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
  28 #include <asm/processor.h>
  29 #include <asm/machdep.h>
  30 #include <asm/kexec.h>
  31 #include <asm/kdump.h>
  32 #include <asm/lmb.h>
  33 #include <asm/firmware.h>
  34 #include <asm/smp.h>
  35
  36 #ifdef DEBUG
  37 #include <asm/udbg.h>
  38 #define DBG(fmt...) udbg_printf(fmt)
  39 #else
  40 #define DBG(fmt...)
  41 #endif
  42
  43 /* This keeps a track of which one is crashing cpu. */
  44 int crashing_cpu = -1;
  45 static cpumask_t cpus_in_crash = CPU_MASK_NONE;
  46 cpumask_t cpus_in_sr = CPU_MASK_NONE;
  47
  48 #ifdef CONFIG_SMP
  49 static atomic_t enter_on_soft_reset = ATOMIC_INIT(0);
  50
  51 void crash_ipi_callback(struct pt_regs *regs)
  52 {
  53         int cpu = smp_processor_id();
  54
  55         if (!cpu_online(cpu))
  56                 return;
  57
  58         hard_irq_disable();
  59         if (!cpu_isset(cpu, cpus_in_crash))
  60                 crash_save_cpu(regs, cpu);
  61         cpu_set(cpu, cpus_in_crash);
  62
  63         /*
  64          * Entered via soft-reset - could be the kdump
  65          * process is invoked using soft-reset or user activated
  66          * it if some CPU did not respond to an IPI.
  67          * For soft-reset, the secondary CPU can enter this func
  68          * twice. 1 - using IPI, and 2. soft-reset.
  69          * Tell the kexec CPU that entered via soft-reset and ready
  70          * to go down.
  71          */
  72         if (cpu_isset(cpu, cpus_in_sr)) {
  73                 cpu_clear(cpu, cpus_in_sr);
  74                 atomic_inc(&enter_on_soft_reset);
  75         }
  76
  77         /*
  78          * Starting the kdump boot.
  79          * This barrier is needed to make sure that all CPUs are stopped.
  80          * If not, soft-reset will be invoked to bring other CPUs.
  81          */
  82         while (!cpu_isset(crashing_cpu, cpus_in_crash))
  83                 cpu_relax();
  84
  85         if (ppc_md.kexec_cpu_down)
  86                 ppc_md.kexec_cpu_down(1, 1);
  87
  88 #ifdef CONFIG_PPC64
  89         kexec_smp_wait();
  90 #else
  91         for (;;);       /* FIXME */
  92 #endif
  93
  94         /* NOTREACHED */
  95 }
  96
  97 /*
  98  * Wait until all CPUs are entered via soft-reset.
  99  */
 100 static void crash_soft_reset_check(int cpu)
 101 {
 102         unsigned int ncpus = num_online_cpus() - 1;/* Excluding the panic cpu */
 103
 104         cpu_clear(cpu, cpus_in_sr);
 105         while (atomic_read(&enter_on_soft_reset) != ncpus)
 106                 cpu_relax();
 107 }
 108
 109
 110 static void crash_kexec_prepare_cpus(int cpu)
 111 {
 112         unsigned int msecs;
 113
 114         unsigned int ncpus = num_online_cpus() - 1;/* Excluding the panic cpu */
 115
 116         crash_send_ipi(crash_ipi_callback);
 117         smp_wmb();
 118
 119         /*
 120          * FIXME: Until we will have the way to stop other CPUSs reliabally,
 121          * the crash CPU will send an IPI and wait for other CPUs to
 122          * respond.
 123          * Delay of at least 10 seconds.
 124          */
 125         printk(KERN_EMERG "Sending IPI to other cpus...\n");
 126         msecs = 10000;
 127         while ((cpus_weight(cpus_in_crash) < ncpus) && (--msecs > 0)) {
 128                 cpu_relax();
 129                 mdelay(1);
 130         }
 131
 132         /* Would it be better to replace the trap vector here? */
 133
 134         /*
 135          * FIXME: In case if we do not get all CPUs, one possibility: ask the
 136          * user to do soft reset such that we get all.
 137          * Soft-reset will be used until better mechanism is implemented.
 138          */
 139         if (cpus_weight(cpus_in_crash) < ncpus) {
 140                 printk(KERN_EMERG "done waiting: %d cpu(s) not responding\n",
 141                         ncpus - cpus_weight(cpus_in_crash));
 142                 printk(KERN_EMERG "Activate soft-reset to stop other cpu(s)\n");
 143                 cpus_in_sr = CPU_MASK_NONE;
 144                 atomic_set(&enter_on_soft_reset, 0);
 145                 while (cpus_weight(cpus_in_crash) < ncpus)
 146                         cpu_relax();
 147         }
 148         /*
 149          * Make sure all CPUs are entered via soft-reset if the kdump is
 150          * invoked using soft-reset.
 151          */
 152         if (cpu_isset(cpu, cpus_in_sr))
 153                 crash_soft_reset_check(cpu);
 154         /* Leave the IPI callback set */
 155 }
 156
 157 /*
 158  * This function will be called by secondary cpus or by kexec cpu
 159  * if soft-reset is activated to stop some CPUs.
 160  */
 161 void crash_kexec_secondary(struct pt_regs *regs)
 162 {
 163         int cpu = smp_processor_id();
 164         unsigned long flags;
 165         int msecs = 5;
 166
 167         local_irq_save(flags);
 168         /* Wait 5ms if the kexec CPU is not entered yet. */
 169         while (crashing_cpu < 0) {
 170                 if (--msecs < 0) {
 171                         /*
 172                          * Either kdump image is not loaded or
 173                          * kdump process is not started - Probably xmon
 174                          * exited using 'x'(exit and recover) or
 175                          * kexec_should_crash() failed for all running tasks.
 176                          */
 177                         cpu_clear(cpu, cpus_in_sr);
 178                         local_irq_restore(flags);
 179                         return;
 180                 }
 181                 mdelay(1);
 182                 cpu_relax();
 183         }
 184         if (cpu == crashing_cpu) {
 185                 /*
 186                  * Panic CPU will enter this func only via soft-reset.
 187                  * Wait until all secondary CPUs entered and
 188                  * then start kexec boot.
 189                  */
 190                 crash_soft_reset_check(cpu);
 191                 cpu_set(crashing_cpu, cpus_in_crash);
 192                 if (ppc_md.kexec_cpu_down)
 193                         ppc_md.kexec_cpu_down(1, 0);
 194                 machine_kexec(kexec_crash_image);
 195                 /* NOTREACHED */
 196         }
 197         crash_ipi_callback(regs);
 198 }
 199
 200 #else
 201 static void crash_kexec_prepare_cpus(int cpu)
 202 {
 203         /*
 204          * move the secondarys to us so that we can copy
 205          * the new kernel 0-0x100 safely
 206          *
 207          * do this if kexec in setup.c ?
 208          */
 209 #ifdef CONFIG_PPC64
 210         smp_release_cpus();
 211 #else
 212         /* FIXME */
 213 #endif
 214 }
 215
 216 void crash_kexec_secondary(struct pt_regs *regs)
 217 {
 218         cpus_in_sr = CPU_MASK_NONE;
 219 }
 220 #endif
 221 #ifdef CONFIG_SPU_BASE
 222
 223 #include <asm/spu.h>
 224 #include <asm/spu_priv1.h>
 225
 226 struct crash_spu_info {
 227         struct spu *spu;
 228         u32 saved_spu_runcntl_RW;
 229         u32 saved_spu_status_R;
 230         u32 saved_spu_npc_RW;
 231         u64 saved_mfc_sr1_RW;
 232         u64 saved_mfc_dar;
 233         u64 saved_mfc_dsisr;
 234 };
 235
 236 #define CRASH_NUM_SPUS  16      /* Enough for current hardware */
 237 static struct crash_spu_info crash_spu_info[CRASH_NUM_SPUS];
 238
 239 static void crash_kexec_stop_spus(void)
 240 {
 241         struct spu *spu;
 242         int i;
 243         u64 tmp;
 244
 245         for (i = 0; i < CRASH_NUM_SPUS; i++) {
 246                 if (!crash_spu_info[i].spu)
 247                         continue;
 248
 249                 spu = crash_spu_info[i].spu;
 250
 251                 crash_spu_info[i].saved_spu_runcntl_RW =
 252                         in_be32(&spu->problem->spu_runcntl_RW);
 253                 crash_spu_info[i].saved_spu_status_R =
 254                         in_be32(&spu->problem->spu_status_R);
 255                 crash_spu_info[i].saved_spu_npc_RW =
 256                         in_be32(&spu->problem->spu_npc_RW);
 257
 258                 crash_spu_info[i].saved_mfc_dar    = spu_mfc_dar_get(spu);
 259                 crash_spu_info[i].saved_mfc_dsisr  = spu_mfc_dsisr_get(spu);
 260                 tmp = spu_mfc_sr1_get(spu);
 261                 crash_spu_info[i].saved_mfc_sr1_RW = tmp;
 262
 263                 tmp &= ~MFC_STATE1_MASTER_RUN_CONTROL_MASK;
 264                 spu_mfc_sr1_set(spu, tmp);
 265
 266                 __delay(200);
 267         }
 268 }
 269
 270 void crash_register_spus(struct list_head *list)
 271 {
 272         struct spu *spu;
 273
 274         list_for_each_entry(spu, list, full_list) {
 275                 if (WARN_ON(spu->number >= CRASH_NUM_SPUS))
 276                         continue;
 277
 278                 crash_spu_info[spu->number].spu = spu;
 279         }
 280 }
 281
 282 #else
 283 static inline void crash_kexec_stop_spus(void)
 284 {
 285 }
 286 #endif /* CONFIG_SPU_BASE */
 287
 288 void default_machine_crash_shutdown(struct pt_regs *regs)
 289 {
 290         unsigned int irq;
 291
 292         /*
 293          * This function is only called after the system
 294          * has panicked or is otherwise in a critical state.
 295          * The minimum amount of code to allow a kexec'd kernel
 296          * to run successfully needs to happen here.
 297          *
 298          * In practice this means stopping other cpus in
 299          * an SMP system.
 300          * The kernel is broken so disable interrupts.
 301          */
 302         hard_irq_disable();
 303
 304         for_each_irq(irq) {
 305                 struct irq_desc *desc = irq_desc + irq;
 306
 307                 if (desc->status & IRQ_INPROGRESS)
 308                         desc->chip->eoi(irq);
 309
 310                 if (!(desc->status & IRQ_DISABLED))
 311                         desc->chip->disable(irq);
 312         }
 313
 314         /*
 315          * Make a note of crashing cpu. Will be used in machine_kexec
 316          * such that another IPI will not be sent.
 317          */
 318         crashing_cpu = smp_processor_id();
 319         crash_save_cpu(regs, crashing_cpu);
 320         crash_kexec_prepare_cpus(crashing_cpu);
 321         cpu_set(crashing_cpu, cpus_in_crash);
 322         crash_kexec_stop_spus();
 323         if (ppc_md.kexec_cpu_down)
 324                 ppc_md.kexec_cpu_down(1, 0);
 325 }