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
  46         for (i = 0; i < kimage->nr_segments; i++) {
  47                 pr_debug("      segment[%lu]: %016lx - %016lx, 0x%lx bytes, %lu pages\n",
  48                         i,
  49                         kimage->segment[i].mem,
  50                         kimage->segment[i].mem + kimage->segment[i].memsz,
  51                         kimage->segment[i].memsz,
  52                         kimage->segment[i].memsz /  PAGE_SIZE);
  53         }
  54 }
  55
  56 void machine_kexec_cleanup(struct kimage *kimage)
  57 {
  58         /* Empty routine needed to avoid build errors. */
  59 }
  60
  61 /**
  62  * machine_kexec_prepare - Prepare for a kexec reboot.
  63  *
  64  * Called from the core kexec code when a kernel image is loaded.
  65  * Forbid loading a kexec kernel if we have no way of hotplugging cpus or cpus
  66  * are stuck in the kernel. This avoids a panic once we hit machine_kexec().
  67  */
  68 int machine_kexec_prepare(struct kimage *kimage)
  69 {
  70         kexec_image_info(kimage);
  71
  72         if (kimage->type != KEXEC_TYPE_CRASH && cpus_are_stuck_in_kernel()) {
  73                 pr_err("Can't kexec: CPUs are stuck in the kernel.\n");
  74                 return -EBUSY;
  75         }
  76
  77         return 0;
  78 }
  79
  80 /**
  81  * kexec_list_flush - Helper to flush the kimage list and source pages to PoC.
  82  */
  83 static void kexec_list_flush(struct kimage *kimage)
  84 {
  85         kimage_entry_t *entry;
  86
  87         for (entry = &kimage->head; ; entry++) {
  88                 unsigned int flag;
  89                 void *addr;
  90
  91                 /* flush the list entries. */
  92                 __flush_dcache_area(entry, sizeof(kimage_entry_t));
  93
  94                 flag = *entry & IND_FLAGS;
  95                 if (flag == IND_DONE)
  96                         break;
  97
  98                 addr = phys_to_virt(*entry & PAGE_MASK);
  99
 100                 switch (flag) {
 101                 case IND_INDIRECTION:
 102                         /* Set entry point just before the new list page. */
 103                         entry = (kimage_entry_t *)addr - 1;
 104                         break;
 105                 case IND_SOURCE:
 106                         /* flush the source pages. */
 107                         __flush_dcache_area(addr, PAGE_SIZE);
 108                         break;
 109                 case IND_DESTINATION:
 110                         break;
 111                 default:
 112                         BUG();
 113                 }
 114         }
 115 }
 116
 117 /**
 118  * kexec_segment_flush - Helper to flush the kimage segments to PoC.
 119  */
 120 static void kexec_segment_flush(const struct kimage *kimage)
 121 {
 122         unsigned long i;
 123
 124         pr_debug("%s:\n", __func__);
 125
 126         for (i = 0; i < kimage->nr_segments; i++) {
 127                 pr_debug("  segment[%lu]: %016lx - %016lx, 0x%lx bytes, %lu pages\n",
 128                         i,
 129                         kimage->segment[i].mem,
 130                         kimage->segment[i].mem + kimage->segment[i].memsz,
 131                         kimage->segment[i].memsz,
 132                         kimage->segment[i].memsz /  PAGE_SIZE);
 133
 134                 __flush_dcache_area(phys_to_virt(kimage->segment[i].mem),
 135                         kimage->segment[i].memsz);
 136         }
 137 }
 138
 139 /**
 140  * machine_kexec - Do the kexec reboot.
 141  *
 142  * Called from the core kexec code for a sys_reboot with LINUX_REBOOT_CMD_KEXEC.
 143  */
 144 void machine_kexec(struct kimage *kimage)
 145 {
 146         phys_addr_t reboot_code_buffer_phys;
 147         void *reboot_code_buffer;
 148         bool in_kexec_crash = (kimage == kexec_crash_image);
 149         bool stuck_cpus = cpus_are_stuck_in_kernel();
 150
 151         /*
 152          * New cpus may have become stuck_in_kernel after we loaded the image.
 153          */
 154         BUG_ON(!in_kexec_crash && (stuck_cpus || (num_online_cpus() > 1)));
 155         WARN(in_kexec_crash && (stuck_cpus || smp_crash_stop_failed()),
 156                 "Some CPUs may be stale, kdump will be unreliable.\n");
 157
 158         reboot_code_buffer_phys = page_to_phys(kimage->control_code_page);
 159         reboot_code_buffer = phys_to_virt(reboot_code_buffer_phys);
 160
 161         kexec_image_info(kimage);
 162
 163         /*
 164          * Copy arm64_relocate_new_kernel to the reboot_code_buffer for use
 165          * after the kernel is shut down.
 166          */
 167         memcpy(reboot_code_buffer, arm64_relocate_new_kernel,
 168                 arm64_relocate_new_kernel_size);
 169
 170         /* Flush the reboot_code_buffer in preparation for its execution. */
 171         __flush_dcache_area(reboot_code_buffer, arm64_relocate_new_kernel_size);
 172
 173         /*
 174          * Although we've killed off the secondary CPUs, we don't update
 175          * the online mask if we're handling a crash kernel and consequently
 176          * need to avoid flush_icache_range(), which will attempt to IPI
 177          * the offline CPUs. Therefore, we must use the __* variant here.
 178          */
 179         __flush_icache_range((uintptr_t)reboot_code_buffer,
 180                              (uintptr_t)reboot_code_buffer +
 181                              arm64_relocate_new_kernel_size);
 182
 183         /* Flush the kimage list and its buffers. */
 184         kexec_list_flush(kimage);
 185
 186         /* Flush the new image if already in place. */
 187         if ((kimage != kexec_crash_image) && (kimage->head & IND_DONE))
 188                 kexec_segment_flush(kimage);
 189
 190         pr_info("Bye!\n");
 191
 192         local_daif_mask();
 193
 194         /*
 195          * cpu_soft_restart will shutdown the MMU, disable data caches, then
 196          * transfer control to the reboot_code_buffer which contains a copy of
 197          * the arm64_relocate_new_kernel routine.  arm64_relocate_new_kernel
 198          * uses physical addressing to relocate the new image to its final
 199          * position and transfers control to the image entry point when the
 200          * relocation is complete.
 201          * In kexec case, kimage->start points to purgatory assuming that
 202          * kernel entry and dtb address are embedded in purgatory by
 203          * userspace (kexec-tools).
 204          * In kexec_file case, the kernel starts directly without purgatory.
 205          */
 206         cpu_soft_restart(reboot_code_buffer_phys, kimage->head, kimage->start,
 207 #ifdef CONFIG_KEXEC_FILE
 208                                                 kimage->arch.dtb_mem);
 209 #else
 210                                                 0);
 211 #endif
 212
 213         BUG(); /* Should never get here. */
 214 }
 215
 216 static void machine_kexec_mask_interrupts(void)
 217 {
 218         unsigned int i;
 219         struct irq_desc *desc;
 220
 221         for_each_irq_desc(i, desc) {
 222                 struct irq_chip *chip;
 223                 int ret;
 224
 225                 chip = irq_desc_get_chip(desc);
 226                 if (!chip)
 227                         continue;
 228
 229                 /*
 230                  * First try to remove the active state. If this
 231                  * fails, try to EOI the interrupt.
 232                  */
 233                 ret = irq_set_irqchip_state(i, IRQCHIP_STATE_ACTIVE, false);
 234
 235                 if (ret && irqd_irq_inprogress(&desc->irq_data) &&
 236                     chip->irq_eoi)
 237                         chip->irq_eoi(&desc->irq_data);
 238
 239                 if (chip->irq_mask)
 240                         chip->irq_mask(&desc->irq_data);
 241
 242                 if (chip->irq_disable && !irqd_irq_disabled(&desc->irq_data))
 243                         chip->irq_disable(&desc->irq_data);
 244         }
 245 }
 246
 247 /**
 248  * machine_crash_shutdown - shutdown non-crashing cpus and save registers
 249  */
 250 void machine_crash_shutdown(struct pt_regs *regs)
 251 {
 252         local_irq_disable();
 253
 254         /* shutdown non-crashing cpus */
 255         crash_smp_send_stop();
 256
 257         /* for crashing cpu */
 258         crash_save_cpu(regs, smp_processor_id());
 259         machine_kexec_mask_interrupts();
 260
 261         pr_info("Starting crashdump kernel...\n");
 262 }
 263
 264 void arch_kexec_protect_crashkres(void)
 265 {
 266         int i;
 267
 268         kexec_segment_flush(kexec_crash_image);
 269
 270         for (i = 0; i < kexec_crash_image->nr_segments; i++)
 271                 set_memory_valid(
 272                         __phys_to_virt(kexec_crash_image->segment[i].mem),
 273                         kexec_crash_image->segment[i].memsz >> PAGE_SHIFT, 0);
 274 }
 275
 276 void arch_kexec_unprotect_crashkres(void)
 277 {
 278         int i;
 279
 280         for (i = 0; i < kexec_crash_image->nr_segments; i++)
 281                 set_memory_valid(
 282                         __phys_to_virt(kexec_crash_image->segment[i].mem),
 283                         kexec_crash_image->segment[i].memsz >> PAGE_SHIFT, 1);
 284 }
 285
 286 #ifdef CONFIG_HIBERNATION
 287 /*
 288  * To preserve the crash dump kernel image, the relevant memory segments
 289  * should be mapped again around the hibernation.
 290  */
 291 void crash_prepare_suspend(void)
 292 {
 293         if (kexec_crash_image)
 294                 arch_kexec_unprotect_crashkres();
 295 }
 296
 297 void crash_post_resume(void)
 298 {
 299         if (kexec_crash_image)
 300                 arch_kexec_protect_crashkres();
 301 }
 302
 303 /*
 304  * crash_is_nosave
 305  *
 306  * Return true only if a page is part of reserved memory for crash dump kernel,
 307  * but does not hold any data of loaded kernel image.
 308  *
 309  * Note that all the pages in crash dump kernel memory have been initially
 310  * marked as Reserved as memory was allocated via memblock_reserve().
 311  *
 312  * In hibernation, the pages which are Reserved and yet "nosave" are excluded
 313  * from the hibernation iamge. crash_is_nosave() does thich check for crash
 314  * dump kernel and will reduce the total size of hibernation image.
 315  */
 316
 317 bool crash_is_nosave(unsigned long pfn)
 318 {
 319         int i;
 320         phys_addr_t addr;
 321
 322         if (!crashk_res.end)
 323                 return false;
 324
 325         /* in reserved memory? */
 326         addr = __pfn_to_phys(pfn);
 327         if ((addr < crashk_res.start) || (crashk_res.end < addr))
 328                 return false;
 329
 330         if (!kexec_crash_image)
 331                 return true;
 332
 333         /* not part of loaded kernel image? */
 334         for (i = 0; i < kexec_crash_image->nr_segments; i++)
 335                 if (addr >= kexec_crash_image->segment[i].mem &&
 336                                 addr < (kexec_crash_image->segment[i].mem +
 337                                         kexec_crash_image->segment[i].memsz))
 338                         return false;
 339
 340         return true;
 341 }
 342
 343 void crash_free_reserved_phys_range(unsigned long begin, unsigned long end)
 344 {
 345         unsigned long addr;
 346         struct page *page;
 347
 348         for (addr = begin; addr < end; addr += PAGE_SIZE) {
 349                 page = phys_to_page(addr);
 350                 free_reserved_page(page);
 351         }
 352 }
 353 #endif /* CONFIG_HIBERNATION */