arch/s390/kernel/machine_kexec.c

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