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/irq.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/lmb.h>
  34 #include <asm/firmware.h>
  35 #include <asm/smp.h>
  36
  37 #ifdef DEBUG
  38 #include <asm/udbg.h>
  39 #define DBG(fmt...) udbg_printf(fmt)
  40 #else
  41 #define DBG(fmt...)
  42 #endif
  43
  44 /* This keeps a track of which one is crashing cpu. */
  45 int crashing_cpu = -1;
  46 static cpumask_t cpus_in_crash = CPU_MASK_NONE;
  47 cpumask_t cpus_in_sr = CPU_MASK_NONE;
  48
  49 #ifdef CONFIG_SMP
  50 static atomic_t enter_on_soft_reset = ATOMIC_INIT(0);
  51
  52 void crash_ipi_callback(struct pt_regs *regs)
  53 {
  54         int cpu = smp_processor_id();
  55
  56         if (!cpu_online(cpu))
  57                 return;
  58
  59         hard_irq_disable();
  60         if (!cpu_isset(cpu, cpus_in_crash))
  61                 crash_save_cpu(regs, cpu);
  62         cpu_set(cpu, cpus_in_crash);
  63
  64         /*
  65          * Entered via soft-reset - could be the kdump
  66          * process is invoked using soft-reset or user activated
  67          * it if some CPU did not respond to an IPI.
  68          * For soft-reset, the secondary CPU can enter this func
  69          * twice. 1 - using IPI, and 2. soft-reset.
  70          * Tell the kexec CPU that entered via soft-reset and ready
  71          * to go down.
  72          */
  73         if (cpu_isset(cpu, cpus_in_sr)) {
  74                 cpu_clear(cpu, cpus_in_sr);
  75                 atomic_inc(&enter_on_soft_reset);
  76         }
  77
  78         /*
  79          * Starting the kdump boot.
  80          * This barrier is needed to make sure that all CPUs are stopped.
  81          * If not, soft-reset will be invoked to bring other CPUs.
  82          */
  83         while (!cpu_isset(crashing_cpu, cpus_in_crash))
  84                 cpu_relax();
  85
  86         if (ppc_md.kexec_cpu_down)
  87                 ppc_md.kexec_cpu_down(1, 1);
  88
  89 #ifdef CONFIG_PPC64
  90         kexec_smp_wait();
  91 #else
  92         for (;;);       /* FIXME */
  93 #endif
  94
  95         /* NOTREACHED */
  96 }
  97
  98 /*
  99  * Wait until all CPUs are entered via soft-reset.
 100  */
 101 static void crash_soft_reset_check(int cpu)
 102 {
 103         unsigned int ncpus = num_online_cpus() - 1;/* Excluding the panic cpu */
 104
 105         cpu_clear(cpu, cpus_in_sr);
 106         while (atomic_read(&enter_on_soft_reset) != ncpus)
 107                 cpu_relax();
 108 }
 109
 110
 111 static void crash_kexec_prepare_cpus(int cpu)
 112 {
 113         unsigned int msecs;
 114
 115         unsigned int ncpus = num_online_cpus() - 1;/* Excluding the panic cpu */
 116
 117         crash_send_ipi(crash_ipi_callback);
 118         smp_wmb();
 119
 120         /*
 121          * FIXME: Until we will have the way to stop other CPUSs reliabally,
 122          * the crash CPU will send an IPI and wait for other CPUs to
 123          * respond.
 124          * Delay of at least 10 seconds.
 125          */
 126         printk(KERN_EMERG "Sending IPI to other cpus...\n");
 127         msecs = 10000;
 128         while ((cpus_weight(cpus_in_crash) < ncpus) && (--msecs > 0)) {
 129                 cpu_relax();
 130                 mdelay(1);
 131         }
 132
 133         /* Would it be better to replace the trap vector here? */
 134
 135         /*
 136          * FIXME: In case if we do not get all CPUs, one possibility: ask the
 137          * user to do soft reset such that we get all.
 138          * Soft-reset will be used until better mechanism is implemented.
 139          */
 140         if (cpus_weight(cpus_in_crash) < ncpus) {
 141                 printk(KERN_EMERG "done waiting: %d cpu(s) not responding\n",
 142                         ncpus - cpus_weight(cpus_in_crash));
 143                 printk(KERN_EMERG "Activate soft-reset to stop other cpu(s)\n");
 144                 cpus_in_sr = CPU_MASK_NONE;
 145                 atomic_set(&enter_on_soft_reset, 0);
 146                 while (cpus_weight(cpus_in_crash) < ncpus)
 147                         cpu_relax();
 148         }
 149         /*
 150          * Make sure all CPUs are entered via soft-reset if the kdump is
 151          * invoked using soft-reset.
 152          */
 153         if (cpu_isset(cpu, cpus_in_sr))
 154                 crash_soft_reset_check(cpu);
 155         /* Leave the IPI callback set */
 156 }
 157
 158 /*
 159  * This function will be called by secondary cpus or by kexec cpu
 160  * if soft-reset is activated to stop some CPUs.
 161  */
 162 void crash_kexec_secondary(struct pt_regs *regs)
 163 {
 164         int cpu = smp_processor_id();
 165         unsigned long flags;
 166         int msecs = 5;
 167
 168         local_irq_save(flags);
 169         /* Wait 5ms if the kexec CPU is not entered yet. */
 170         while (crashing_cpu < 0) {
 171                 if (--msecs < 0) {
 172                         /*
 173                          * Either kdump image is not loaded or
 174                          * kdump process is not started - Probably xmon
 175                          * exited using 'x'(exit and recover) or
 176                          * kexec_should_crash() failed for all running tasks.
 177                          */
 178                         cpu_clear(cpu, cpus_in_sr);
 179                         local_irq_restore(flags);
 180                         return;
 181                 }
 182                 mdelay(1);
 183                 cpu_relax();
 184         }
 185         if (cpu == crashing_cpu) {
 186                 /*
 187                  * Panic CPU will enter this func only via soft-reset.
 188                  * Wait until all secondary CPUs entered and
 189                  * then start kexec boot.
 190                  */
 191                 crash_soft_reset_check(cpu);
 192                 cpu_set(crashing_cpu, cpus_in_crash);
 193                 if (ppc_md.kexec_cpu_down)
 194                         ppc_md.kexec_cpu_down(1, 0);
 195                 machine_kexec(kexec_crash_image);
 196                 /* NOTREACHED */
 197         }
 198         crash_ipi_callback(regs);
 199 }
 200
 201 #else
 202 static void crash_kexec_prepare_cpus(int cpu)
 203 {
 204         /*
 205          * move the secondarys to us so that we can copy
 206          * the new kernel 0-0x100 safely
 207          *
 208          * do this if kexec in setup.c ?
 209          */
 210 #ifdef CONFIG_PPC64
 211         smp_release_cpus();
 212 #else
 213         /* FIXME */
 214 #endif
 215 }
 216
 217 void crash_kexec_secondary(struct pt_regs *regs)
 218 {
 219         cpus_in_sr = CPU_MASK_NONE;
 220 }
 221 #endif
 222 #ifdef CONFIG_SPU_BASE
 223
 224 #include <asm/spu.h>
 225 #include <asm/spu_priv1.h>
 226
 227 struct crash_spu_info {
 228         struct spu *spu;
 229         u32 saved_spu_runcntl_RW;
 230         u32 saved_spu_status_R;
 231         u32 saved_spu_npc_RW;
 232         u64 saved_mfc_sr1_RW;
 233         u64 saved_mfc_dar;
 234         u64 saved_mfc_dsisr;
 235 };
 236
 237 #define CRASH_NUM_SPUS  16      /* Enough for current hardware */
 238 static struct crash_spu_info crash_spu_info[CRASH_NUM_SPUS];
 239
 240 static void crash_kexec_stop_spus(void)
 241 {
 242         struct spu *spu;
 243         int i;
 244         u64 tmp;
 245
 246         for (i = 0; i < CRASH_NUM_SPUS; i++) {
 247                 if (!crash_spu_info[i].spu)
 248                         continue;
 249
 250                 spu = crash_spu_info[i].spu;
 251
 252                 crash_spu_info[i].saved_spu_runcntl_RW =
 253                         in_be32(&spu->problem->spu_runcntl_RW);
 254                 crash_spu_info[i].saved_spu_status_R =
 255                         in_be32(&spu->problem->spu_status_R);
 256                 crash_spu_info[i].saved_spu_npc_RW =
 257                         in_be32(&spu->problem->spu_npc_RW);
 258
 259                 crash_spu_info[i].saved_mfc_dar    = spu_mfc_dar_get(spu);
 260                 crash_spu_info[i].saved_mfc_dsisr  = spu_mfc_dsisr_get(spu);
 261                 tmp = spu_mfc_sr1_get(spu);
 262                 crash_spu_info[i].saved_mfc_sr1_RW = tmp;
 263
 264                 tmp &= ~MFC_STATE1_MASTER_RUN_CONTROL_MASK;
 265                 spu_mfc_sr1_set(spu, tmp);
 266
 267                 __delay(200);
 268         }
 269 }
 270
 271 void crash_register_spus(struct list_head *list)
 272 {
 273         struct spu *spu;
 274
 275         list_for_each_entry(spu, list, full_list) {
 276                 if (WARN_ON(spu->number >= CRASH_NUM_SPUS))
 277                         continue;
 278
 279                 crash_spu_info[spu->number].spu = spu;
 280         }
 281 }
 282
 283 #else
 284 static inline void crash_kexec_stop_spus(void)
 285 {
 286 }
 287 #endif /* CONFIG_SPU_BASE */
 288
 289 void default_machine_crash_shutdown(struct pt_regs *regs)
 290 {
 291         unsigned int irq;
 292
 293         /*
 294          * This function is only called after the system
 295          * has panicked or is otherwise in a critical state.
 296          * The minimum amount of code to allow a kexec'd kernel
 297          * to run successfully needs to happen here.
 298          *
 299          * In practice this means stopping other cpus in
 300          * an SMP system.
 301          * The kernel is broken so disable interrupts.
 302          */
 303         hard_irq_disable();
 304
 305         for_each_irq(irq) {
 306                 struct irq_desc *desc = irq_desc + irq;
 307
 308                 if (desc->status & IRQ_INPROGRESS)
 309                         desc->chip->eoi(irq);
 310
 311                 if (!(desc->status & IRQ_DISABLED))
 312                         desc->chip->disable(irq);
 313         }
 314
 315         /*
 316          * Make a note of crashing cpu. Will be used in machine_kexec
 317          * such that another IPI will not be sent.
 318          */
 319         crashing_cpu = smp_processor_id();
 320         crash_save_cpu(regs, crashing_cpu);
 321         crash_kexec_prepare_cpus(crashing_cpu);
 322         cpu_set(crashing_cpu, cpus_in_crash);
 323         crash_kexec_stop_spus();
 324         if (ppc_md.kexec_cpu_down)
 325                 ppc_md.kexec_cpu_down(1, 0);
 326 }