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