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/lmb.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 /*
 166  * This function will be called by secondary cpus or by kexec cpu
 167  * if soft-reset is activated to stop some CPUs.
 168  */
 169 void crash_kexec_secondary(struct pt_regs *regs)
 170 {
 171         int cpu = smp_processor_id();
 172         unsigned long flags;
 173         int msecs = 5;
 174
 175         local_irq_save(flags);
 176         /* Wait 5ms if the kexec CPU is not entered yet. */
 177         while (crashing_cpu < 0) {
 178                 if (--msecs < 0) {
 179                         /*
 180                          * Either kdump image is not loaded or
 181                          * kdump process is not started - Probably xmon
 182                          * exited using 'x'(exit and recover) or
 183                          * kexec_should_crash() failed for all running tasks.
 184                          */
 185                         cpu_clear(cpu, cpus_in_sr);
 186                         local_irq_restore(flags);
 187                         return;
 188                 }
 189                 mdelay(1);
 190                 cpu_relax();
 191         }
 192         if (cpu == crashing_cpu) {
 193                 /*
 194                  * Panic CPU will enter this func only via soft-reset.
 195                  * Wait until all secondary CPUs entered and
 196                  * then start kexec boot.
 197                  */
 198                 crash_soft_reset_check(cpu);
 199                 cpu_set(crashing_cpu, cpus_in_crash);
 200                 if (ppc_md.kexec_cpu_down)
 201                         ppc_md.kexec_cpu_down(1, 0);
 202                 machine_kexec(kexec_crash_image);
 203                 /* NOTREACHED */
 204         }
 205         crash_ipi_callback(regs);
 206 }
 207
 208 #else
 209 static void crash_kexec_prepare_cpus(int cpu)
 210 {
 211         /*
 212          * move the secondarys 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         cpus_in_sr = CPU_MASK_NONE;
 227 }
 228 #endif
 229 #ifdef CONFIG_SPU_BASE
 230
 231 #include <asm/spu.h>
 232 #include <asm/spu_priv1.h>
 233
 234 struct crash_spu_info {
 235         struct spu *spu;
 236         u32 saved_spu_runcntl_RW;
 237         u32 saved_spu_status_R;
 238         u32 saved_spu_npc_RW;
 239         u64 saved_mfc_sr1_RW;
 240         u64 saved_mfc_dar;
 241         u64 saved_mfc_dsisr;
 242 };
 243
 244 #define CRASH_NUM_SPUS  16      /* Enough for current hardware */
 245 static struct crash_spu_info crash_spu_info[CRASH_NUM_SPUS];
 246
 247 static void crash_kexec_stop_spus(void)
 248 {
 249         struct spu *spu;
 250         int i;
 251         u64 tmp;
 252
 253         for (i = 0; i < CRASH_NUM_SPUS; i++) {
 254                 if (!crash_spu_info[i].spu)
 255                         continue;
 256
 257                 spu = crash_spu_info[i].spu;
 258
 259                 crash_spu_info[i].saved_spu_runcntl_RW =
 260                         in_be32(&spu->problem->spu_runcntl_RW);
 261                 crash_spu_info[i].saved_spu_status_R =
 262                         in_be32(&spu->problem->spu_status_R);
 263                 crash_spu_info[i].saved_spu_npc_RW =
 264                         in_be32(&spu->problem->spu_npc_RW);
 265
 266                 crash_spu_info[i].saved_mfc_dar    = spu_mfc_dar_get(spu);
 267                 crash_spu_info[i].saved_mfc_dsisr  = spu_mfc_dsisr_get(spu);
 268                 tmp = spu_mfc_sr1_get(spu);
 269                 crash_spu_info[i].saved_mfc_sr1_RW = tmp;
 270
 271                 tmp &= ~MFC_STATE1_MASTER_RUN_CONTROL_MASK;
 272                 spu_mfc_sr1_set(spu, tmp);
 273
 274                 __delay(200);
 275         }
 276 }
 277
 278 void crash_register_spus(struct list_head *list)
 279 {
 280         struct spu *spu;
 281
 282         list_for_each_entry(spu, list, full_list) {
 283                 if (WARN_ON(spu->number >= CRASH_NUM_SPUS))
 284                         continue;
 285
 286                 crash_spu_info[spu->number].spu = spu;
 287         }
 288 }
 289
 290 #else
 291 static inline void crash_kexec_stop_spus(void)
 292 {
 293 }
 294 #endif /* CONFIG_SPU_BASE */
 295
 296 /*
 297  * Register a function to be called on shutdown.  Only use this if you
 298  * can't reset your device in the second kernel.
 299  */
 300 int crash_shutdown_register(crash_shutdown_t handler)
 301 {
 302         unsigned int i, rc;
 303
 304         spin_lock(&crash_handlers_lock);
 305         for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
 306                 if (!crash_shutdown_handles[i]) {
 307                         /* Insert handle at first empty entry */
 308                         crash_shutdown_handles[i] = handler;
 309                         rc = 0;
 310                         break;
 311                 }
 312
 313         if (i == CRASH_HANDLER_MAX) {
 314                 printk(KERN_ERR "Crash shutdown handles full, "
 315                        "not registered.\n");
 316                 rc = 1;
 317         }
 318
 319         spin_unlock(&crash_handlers_lock);
 320         return rc;
 321 }
 322 EXPORT_SYMBOL(crash_shutdown_register);
 323
 324 int crash_shutdown_unregister(crash_shutdown_t handler)
 325 {
 326         unsigned int i, rc;
 327
 328         spin_lock(&crash_handlers_lock);
 329         for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
 330                 if (crash_shutdown_handles[i] == handler)
 331                         break;
 332
 333         if (i == CRASH_HANDLER_MAX) {
 334                 printk(KERN_ERR "Crash shutdown handle not found\n");
 335                 rc = 1;
 336         } else {
 337                 /* Shift handles down */
 338                 for (; crash_shutdown_handles[i]; i++)
 339                         crash_shutdown_handles[i] =
 340                                 crash_shutdown_handles[i+1];
 341                 rc = 0;
 342         }
 343
 344         spin_unlock(&crash_handlers_lock);
 345         return rc;
 346 }
 347 EXPORT_SYMBOL(crash_shutdown_unregister);
 348
 349 static unsigned long crash_shutdown_buf[JMP_BUF_LEN];
 350
 351 static int handle_fault(struct pt_regs *regs)
 352 {
 353         longjmp(crash_shutdown_buf, 1);
 354         return 0;
 355 }
 356
 357 void default_machine_crash_shutdown(struct pt_regs *regs)
 358 {
 359         unsigned int i;
 360         int (*old_handler)(struct pt_regs *regs);
 361
 362
 363         /*
 364          * This function is only called after the system
 365          * has panicked or is otherwise in a critical state.
 366          * The minimum amount of code to allow a kexec'd kernel
 367          * to run successfully needs to happen here.
 368          *
 369          * In practice this means stopping other cpus in
 370          * an SMP system.
 371          * The kernel is broken so disable interrupts.
 372          */
 373         hard_irq_disable();
 374
 375         for_each_irq(i) {
 376                 struct irq_desc *desc = irq_desc + i;
 377
 378                 if (desc->status & IRQ_INPROGRESS)
 379                         desc->chip->eoi(i);
 380
 381                 if (!(desc->status & IRQ_DISABLED))
 382                         desc->chip->disable(i);
 383         }
 384
 385         /*
 386          * Call registered shutdown routines savely.  Swap out
 387          * __debugger_fault_handler, and replace on exit.
 388          */
 389         old_handler = __debugger_fault_handler;
 390         __debugger_fault_handler = handle_fault;
 391         for (i = 0; crash_shutdown_handles[i]; i++) {
 392                 if (setjmp(crash_shutdown_buf) == 0) {
 393                         /*
 394                          * Insert syncs and delay to ensure
 395                          * instructions in the dangerous region don't
 396                          * leak away from this protected region.
 397                          */
 398                         asm volatile("sync; isync");
 399                         /* dangerous region */
 400                         crash_shutdown_handles[i]();
 401                         asm volatile("sync; isync");
 402                 }
 403         }
 404         __debugger_fault_handler = old_handler;
 405
 406         /*
 407          * Make a note of crashing cpu. Will be used in machine_kexec
 408          * such that another IPI will not be sent.
 409          */
 410         crashing_cpu = smp_processor_id();
 411         crash_save_cpu(regs, crashing_cpu);
 412         crash_kexec_prepare_cpus(crashing_cpu);
 413         cpu_set(crashing_cpu, cpus_in_crash);
 414         crash_kexec_stop_spus();
 415         if (ppc_md.kexec_cpu_down)
 416                 ppc_md.kexec_cpu_down(1, 0);
 417 }