linux/arch/s390/kernel/machine_kexec.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright IBM Corp. 2005, 2011
   4 *
   5 * Author(s): Rolf Adelsberger,
   6 *            Heiko Carstens <heiko.carstens@de.ibm.com>
   7 *            Michael Holzheu <holzheu@linux.vnet.ibm.com>
   8 */
   9
  10#include <linux/device.h>
  11#include <linux/mm.h>
  12#include <linux/kexec.h>
  13#include <linux/delay.h>
  14#include <linux/reboot.h>
  15#include <linux/ftrace.h>
  16#include <linux/debug_locks.h>
  17#include <linux/suspend.h>
  18#include <asm/cio.h>
  19#include <asm/setup.h>
  20#include <asm/pgtable.h>
  21#include <asm/pgalloc.h>
  22#include <asm/smp.h>
  23#include <asm/ipl.h>
  24#include <asm/diag.h>
  25#include <asm/elf.h>
  26#include <asm/asm-offsets.h>
  27#include <asm/cacheflush.h>
  28#include <asm/os_info.h>
  29#include <asm/set_memory.h>
  30#include <asm/switch_to.h>
  31#include <asm/nmi.h>
  32
  33typedef void (*relocate_kernel_t)(kimage_entry_t *, unsigned long);
  34
  35extern const unsigned char relocate_kernel[];
  36extern const unsigned long long relocate_kernel_len;
  37
  38#ifdef CONFIG_CRASH_DUMP
  39
  40/*
  41 * PM notifier callback for kdump
  42 */
  43static int machine_kdump_pm_cb(struct notifier_block *nb, unsigned long action,
  44                               void *ptr)
  45{
  46        switch (action) {
  47        case PM_SUSPEND_PREPARE:
  48        case PM_HIBERNATION_PREPARE:
  49                if (kexec_crash_image)
  50                        arch_kexec_unprotect_crashkres();
  51                break;
  52        case PM_POST_SUSPEND:
  53        case PM_POST_HIBERNATION:
  54                if (kexec_crash_image)
  55                        arch_kexec_protect_crashkres();
  56                break;
  57        default:
  58                return NOTIFY_DONE;
  59        }
  60        return NOTIFY_OK;
  61}
  62
  63static int __init machine_kdump_pm_init(void)
  64{
  65        pm_notifier(machine_kdump_pm_cb, 0);
  66        return 0;
  67}
  68arch_initcall(machine_kdump_pm_init);
  69
  70/*
  71 * Reset the system, copy boot CPU registers to absolute zero,
  72 * and jump to the kdump image
  73 */
  74static void __do_machine_kdump(void *image)
  75{
  76        int (*start_kdump)(int);
  77        unsigned long prefix;
  78
  79        /* store_status() saved the prefix register to lowcore */
  80        prefix = (unsigned long) S390_lowcore.prefixreg_save_area;
  81
  82        /* Now do the reset  */
  83        s390_reset_system();
  84
  85        /*
  86         * Copy dump CPU store status info to absolute zero.
  87         * This need to be done *after* s390_reset_system set the
  88         * prefix register of this CPU to zero
  89         */
  90        memcpy((void *) __LC_FPREGS_SAVE_AREA,
  91               (void *)(prefix + __LC_FPREGS_SAVE_AREA), 512);
  92
  93        __load_psw_mask(PSW_MASK_BASE | PSW_DEFAULT_KEY | PSW_MASK_EA | PSW_MASK_BA);
  94        start_kdump = (void *)((struct kimage *) image)->start;
  95        start_kdump(1);
  96
  97        /* Die if start_kdump returns */
  98        disabled_wait((unsigned long) __builtin_return_address(0));
  99}
 100
 101/*
 102 * Start kdump: create a LGR log entry, store status of all CPUs and
 103 * branch to __do_machine_kdump.
 104 */
 105static noinline void __machine_kdump(void *image)
 106{
 107        struct mcesa *mcesa;
 108        union ctlreg2 cr2_old, cr2_new;
 109        int this_cpu, cpu;
 110
 111        lgr_info_log();
 112        /* Get status of the other CPUs */
 113        this_cpu = smp_find_processor_id(stap());
 114        for_each_online_cpu(cpu) {
 115                if (cpu == this_cpu)
 116                        continue;
 117                if (smp_store_status(cpu))
 118                        continue;
 119        }
 120        /* Store status of the boot CPU */
 121        mcesa = (struct mcesa *)(S390_lowcore.mcesad & MCESA_ORIGIN_MASK);
 122        if (MACHINE_HAS_VX)
 123                save_vx_regs((__vector128 *) mcesa->vector_save_area);
 124        if (MACHINE_HAS_GS) {
 125                __ctl_store(cr2_old.val, 2, 2);
 126                cr2_new = cr2_old;
 127                cr2_new.gse = 1;
 128                __ctl_load(cr2_new.val, 2, 2);
 129                save_gs_cb((struct gs_cb *) mcesa->guarded_storage_save_area);
 130                __ctl_load(cr2_old.val, 2, 2);
 131        }
 132        /*
 133         * To create a good backchain for this CPU in the dump store_status
 134         * is passed the address of a function. The address is saved into
 135         * the PSW save area of the boot CPU and the function is invoked as
 136         * a tail call of store_status. The backchain in the dump will look
 137         * like this:
 138         *   restart_int_handler ->  __machine_kexec -> __do_machine_kdump
 139         * The call to store_status() will not return.
 140         */
 141        store_status(__do_machine_kdump, image);
 142}
 143#endif
 144
 145/*
 146 * Check if kdump checksums are valid: We call purgatory with parameter "0"
 147 */
 148static bool kdump_csum_valid(struct kimage *image)
 149{
 150#ifdef CONFIG_CRASH_DUMP
 151        int (*start_kdump)(int) = (void *)image->start;
 152        int rc;
 153
 154        __arch_local_irq_stnsm(0xfb); /* disable DAT */
 155        rc = start_kdump(0);
 156        __arch_local_irq_stosm(0x04); /* enable DAT */
 157        return rc == 0;
 158#else
 159        return false;
 160#endif
 161}
 162
 163#ifdef CONFIG_CRASH_DUMP
 164
 165void crash_free_reserved_phys_range(unsigned long begin, unsigned long end)
 166{
 167        unsigned long addr, size;
 168
 169        for (addr = begin; addr < end; addr += PAGE_SIZE)
 170                free_reserved_page(pfn_to_page(addr >> PAGE_SHIFT));
 171        size = begin - crashk_res.start;
 172        if (size)
 173                os_info_crashkernel_add(crashk_res.start, size);
 174        else
 175                os_info_crashkernel_add(0, 0);
 176}
 177
 178static void crash_protect_pages(int protect)
 179{
 180        unsigned long size;
 181
 182        if (!crashk_res.end)
 183                return;
 184        size = resource_size(&crashk_res);
 185        if (protect)
 186                set_memory_ro(crashk_res.start, size >> PAGE_SHIFT);
 187        else
 188                set_memory_rw(crashk_res.start, size >> PAGE_SHIFT);
 189}
 190
 191void arch_kexec_protect_crashkres(void)
 192{
 193        crash_protect_pages(1);
 194}
 195
 196void arch_kexec_unprotect_crashkres(void)
 197{
 198        crash_protect_pages(0);
 199}
 200
 201#endif
 202
 203/*
 204 * Give back memory to hypervisor before new kdump is loaded
 205 */
 206static int machine_kexec_prepare_kdump(void)
 207{
 208#ifdef CONFIG_CRASH_DUMP
 209        if (MACHINE_IS_VM)
 210                diag10_range(PFN_DOWN(crashk_res.start),
 211                             PFN_DOWN(crashk_res.end - crashk_res.start + 1));
 212        return 0;
 213#else
 214        return -EINVAL;
 215#endif
 216}
 217
 218int machine_kexec_prepare(struct kimage *image)
 219{
 220        void *reboot_code_buffer;
 221
 222        if (image->type == KEXEC_TYPE_CRASH)
 223                return machine_kexec_prepare_kdump();
 224
 225        /* We don't support anything but the default image type for now. */
 226        if (image->type != KEXEC_TYPE_DEFAULT)
 227                return -EINVAL;
 228
 229        /* Get the destination where the assembler code should be copied to.*/
 230        reboot_code_buffer = (void *) page_to_phys(image->control_code_page);
 231
 232        /* Then copy it */
 233        memcpy(reboot_code_buffer, relocate_kernel, relocate_kernel_len);
 234        return 0;
 235}
 236
 237void machine_kexec_cleanup(struct kimage *image)
 238{
 239}
 240
 241void arch_crash_save_vmcoreinfo(void)
 242{
 243        VMCOREINFO_SYMBOL(lowcore_ptr);
 244        VMCOREINFO_SYMBOL(high_memory);
 245        VMCOREINFO_LENGTH(lowcore_ptr, NR_CPUS);
 246        vmcoreinfo_append_str("SDMA=%lx\n", __sdma);
 247        vmcoreinfo_append_str("EDMA=%lx\n", __edma);
 248        vmcoreinfo_append_str("KERNELOFFSET=%lx\n", kaslr_offset());
 249        mem_assign_absolute(S390_lowcore.vmcore_info, paddr_vmcoreinfo_note());
 250}
 251
 252void machine_shutdown(void)
 253{
 254}
 255
 256void machine_crash_shutdown(struct pt_regs *regs)
 257{
 258        set_os_info_reipl_block();
 259}
 260
 261/*
 262 * Do normal kexec
 263 */
 264static void __do_machine_kexec(void *data)
 265{
 266        relocate_kernel_t data_mover;
 267        struct kimage *image = data;
 268
 269        s390_reset_system();
 270        data_mover = (relocate_kernel_t) page_to_phys(image->control_code_page);
 271
 272        __arch_local_irq_stnsm(0xfb); /* disable DAT - avoid no-execute */
 273        /* Call the moving routine */
 274        (*data_mover)(&image->head, image->start);
 275
 276        /* Die if kexec returns */
 277        disabled_wait((unsigned long) __builtin_return_address(0));
 278}
 279
 280/*
 281 * Reset system and call either kdump or normal kexec
 282 */
 283static void __machine_kexec(void *data)
 284{
 285        __arch_local_irq_stosm(0x04); /* enable DAT */
 286        pfault_fini();
 287        tracing_off();
 288        debug_locks_off();
 289#ifdef CONFIG_CRASH_DUMP
 290        if (((struct kimage *) data)->type == KEXEC_TYPE_CRASH)
 291                __machine_kdump(data);
 292#endif
 293        __do_machine_kexec(data);
 294}
 295
 296/*
 297 * Do either kdump or normal kexec. In case of kdump we first ask
 298 * purgatory, if kdump checksums are valid.
 299 */
 300void machine_kexec(struct kimage *image)
 301{
 302        if (image->type == KEXEC_TYPE_CRASH && !kdump_csum_valid(image))
 303                return;
 304        tracer_disable();
 305        smp_send_stop();
 306        smp_call_ipl_cpu(__machine_kexec, image);
 307}
 308