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