| 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 | static u32 *append_elf_note(u32 *buf, char *name, unsigned type, void *data, |
| 50 | size_t data_len) |
| 51 | { |
| 52 | struct elf_note note; |
| 53 | |
| 54 | note.n_namesz = strlen(name) + 1; |
| 55 | note.n_descsz = data_len; |
| 56 | note.n_type = type; |
| 57 | memcpy(buf, ¬e, sizeof(note)); |
| 58 | buf += (sizeof(note) +3)/4; |
| 59 | memcpy(buf, name, note.n_namesz); |
| 60 | buf += (note.n_namesz + 3)/4; |
| 61 | memcpy(buf, data, note.n_descsz); |
| 62 | buf += (note.n_descsz + 3)/4; |
| 63 | |
| 64 | return buf; |
| 65 | } |
| 66 | |
| 67 | static void final_note(u32 *buf) |
| 68 | { |
| 69 | struct elf_note note; |
| 70 | |
| 71 | note.n_namesz = 0; |
| 72 | note.n_descsz = 0; |
| 73 | note.n_type = 0; |
| 74 | memcpy(buf, ¬e, sizeof(note)); |
| 75 | } |
| 76 | |
| 77 | static void crash_save_this_cpu(struct pt_regs *regs, int cpu) |
| 78 | { |
| 79 | struct elf_prstatus prstatus; |
| 80 | u32 *buf; |
| 81 | |
| 82 | if ((cpu < 0) || (cpu >= NR_CPUS)) |
| 83 | return; |
| 84 | |
| 85 | /* Using ELF notes here is opportunistic. |
| 86 | * I need a well defined structure format |
| 87 | * for the data I pass, and I need tags |
| 88 | * on the data to indicate what information I have |
| 89 | * squirrelled away. ELF notes happen to provide |
| 90 | * all of that that no need to invent something new. |
| 91 | */ |
| 92 | buf = (u32*)per_cpu_ptr(crash_notes, cpu); |
| 93 | if (!buf) |
| 94 | return; |
| 95 | |
| 96 | memset(&prstatus, 0, sizeof(prstatus)); |
| 97 | prstatus.pr_pid = current->pid; |
| 98 | elf_core_copy_regs(&prstatus.pr_reg, regs); |
| 99 | buf = append_elf_note(buf, "CORE", NT_PRSTATUS, &prstatus, |
| 100 | sizeof(prstatus)); |
| 101 | final_note(buf); |
| 102 | } |
| 103 | |
| 104 | #ifdef CONFIG_SMP |
| 105 | static atomic_t enter_on_soft_reset = ATOMIC_INIT(0); |
| 106 | |
| 107 | void crash_ipi_callback(struct pt_regs *regs) |
| 108 | { |
| 109 | int cpu = smp_processor_id(); |
| 110 | |
| 111 | if (!cpu_online(cpu)) |
| 112 | return; |
| 113 | |
| 114 | hard_irq_disable(); |
| 115 | if (!cpu_isset(cpu, cpus_in_crash)) |
| 116 | crash_save_this_cpu(regs, cpu); |
| 117 | cpu_set(cpu, cpus_in_crash); |
| 118 | |
| 119 | /* |
| 120 | * Entered via soft-reset - could be the kdump |
| 121 | * process is invoked using soft-reset or user activated |
| 122 | * it if some CPU did not respond to an IPI. |
| 123 | * For soft-reset, the secondary CPU can enter this func |
| 124 | * twice. 1 - using IPI, and 2. soft-reset. |
| 125 | * Tell the kexec CPU that entered via soft-reset and ready |
| 126 | * to go down. |
| 127 | */ |
| 128 | if (cpu_isset(cpu, cpus_in_sr)) { |
| 129 | cpu_clear(cpu, cpus_in_sr); |
| 130 | atomic_inc(&enter_on_soft_reset); |
| 131 | } |
| 132 | |
| 133 | /* |
| 134 | * Starting the kdump boot. |
| 135 | * This barrier is needed to make sure that all CPUs are stopped. |
| 136 | * If not, soft-reset will be invoked to bring other CPUs. |
| 137 | */ |
| 138 | while (!cpu_isset(crashing_cpu, cpus_in_crash)) |
| 139 | cpu_relax(); |
| 140 | |
| 141 | if (ppc_md.kexec_cpu_down) |
| 142 | ppc_md.kexec_cpu_down(1, 1); |
| 143 | |
| 144 | #ifdef CONFIG_PPC64 |
| 145 | kexec_smp_wait(); |
| 146 | #else |
| 147 | for (;;); /* FIXME */ |
| 148 | #endif |
| 149 | |
| 150 | /* NOTREACHED */ |
| 151 | } |
| 152 | |
| 153 | /* |
| 154 | * Wait until all CPUs are entered via soft-reset. |
| 155 | */ |
| 156 | static void crash_soft_reset_check(int cpu) |
| 157 | { |
| 158 | unsigned int ncpus = num_online_cpus() - 1;/* Excluding the panic cpu */ |
| 159 | |
| 160 | cpu_clear(cpu, cpus_in_sr); |
| 161 | while (atomic_read(&enter_on_soft_reset) != ncpus) |
| 162 | cpu_relax(); |
| 163 | } |
| 164 | |
| 165 | |
| 166 | static void crash_kexec_prepare_cpus(int cpu) |
| 167 | { |
| 168 | unsigned int msecs; |
| 169 | |
| 170 | unsigned int ncpus = num_online_cpus() - 1;/* Excluding the panic cpu */ |
| 171 | |
| 172 | crash_send_ipi(crash_ipi_callback); |
| 173 | smp_wmb(); |
| 174 | |
| 175 | /* |
| 176 | * FIXME: Until we will have the way to stop other CPUSs reliabally, |
| 177 | * the crash CPU will send an IPI and wait for other CPUs to |
| 178 | * respond. |
| 179 | * Delay of at least 10 seconds. |
| 180 | */ |
| 181 | printk(KERN_EMERG "Sending IPI to other cpus...\n"); |
| 182 | msecs = 10000; |
| 183 | while ((cpus_weight(cpus_in_crash) < ncpus) && (--msecs > 0)) { |
| 184 | cpu_relax(); |
| 185 | mdelay(1); |
| 186 | } |
| 187 | |
| 188 | /* Would it be better to replace the trap vector here? */ |
| 189 | |
| 190 | /* |
| 191 | * FIXME: In case if we do not get all CPUs, one possibility: ask the |
| 192 | * user to do soft reset such that we get all. |
| 193 | * Soft-reset will be used until better mechanism is implemented. |
| 194 | */ |
| 195 | if (cpus_weight(cpus_in_crash) < ncpus) { |
| 196 | printk(KERN_EMERG "done waiting: %d cpu(s) not responding\n", |
| 197 | ncpus - cpus_weight(cpus_in_crash)); |
| 198 | printk(KERN_EMERG "Activate soft-reset to stop other cpu(s)\n"); |
| 199 | cpus_in_sr = CPU_MASK_NONE; |
| 200 | atomic_set(&enter_on_soft_reset, 0); |
| 201 | while (cpus_weight(cpus_in_crash) < ncpus) |
| 202 | cpu_relax(); |
| 203 | } |
| 204 | /* |
| 205 | * Make sure all CPUs are entered via soft-reset if the kdump is |
| 206 | * invoked using soft-reset. |
| 207 | */ |
| 208 | if (cpu_isset(cpu, cpus_in_sr)) |
| 209 | crash_soft_reset_check(cpu); |
| 210 | /* Leave the IPI callback set */ |
| 211 | } |
| 212 | |
| 213 | /* |
| 214 | * This function will be called by secondary cpus or by kexec cpu |
| 215 | * if soft-reset is activated to stop some CPUs. |
| 216 | */ |
| 217 | void crash_kexec_secondary(struct pt_regs *regs) |
| 218 | { |
| 219 | int cpu = smp_processor_id(); |
| 220 | unsigned long flags; |
| 221 | int msecs = 5; |
| 222 | |
| 223 | local_irq_save(flags); |
| 224 | /* Wait 5ms if the kexec CPU is not entered yet. */ |
| 225 | while (crashing_cpu < 0) { |
| 226 | if (--msecs < 0) { |
| 227 | /* |
| 228 | * Either kdump image is not loaded or |
| 229 | * kdump process is not started - Probably xmon |
| 230 | * exited using 'x'(exit and recover) or |
| 231 | * kexec_should_crash() failed for all running tasks. |
| 232 | */ |
| 233 | cpu_clear(cpu, cpus_in_sr); |
| 234 | local_irq_restore(flags); |
| 235 | return; |
| 236 | } |
| 237 | mdelay(1); |
| 238 | cpu_relax(); |
| 239 | } |
| 240 | if (cpu == crashing_cpu) { |
| 241 | /* |
| 242 | * Panic CPU will enter this func only via soft-reset. |
| 243 | * Wait until all secondary CPUs entered and |
| 244 | * then start kexec boot. |
| 245 | */ |
| 246 | crash_soft_reset_check(cpu); |
| 247 | cpu_set(crashing_cpu, cpus_in_crash); |
| 248 | if (ppc_md.kexec_cpu_down) |
| 249 | ppc_md.kexec_cpu_down(1, 0); |
| 250 | machine_kexec(kexec_crash_image); |
| 251 | /* NOTREACHED */ |
| 252 | } |
| 253 | crash_ipi_callback(regs); |
| 254 | } |
| 255 | |
| 256 | #else |
| 257 | static void crash_kexec_prepare_cpus(int cpu) |
| 258 | { |
| 259 | /* |
| 260 | * move the secondarys to us so that we can copy |
| 261 | * the new kernel 0-0x100 safely |
| 262 | * |
| 263 | * do this if kexec in setup.c ? |
| 264 | */ |
| 265 | #ifdef CONFIG_PPC64 |
| 266 | smp_release_cpus(); |
| 267 | #else |
| 268 | /* FIXME */ |
| 269 | #endif |
| 270 | } |
| 271 | |
| 272 | void crash_kexec_secondary(struct pt_regs *regs) |
| 273 | { |
| 274 | cpus_in_sr = CPU_MASK_NONE; |
| 275 | } |
| 276 | #endif |
| 277 | |
| 278 | void default_machine_crash_shutdown(struct pt_regs *regs) |
| 279 | { |
| 280 | unsigned int irq; |
| 281 | |
| 282 | /* |
| 283 | * This function is only called after the system |
| 284 | * has panicked or is otherwise in a critical state. |
| 285 | * The minimum amount of code to allow a kexec'd kernel |
| 286 | * to run successfully needs to happen here. |
| 287 | * |
| 288 | * In practice this means stopping other cpus in |
| 289 | * an SMP system. |
| 290 | * The kernel is broken so disable interrupts. |
| 291 | */ |
| 292 | hard_irq_disable(); |
| 293 | |
| 294 | for_each_irq(irq) { |
| 295 | struct irq_desc *desc = irq_desc + irq; |
| 296 | |
| 297 | if (desc->status & IRQ_INPROGRESS) |
| 298 | desc->chip->eoi(irq); |
| 299 | |
| 300 | if (!(desc->status & IRQ_DISABLED)) |
| 301 | desc->chip->disable(irq); |
| 302 | } |
| 303 | |
| 304 | /* |
| 305 | * Make a note of crashing cpu. Will be used in machine_kexec |
| 306 | * such that another IPI will not be sent. |
| 307 | */ |
| 308 | crashing_cpu = smp_processor_id(); |
| 309 | crash_save_this_cpu(regs, crashing_cpu); |
| 310 | crash_kexec_prepare_cpus(crashing_cpu); |
| 311 | cpu_set(crashing_cpu, cpus_in_crash); |
| 312 | if (ppc_md.kexec_cpu_down) |
| 313 | ppc_md.kexec_cpu_down(1, 0); |
| 314 | } |