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