arch/arm64/kernel/machine_kexec.c

   1 /*
   2  * kexec for arm64
   3  *
   4  * Copyright (C) Linaro.
   5  * Copyright (C) Huawei Futurewei Technologies.
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  */
  11
  12 #include <linux/interrupt.h>
  13 #include <linux/irq.h>
  14 #include <linux/kernel.h>
  15 #include <linux/kexec.h>
  16 #include <linux/page-flags.h>
  17 #include <linux/smp.h>
  18
  19 #include <asm/cacheflush.h>
  20 #include <asm/cpu_ops.h>
  21 #include <asm/daifflags.h>
  22 #include <asm/memory.h>
  23 #include <asm/mmu.h>
  24 #include <asm/mmu_context.h>
  25 #include <asm/page.h>
  26
  27 #include "cpu-reset.h"
  28
  29 /* Global variables for the arm64_relocate_new_kernel routine. */
  30 extern const unsigned char arm64_relocate_new_kernel[];
  31 extern const unsigned long arm64_relocate_new_kernel_size;
  32
  33 /**
  34  * kexec_image_info - For debugging output.
  35  */
  36 #define kexec_image_info(_i) _kexec_image_info(__func__, __LINE__, _i)
  37 static void _kexec_image_info(const char *func, int line,
  38         const struct kimage *kimage)
  39 {
  40         unsigned long i;
  41
  42         pr_debug("%s:%d:\n", func, line);
  43         pr_debug("  kexec kimage info:\n");
  44         pr_debug("    type:        %d\n", kimage->type);
  45         pr_debug("    start:       %lx\n", kimage->start);
  46         pr_debug("    head:        %lx\n", kimage->head);
  47         pr_debug("    nr_segments: %lu\n", kimage->nr_segments);
  48
  49         for (i = 0; i < kimage->nr_segments; i++) {
  50                 pr_debug("      segment[%lu]: %016lx - %016lx, 0x%lx bytes, %lu pages\n",
  51                         i,
  52                         kimage->segment[i].mem,
  53                         kimage->segment[i].mem + kimage->segment[i].memsz,
  54                         kimage->segment[i].memsz,
  55                         kimage->segment[i].memsz /  PAGE_SIZE);
  56         }
  57 }
  58
  59 void machine_kexec_cleanup(struct kimage *kimage)
  60 {
  61         /* Empty routine needed to avoid build errors. */
  62 }
  63
  64 /**
  65  * machine_kexec_prepare - Prepare for a kexec reboot.
  66  *
  67  * Called from the core kexec code when a kernel image is loaded.
  68  * Forbid loading a kexec kernel if we have no way of hotplugging cpus or cpus
  69  * are stuck in the kernel. This avoids a panic once we hit machine_kexec().
  70  */
  71 int machine_kexec_prepare(struct kimage *kimage)
  72 {
  73         kexec_image_info(kimage);
  74
  75         if (kimage->type != KEXEC_TYPE_CRASH && cpus_are_stuck_in_kernel()) {
  76                 pr_err("Can't kexec: CPUs are stuck in the kernel.\n");
  77                 return -EBUSY;
  78         }
  79
  80         return 0;
  81 }
  82
  83 /**
  84  * kexec_list_flush - Helper to flush the kimage list and source pages to PoC.
  85  */
  86 static void kexec_list_flush(struct kimage *kimage)
  87 {
  88         kimage_entry_t *entry;
  89
  90         for (entry = &kimage->head; ; entry++) {
  91                 unsigned int flag;
  92                 void *addr;
  93
  94                 /* flush the list entries. */
  95                 __flush_dcache_area(entry, sizeof(kimage_entry_t));
  96
  97                 flag = *entry & IND_FLAGS;
  98                 if (flag == IND_DONE)
  99                         break;
 100
 101                 addr = phys_to_virt(*entry & PAGE_MASK);
 102
 103                 switch (flag) {
 104                 case IND_INDIRECTION:
 105                         /* Set entry point just before the new list page. */
 106                         entry = (kimage_entry_t *)addr - 1;
 107                         break;
 108                 case IND_SOURCE:
 109                         /* flush the source pages. */
 110                         __flush_dcache_area(addr, PAGE_SIZE);
 111                         break;
 112                 case IND_DESTINATION:
 113                         break;
 114                 default:
 115                         BUG();
 116                 }
 117         }
 118 }
 119
 120 /**
 121  * kexec_segment_flush - Helper to flush the kimage segments to PoC.
 122  */
 123 static void kexec_segment_flush(const struct kimage *kimage)
 124 {
 125         unsigned long i;
 126
 127         pr_debug("%s:\n", __func__);
 128
 129         for (i = 0; i < kimage->nr_segments; i++) {
 130                 pr_debug("  segment[%lu]: %016lx - %016lx, 0x%lx bytes, %lu pages\n",
 131                         i,
 132                         kimage->segment[i].mem,
 133                         kimage->segment[i].mem + kimage->segment[i].memsz,
 134                         kimage->segment[i].memsz,
 135                         kimage->segment[i].memsz /  PAGE_SIZE);
 136
 137                 __flush_dcache_area(phys_to_virt(kimage->segment[i].mem),
 138                         kimage->segment[i].memsz);
 139         }
 140 }
 141
 142 /**
 143  * machine_kexec - Do the kexec reboot.
 144  *
 145  * Called from the core kexec code for a sys_reboot with LINUX_REBOOT_CMD_KEXEC.
 146  */
 147 void machine_kexec(struct kimage *kimage)
 148 {
 149         phys_addr_t reboot_code_buffer_phys;
 150         void *reboot_code_buffer;
 151         bool in_kexec_crash = (kimage == kexec_crash_image);
 152         bool stuck_cpus = cpus_are_stuck_in_kernel();
 153
 154         /*
 155          * New cpus may have become stuck_in_kernel after we loaded the image.
 156          */
 157         BUG_ON(!in_kexec_crash && (stuck_cpus || (num_online_cpus() > 1)));
 158         WARN(in_kexec_crash && (stuck_cpus || smp_crash_stop_failed()),
 159                 "Some CPUs may be stale, kdump will be unreliable.\n");
 160
 161         reboot_code_buffer_phys = page_to_phys(kimage->control_code_page);
 162         reboot_code_buffer = phys_to_virt(reboot_code_buffer_phys);
 163
 164         kexec_image_info(kimage);
 165
 166         pr_debug("%s:%d: control_code_page:        %p\n", __func__, __LINE__,
 167                 kimage->control_code_page);
 168         pr_debug("%s:%d: reboot_code_buffer_phys:  %pa\n", __func__, __LINE__,
 169                 &reboot_code_buffer_phys);
 170         pr_debug("%s:%d: reboot_code_buffer:       %p\n", __func__, __LINE__,
 171                 reboot_code_buffer);
 172         pr_debug("%s:%d: relocate_new_kernel:      %p\n", __func__, __LINE__,
 173                 arm64_relocate_new_kernel);
 174         pr_debug("%s:%d: relocate_new_kernel_size: 0x%lx(%lu) bytes\n",
 175                 __func__, __LINE__, arm64_relocate_new_kernel_size,
 176                 arm64_relocate_new_kernel_size);
 177
 178         /*
 179          * Copy arm64_relocate_new_kernel to the reboot_code_buffer for use
 180          * after the kernel is shut down.
 181          */
 182         memcpy(reboot_code_buffer, arm64_relocate_new_kernel,
 183                 arm64_relocate_new_kernel_size);
 184
 185         /* Flush the reboot_code_buffer in preparation for its execution. */
 186         __flush_dcache_area(reboot_code_buffer, arm64_relocate_new_kernel_size);
 187
 188         /*
 189          * Although we've killed off the secondary CPUs, we don't update
 190          * the online mask if we're handling a crash kernel and consequently
 191          * need to avoid flush_icache_range(), which will attempt to IPI
 192          * the offline CPUs. Therefore, we must use the __* variant here.
 193          */
 194         __flush_icache_range((uintptr_t)reboot_code_buffer,
 195                              arm64_relocate_new_kernel_size);
 196
 197         /* Flush the kimage list and its buffers. */
 198         kexec_list_flush(kimage);
 199
 200         /* Flush the new image if already in place. */
 201         if ((kimage != kexec_crash_image) && (kimage->head & IND_DONE))
 202                 kexec_segment_flush(kimage);
 203
 204         pr_info("Bye!\n");
 205
 206         local_daif_mask();
 207
 208         /*
 209          * cpu_soft_restart will shutdown the MMU, disable data caches, then
 210          * transfer control to the reboot_code_buffer which contains a copy of
 211          * the arm64_relocate_new_kernel routine.  arm64_relocate_new_kernel
 212          * uses physical addressing to relocate the new image to its final
 213          * position and transfers control to the image entry point when the
 214          * relocation is complete.
 215          */
 216
 217         cpu_soft_restart(reboot_code_buffer_phys, kimage->head, kimage->start, 0);
 218
 219         BUG(); /* Should never get here. */
 220 }
 221
 222 static void machine_kexec_mask_interrupts(void)
 223 {
 224         unsigned int i;
 225         struct irq_desc *desc;
 226
 227         for_each_irq_desc(i, desc) {
 228                 struct irq_chip *chip;
 229                 int ret;
 230
 231                 chip = irq_desc_get_chip(desc);
 232                 if (!chip)
 233                         continue;
 234
 235                 /*
 236                  * First try to remove the active state. If this
 237                  * fails, try to EOI the interrupt.
 238                  */
 239                 ret = irq_set_irqchip_state(i, IRQCHIP_STATE_ACTIVE, false);
 240
 241                 if (ret && irqd_irq_inprogress(&desc->irq_data) &&
 242                     chip->irq_eoi)
 243                         chip->irq_eoi(&desc->irq_data);
 244
 245                 if (chip->irq_mask)
 246                         chip->irq_mask(&desc->irq_data);
 247
 248                 if (chip->irq_disable && !irqd_irq_disabled(&desc->irq_data))
 249                         chip->irq_disable(&desc->irq_data);
 250         }
 251 }
 252
 253 /**
 254  * machine_crash_shutdown - shutdown non-crashing cpus and save registers
 255  */
 256 void machine_crash_shutdown(struct pt_regs *regs)
 257 {
 258         local_irq_disable();
 259
 260         /* shutdown non-crashing cpus */
 261         crash_smp_send_stop();
 262
 263         /* for crashing cpu */
 264         crash_save_cpu(regs, smp_processor_id());
 265         machine_kexec_mask_interrupts();
 266
 267         pr_info("Starting crashdump kernel...\n");
 268 }
 269
 270 void arch_kexec_protect_crashkres(void)
 271 {
 272         int i;
 273
 274         kexec_segment_flush(kexec_crash_image);
 275
 276         for (i = 0; i < kexec_crash_image->nr_segments; i++)
 277                 set_memory_valid(
 278                         __phys_to_virt(kexec_crash_image->segment[i].mem),
 279                         kexec_crash_image->segment[i].memsz >> PAGE_SHIFT, 0);
 280 }
 281
 282 void arch_kexec_unprotect_crashkres(void)
 283 {
 284         int i;
 285
 286         for (i = 0; i < kexec_crash_image->nr_segments; i++)
 287                 set_memory_valid(
 288                         __phys_to_virt(kexec_crash_image->segment[i].mem),
 289                         kexec_crash_image->segment[i].memsz >> PAGE_SHIFT, 1);
 290 }
 291
 292 #ifdef CONFIG_HIBERNATION
 293 /*
 294  * To preserve the crash dump kernel image, the relevant memory segments
 295  * should be mapped again around the hibernation.
 296  */
 297 void crash_prepare_suspend(void)
 298 {
 299         if (kexec_crash_image)
 300                 arch_kexec_unprotect_crashkres();
 301 }
 302
 303 void crash_post_resume(void)
 304 {
 305         if (kexec_crash_image)
 306                 arch_kexec_protect_crashkres();
 307 }
 308
 309 /*
 310  * crash_is_nosave
 311  *
 312  * Return true only if a page is part of reserved memory for crash dump kernel,
 313  * but does not hold any data of loaded kernel image.
 314  *
 315  * Note that all the pages in crash dump kernel memory have been initially
 316  * marked as Reserved in kexec_reserve_crashkres_pages().
 317  *
 318  * In hibernation, the pages which are Reserved and yet "nosave" are excluded
 319  * from the hibernation iamge. crash_is_nosave() does thich check for crash
 320  * dump kernel and will reduce the total size of hibernation image.
 321  */
 322
 323 bool crash_is_nosave(unsigned long pfn)
 324 {
 325         int i;
 326         phys_addr_t addr;
 327
 328         if (!crashk_res.end)
 329                 return false;
 330
 331         /* in reserved memory? */
 332         addr = __pfn_to_phys(pfn);
 333         if ((addr < crashk_res.start) || (crashk_res.end < addr))
 334                 return false;
 335
 336         if (!kexec_crash_image)
 337                 return true;
 338
 339         /* not part of loaded kernel image? */
 340         for (i = 0; i < kexec_crash_image->nr_segments; i++)
 341                 if (addr >= kexec_crash_image->segment[i].mem &&
 342                                 addr < (kexec_crash_image->segment[i].mem +
 343                                         kexec_crash_image->segment[i].memsz))
 344                         return false;
 345
 346         return true;
 347 }
 348
 349 void crash_free_reserved_phys_range(unsigned long begin, unsigned long end)
 350 {
 351         unsigned long addr;
 352         struct page *page;
 353
 354         for (addr = begin; addr < end; addr += PAGE_SIZE) {
 355                 page = phys_to_page(addr);
 356                 ClearPageReserved(page);
 357                 free_reserved_page(page);
 358         }
 359 }
 360 #endif /* CONFIG_HIBERNATION */
 361
 362 void arch_crash_save_vmcoreinfo(void)
 363 {
 364         VMCOREINFO_NUMBER(VA_BITS);
 365         /* Please note VMCOREINFO_NUMBER() uses "%d", not "%x" */
 366         vmcoreinfo_append_str("NUMBER(kimage_voffset)=0x%llx\n",
 367                                                 kimage_voffset);
 368         vmcoreinfo_append_str("NUMBER(PHYS_OFFSET)=0x%llx\n",
 369                                                 PHYS_OFFSET);
 370         vmcoreinfo_append_str("KERNELOFFSET=%lx\n", kaslr_offset());
 371 }