t handler)
 353 {
 354         unsigned int i, rc;
 355
 356         spin_lock(&crash_handlers_lock);
 357         for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
 358                 if (crash_shutdown_handles[i] == handler)
 359                         break;
 360
 361         if (i == CRASH_HANDLER_MAX) {
 362                 printk(KERN_ERR "Crash shutdown handle not found\n");
 363                 rc = 1;
 364         } else {
 365                 /* Shift handles down */
 366                 for (; crash_shutdown_handles[i]; i++)
 367                         crash_shutdown_handles[i] =
 368                                 crash_shutdown_handles[i+1];
 369                 rc = 0;
 370         }
 371
 372         spin_unlock(&crash_handlers_lock);
 373         return rc;
 374 }
 375 EXPORT_SYMBOL(crash_shutdown_unregister);
 376
 377 static unsigned long crash_shutdown_buf[JMP_BUF_LEN];
 378 static int crash_shutdown_cpu = -1;
 379
 380 static int handle_fault(struct pt_regs *regs)
 381 {
 382         if (crash_shutdown_cpu == smp_processor_id())
 383                 longjmp(crash_shutdown_buf, 1);
 384         return 0;
 385 }
 386
 387 void default_machine_crash_shutdown(struct pt_regs *regs)
 388 {
 389         unsigned int i;
 390         int (*old_handler)(struct pt_regs *regs);
 391
 392
 393         /*
 394          * This function is only called after the system
 395          * has panicked or is otherwise in a critical state.
 396          * The minimum amount of code to allow a kexec'd kernel
 397          * to run successfully needs to happen here.
 398          *
 399          * In practice this means stopping other cpus in
 400          * an SMP system.
 401          * The kernel is broken so disable interrupts.
 402          */
 403         hard_irq_disable();
 404
 405         /*
 406          * Make a note of crashing cpu. Will be used in machine_kexec
 407          * such that another IPI will not be sent.
 408          */
 409         crashing_cpu = smp_processor_id();
 410         crash_save_cpu(regs, crashing_cpu);
 411         crash_kexec_prepare_cpus(crashing_cpu);
 412         cpu_set(crashing_cpu, cpus_in_crash);
 413 #if defined(CONFIG_PPC_STD_MMU_64) && defined(CONFIG_SMP)
 414         crash_kexec_wait_realmode(crashing_cpu);
 415 #endif
 416
 417         machine_kexec_mask_interrupts();
 418
 419         /*
 420          * Call registered shutdown routines savely.  Swap out
 421          * __debugger_fault_handler, and replace on exit.
 422          */
 423         old_handler = __debugger_fault_handler;
 424         __debugger_fault_handler = handle_fault;
 425         crash_shutdown_cpu = smp_processor_id();
 426         for (i = 0; crash_shutdown_handles[i]; i++) {
 427                 if (setjmp(crash_shutdown_buf) == 0) {
 428                         /*
 429                          * Insert syncs and delay to ensure
 430                          * instructions in the dangerous region don't
 431                          * leak away from this protected region.
 432                          */
 433                         asm volatile("sync; isync");
 434                         /* dangerous region */
 435                         crash_shutdown_handles[i]();
 436                         asm volatile("sync; isync");
 437                 }
 438         }
 439         crash_shutdown_cpu = -1;
 440         __debugger_fault_handler = old_handler;
 441
 442         crash_kexec_stop_spus();
 443
 444         if (ppc_md.kexec_cpu_down)
 445                 ppc_md.kexec_cpu_down(1, 0);
 446 }