arch/arm64/kernel/machine_kexec.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * kexec for arm64
   4  *
   5  * Copyright (C) Linaro.
   6  * Copyright (C) Huawei Futurewei Technologies.
   7  */
   8
   9 #include <linux/interrupt.h>
  10 #include <linux/irq.h>
  11 #include <linux/kernel.h>
  12 #include <linux/kexec.h>
  13 #include <linux/page-flags.h>
  14 #include <linux/smp.h>
  15
  16 #include <asm/cacheflush.h>
  17 #include <asm/cpu_ops.h>
  18 #include <asm/daifflags.h>
  19 #include <asm/memory.h>
  20 #include <asm/mmu.h>
  21 #include <asm/mmu_context.h>
  22 #include <asm/page.h>
  23
  24 #include "cpu-reset.h"
  25
  26 /* Global variables for the arm64_relocate_new_kernel routine. */
  27 extern const unsigned char arm64_relocate_new_kernel[];
  28 extern const unsigned long arm64_relocate_new_kernel_size;
  29
  30 /**
  31  * kexec_image_info - For debugging output.
  32  */
  33 #define kexec_image_info(_i) _kexec_image_info(__func__, __LINE__, _i)
  34 static void _kexec_image_info(const char *func, int line,
  35         const struct kimage *kimage)
  36 {
  37         unsigned long i;
  38
  39         pr_debug("%s:%d:\n", func, line);
  40         pr_debug("  kexec kimage info:\n");
  41         pr_debug("    type:        %d\n", kimage->type);
  42         pr_debug("    start:       %lx\n", kimage->start);
  43         pr_debug("    head:        %lx\n", kimage->head);
  44         pr_debug("    nr_segments: %lu\n", kimage->nr_segments);
  45         pr_debug("    kern_reloc: %pa\n", &kimage->arch.kern_reloc);
  46
  47         for (i = 0; i < kimage->nr_segments; i++) {
  48                 pr_debug("      segment[%lu]: %016lx - %016lx, 0x%lx bytes, %lu pages\n",
  49                         i,
  50                         kimage->segment[i].mem,
  51                         kimage->segment[i].mem + kimage->segment[i].memsz,
  52                         kimage->segment[i].memsz,
  53                         kimage->segment[i].memsz /  PAGE_SIZE);
  54         }
  55 }
  56
  57 void machine_kexec_cleanup(struct kimage *kimage)
  58 {
  59         /* Empty routine needed to avoid build errors. */
  60 }
  61
  62 int machine_kexec_post_load(struct kimage *kimage)
  63 {
  64         void *reloc_code = page_to_virt(kimage->control_code_page);
  65
  66         memcpy(reloc_code, arm64_relocate_new_kernel,
  67                arm64_relocate_new_kernel_size);
  68         kimage->arch.kern_reloc = __pa(reloc_code);
  69         kexec_image_info(kimage);
  70
  71         /*
  72          * For execution with the MMU off, reloc_code needs to be cleaned to the
  73          * PoC and invalidated from the I-cache.
  74          */
  75         dcache_clean_inval_poc((unsigned long)reloc_code,
  76                             (unsigned long)reloc_code +
  77                                     arm64_relocate_new_kernel_size);
  78         icache_inval_pou((uintptr_t)reloc_code,
  79                                 (uintptr_t)reloc_code +
  80                                         arm64_relocate_new_kernel_size);
  81
  82         return 0;
  83 }
  84
  85 /**
  86  * machine_kexec_prepare - Prepare for a kexec reboot.
  87  *
  88  * Called from the core kexec code when a kernel image is loaded.
  89  * Forbid loading a kexec kernel if we have no way of hotplugging cpus or cpus
  90  * are stuck in the kernel. This avoids a panic once we hit machine_kexec().
  91  */
  92 int machine_kexec_prepare(struct kimage *kimage)
  93 {
  94         if (kimage->type != KEXEC_TYPE_CRASH && cpus_are_stuck_in_kernel()) {
  95                 pr_err("Can't kexec: CPUs are stuck in the kernel.\n");
  96                 return -EBUSY;
  97         }
  98
  99         return 0;
 100 }
 101
 102 /**
 103  * kexec_list_flush - Helper to flush the kimage list and source pages to PoC.
 104  */
 105 static void kexec_list_flush(struct kimage *kimage)
 106 {
 107         kimage_entry_t *entry;
 108
 109         for (entry = &kimage->head; ; entry++) {
 110                 unsigned int flag;
 111                 unsigned long addr;
 112
 113                 /* flush the list entries. */
 114                 dcache_clean_inval_poc((unsigned long)entry,
 115                                     (unsigned long)entry +
 116                                             sizeof(kimage_entry_t));
 117
 118                 flag = *entry & IND_FLAGS;
 119                 if (flag == IND_DONE)
 120                         break;
 121
 122                 addr = (unsigned long)phys_to_virt(*entry & PAGE_MASK);
 123
 124                 switch (flag) {
 125                 case IND_INDIRECTION:
 126                         /* Set entry point just before the new list page. */
 127                         entry = (kimage_entry_t *)addr - 1;
 128                         break;
 129                 case IND_SOURCE:
 130                         /* flush the source pages. */
 131                         dcache_clean_inval_poc(addr, addr + PAGE_SIZE);
 132                         break;
 133                 case IND_DESTINATION:
 134                         break;
 135                 default:
 136                         BUG();
 137                 }
 138         }
 139 }
 140
 141 /**
 142  * kexec_segment_flush - Helper to flush the kimage segments to PoC.
 143  */
 144 static void kexec_segment_flush(const struct kimage *kimage)
 145 {
 146         unsigned long i;
 147
 148         pr_debug("%s:\n", __func__);
 149
 150         for (i = 0; i < kimage->nr_segments; i++) {
 151                 pr_debug("  segment[%lu]: %016lx - %016lx, 0x%lx bytes, %lu pages\n",
 152                         i,
 153                         kimage->segment[i].mem,
 154                         kimage->segment[i].mem + kimage->segment[i].memsz,
 155                         kimage->segment[i].memsz,
 156                         kimage->segment[i].memsz /  PAGE_SIZE);
 157
 158                 dcache_clean_inval_poc(
 159                         (unsigned long)phys_to_virt(kimage->segment[i].mem),
 160                         (unsigned long)phys_to_virt(kimage->segment[i].mem) +
 161                                 kimage->segment[i].memsz);
 162         }
 163 }
 164
 165 /**
 166  * machine_kexec - Do the kexec reboot.
 167  *
 168  * Called from the core kexec code for a sys_reboot with LINUX_REBOOT_CMD_KEXEC.
 169  */
 170 void machine_kexec(struct kimage *kimage)
 171 {
 172         bool in_kexec_crash = (kimage == kexec_crash_image);
 173         bool stuck_cpus = cpus_are_stuck_in_kernel();
 174
 175         /*
 176          * New cpus may have become stuck_in_kernel after we loaded the image.
 177          */
 178         BUG_ON(!in_kexec_crash && (stuck_cpus || (num_online_cpus() > 1)));
 179         WARN(in_kexec_crash && (stuck_cpus || smp_crash_stop_failed()),
 180                 "Some CPUs may be stale, kdump will be unreliable.\n");
 181
 182         /* Flush the kimage list and its buffers. */
 183         kexec_list_flush(kimage);
 184
 185         /* Flush the new image if already in place. */
 186         if ((kimage != kexec_crash_image) && (kimage->head & IND_DONE))
 187                 kexec_segment_flush(kimage);
 188
 189         pr_info("Bye!\n");
 190
 191         local_daif_mask();
 192
 193         /*
 194          * cpu_soft_restart will shutdown the MMU, disable data caches, then
 195          * transfer control to the kern_reloc which contains a copy of
 196          * the arm64_relocate_new_kernel routine.  arm64_relocate_new_kernel
 197          * uses physical addressing to relocate the new image to its final
 198          * position and transfers control to the image entry point when the
 199          * relocation is complete.
 200          * In kexec case, kimage->start points to purgatory assuming that
 201          * kernel entry and dtb address are embedded in purgatory by
 202          * userspace (kexec-tools).
 203          * In kexec_file case, the kernel starts directly without purgatory.
 204          */
 205         cpu_soft_restart(kimage->arch.kern_reloc, kimage->head, kimage->start,
 206                          kimage->arch.dtb_mem);
 207
 208         BUG(); /* Should never get here. */
 209 }
 210
 211 static void machine_kexec_mask_interrupts(void)
 212 {
 213         unsigned int i;
 214         struct irq_desc *desc;
 215
 216         for_each_irq_desc(i, desc) {
 217                 struct irq_chip *chip;
 218                 int ret;
 219
 220                 chip = irq_desc_get_chip(desc);
 221                 if (!chip)
 222                         continue;
 223
 224                 /*
 225                  * First try to remove the active state. If this
 226                  * fails, try to EOI the interrupt.
 227                  */
 228                 ret = irq_set_irqchip_state(i, IRQCHIP_STATE_ACTIVE, false);
 229
 230                 if (ret && irqd_irq_inprogress(&desc->irq_data) &&
 231                     chip->irq_eoi)
 232                         chip->irq_eoi(&desc->irq_data);
 233
 234                 if (chip->irq_mask)
 235                         chip->irq_mask(&desc->irq_data);
 236
 237                 if (chip->irq_disable && !irqd_irq_disabled(&desc->irq_data))
 238                         chip->irq_disable(&desc->irq_data);
 239         }
 240 }
 241
 242 /**
 243  * machine_crash_shutdown - shutdown non-crashing cpus and save registers
 244  */
 245 void machine_crash_shutdown(struct pt_regs *regs)
 246 {
 247         local_irq_disable();
 248
 249         /* shutdown non-crashing cpus */
 250         crash_smp_send_stop();
 251
 252         /* for crashing cpu */
 253         crash_save_cpu(regs, smp_processor_id());
 254         machine_kexec_mask_interrupts();
 255
 256         pr_info("Starting crashdump kernel...\n");
 257 }
 258
 259 void arch_kexec_protect_crashkres(void)
 260 {
 261         int i;
 262
 263         kexec_segment_flush(kexec_crash_image);
 264
 265         for (i = 0; i < kexec_crash_image->nr_segments; i++)
 266                 set_memory_valid(
 267                         __phys_to_virt(kexec_crash_image->segment[i].mem),
 268                         kexec_crash_image->segment[i].memsz >> PAGE_SHIFT, 0);
 269 }
 270
 271 void arch_kexec_unprotect_crashkres(void)
 272 {
 273         int i;
 274
 275         for (i = 0; i < kexec_crash_image->nr_segments; i++)
 276                 set_memory_valid(
 277                         __phys_to_virt(kexec_crash_image->segment[i].mem),
 278                         kexec_crash_image->segment[i].memsz >> PAGE_SHIFT, 1);
 279 }
 280
 281 #ifdef CONFIG_HIBERNATION
 282 /*
 283  * To preserve the crash dump kernel image, the relevant memory segments
 284  * should be mapped again around the hibernation.
 285  */
 286 void crash_prepare_suspend(void)
 287 {
 288         if (kexec_crash_image)
 289                 arch_kexec_unprotect_crashkres();
 290 }
 291
 292 void crash_post_resume(void)
 293 {
 294         if (kexec_crash_image)
 295                 arch_kexec_protect_crashkres();
 296 }
 297
 298 /*
 299  * crash_is_nosave
 300  *
 301  * Return true only if a page is part of reserved memory for crash dump kernel,
 302  * but does not hold any data of loaded kernel image.
 303  *
 304  * Note that all the pages in crash dump kernel memory have been initially
 305  * marked as Reserved as memory was allocated via memblock_reserve().
 306  *
 307  * In hibernation, the pages which are Reserved and yet "nosave" are excluded
 308  * from the hibernation iamge. crash_is_nosave() does thich check for crash
 309  * dump kernel and will reduce the total size of hibernation image.
 310  */
 311
 312 bool crash_is_nosave(unsigned long pfn)
 313 {
 314         int i;
 315         phys_addr_t addr;
 316
 317         if (!crashk_res.end)
 318                 return false;
 319
 320         /* in reserved memory? */
 321         addr = __pfn_to_phys(pfn);
 322         if ((addr < crashk_res.start) || (crashk_res.end < addr))
 323                 return false;
 324
 325         if (!kexec_crash_image)
 326                 return true;
 327
 328         /* not part of loaded kernel image? */
 329         for (i = 0; i < kexec_crash_image->nr_segments; i++)
 330                 if (addr >= kexec_crash_image->segment[i].mem &&
 331                                 addr < (kexec_crash_image->segment[i].mem +
 332                                         kexec_crash_image->segment[i].memsz))
 333                         return false;
 334
 335         return true;
 336 }
 337
 338 void crash_free_reserved_phys_range(unsigned long begin, unsigned long end)
 339 {
 340         unsigned long addr;
 341         struct page *page;
 342
 343         for (addr = begin; addr < end; addr += PAGE_SIZE) {
 344                 page = phys_to_page(addr);
 345                 free_reserved_page(page);
 346         }
 347 }
 348 #endif /* CONFIG_HIBERNATION */