1 /* 2 * Copyright (C) 1991, 1992 Linus Torvalds 3 * Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs 4 */ 5 #include <linux/kallsyms.h> 6 #include <linux/kprobes.h> 7 #include <linux/uaccess.h> 8 #include <linux/utsname.h> 9 #include <linux/hardirq.h> 10 #include <linux/kdebug.h> 11 #include <linux/module.h> 12 #include <linux/ptrace.h> 13 #include <linux/sched/debug.h> 14 #include <linux/sched/task_stack.h> 15 #include <linux/ftrace.h> 16 #include <linux/kexec.h> 17 #include <linux/bug.h> 18 #include <linux/nmi.h> 19 #include <linux/sysfs.h> 20 21 #include <asm/cpu_entry_area.h> 22 #include <asm/stacktrace.h> 23 #include <asm/unwind.h> 24 25 #define OPCODE_BUFSIZE 64 26 27 int panic_on_unrecovered_nmi; 28 int panic_on_io_nmi; 29 static int die_counter; 30 31 static struct pt_regs exec_summary_regs; 32 33 bool in_task_stack(unsigned long *stack, struct task_struct *task, 34 struct stack_info *info) 35 { 36 unsigned long *begin = task_stack_page(task); 37 unsigned long *end = task_stack_page(task) + THREAD_SIZE; 38 39 if (stack < begin || stack >= end) 40 return false; 41 42 info->type = STACK_TYPE_TASK; 43 info->begin = begin; 44 info->end = end; 45 info->next_sp = NULL; 46 47 return true; 48 } 49 50 bool in_entry_stack(unsigned long *stack, struct stack_info *info) 51 { 52 struct entry_stack *ss = cpu_entry_stack(smp_processor_id()); 53 54 void *begin = ss; 55 void *end = ss + 1; 56 57 if ((void *)stack < begin || (void *)stack >= end) 58 return false; 59 60 info->type = STACK_TYPE_ENTRY; 61 info->begin = begin; 62 info->end = end; 63 info->next_sp = NULL; 64 65 return true; 66 } 67 68 static void printk_stack_address(unsigned long address, int reliable, 69 char *log_lvl) 70 { 71 touch_nmi_watchdog(); 72 printk("%s %s%pB\n", log_lvl, reliable ? "" : "? ", (void *)address); 73 } 74 75 /* 76 * There are a couple of reasons for the 2/3rd prologue, courtesy of Linus: 77 * 78 * In case where we don't have the exact kernel image (which, if we did, we can 79 * simply disassemble and navigate to the RIP), the purpose of the bigger 80 * prologue is to have more context and to be able to correlate the code from 81 * the different toolchains better. 82 * 83 * In addition, it helps in recreating the register allocation of the failing 84 * kernel and thus make sense of the register dump. 85 * 86 * What is more, the additional complication of a variable length insn arch like 87 * x86 warrants having longer byte sequence before rIP so that the disassembler 88 * can "sync" up properly and find instruction boundaries when decoding the 89 * opcode bytes. 90 * 91 * Thus, the 2/3rds prologue and 64 byte OPCODE_BUFSIZE is just a random 92 * guesstimate in attempt to achieve all of the above. 93 */ 94 void show_opcodes(u8 *rip, const char *loglvl) 95 { 96 unsigned int code_prologue = OPCODE_BUFSIZE * 2 / 3; 97 u8 opcodes[OPCODE_BUFSIZE]; 98 u8 *ip; 99 int i; 100 101 printk("%sCode: ", loglvl); 102 103 ip = (u8 *)rip - code_prologue; 104 if (probe_kernel_read(opcodes, ip, OPCODE_BUFSIZE)) { 105 pr_cont("Bad RIP value.\n"); 106 return; 107 } 108 109 for (i = 0; i < OPCODE_BUFSIZE; i++, ip++) { 110 if (ip == rip) 111 pr_cont("<%02x> ", opcodes[i]); 112 else 113 pr_cont("%02x ", opcodes[i]); 114 } 115 pr_cont("\n"); 116 } 117 118 void show_ip(struct pt_regs *regs, const char *loglvl) 119 { 120 #ifdef CONFIG_X86_32 121 printk("%sEIP: %pS\n", loglvl, (void *)regs->ip); 122 #else 123 printk("%sRIP: %04x:%pS\n", loglvl, (int)regs->cs, (void *)regs->ip); 124 #endif 125 show_opcodes((u8 *)regs->ip, loglvl); 126 } 127 128 void show_iret_regs(struct pt_regs *regs) 129 { 130 show_ip(regs, KERN_DEFAULT); 131 printk(KERN_DEFAULT "RSP: %04x:%016lx EFLAGS: %08lx", (int)regs->ss, 132 regs->sp, regs->flags); 133 } 134 135 static void show_regs_if_on_stack(struct stack_info *info, struct pt_regs *regs, 136 bool partial) 137 { 138 /* 139 * These on_stack() checks aren't strictly necessary: the unwind code 140 * has already validated the 'regs' pointer. The checks are done for 141 * ordering reasons: if the registers are on the next stack, we don't 142 * want to print them out yet. Otherwise they'll be shown as part of 143 * the wrong stack. Later, when show_trace_log_lvl() switches to the 144 * next stack, this function will be called again with the same regs so 145 * they can be printed in the right context. 146 */ 147 if (!partial && on_stack(info, regs, sizeof(*regs))) { 148 __show_regs(regs, 0); 149 150 } else if (partial && on_stack(info, (void *)regs + IRET_FRAME_OFFSET, 151 IRET_FRAME_SIZE)) { 152 /* 153 * When an interrupt or exception occurs in entry code, the 154 * full pt_regs might not have been saved yet. In that case 155 * just print the iret frame. 156 */ 157 show_iret_regs(regs); 158 } 159 } 160 161 void show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs, 162 unsigned long *stack, char *log_lvl) 163 { 164 struct unwind_state state; 165 struct stack_info stack_info = {0}; 166 unsigned long visit_mask = 0; 167 int graph_idx = 0; 168 bool partial = false; 169 170 printk("%sCall Trace:\n", log_lvl); 171 172 unwind_start(&state, task, regs, stack); 173 stack = stack ? : get_stack_pointer(task, regs); 174 regs = unwind_get_entry_regs(&state, &partial); 175 176 /* 177 * Iterate through the stacks, starting with the current stack pointer. 178 * Each stack has a pointer to the next one. 179 * 180 * x86-64 can have several stacks: 181 * - task stack 182 * - interrupt stack 183 * - HW exception stacks (double fault, nmi, debug, mce) 184 * - entry stack 185 * 186 * x86-32 can have up to four stacks: 187 * - task stack 188 * - softirq stack 189 * - hardirq stack 190 * - entry stack 191 */ 192 for ( ; stack; stack = PTR_ALIGN(stack_info.next_sp, sizeof(long))) { 193 const char *stack_name; 194 195 if (get_stack_info(stack, task, &stack_info, &visit_mask)) { 196 /* 197 * We weren't on a valid stack. It's possible that 198 * we overflowed a valid stack into a guard page. 199 * See if the next page up is valid so that we can 200 * generate some kind of backtrace if this happens. 201 */ 202 stack = (unsigned long *)PAGE_ALIGN((unsigned long)stack); 203 if (get_stack_info(stack, task, &stack_info, &visit_mask)) 204 break; 205 } 206 207 stack_name = stack_type_name(stack_info.type); 208 if (stack_name) 209 printk("%s <%s>\n", log_lvl, stack_name); 210 211 if (regs) 212 show_regs_if_on_stack(&stack_info, regs, partial); 213 214 /* 215 * Scan the stack, printing any text addresses we find. At the 216 * same time, follow proper stack frames with the unwinder. 217 * 218 * Addresses found during the scan which are not reported by 219 * the unwinder are considered to be additional clues which are 220 * sometimes useful for debugging and are prefixed with '?'. 221 * This also serves as a failsafe option in case the unwinder 222 * goes off in the weeds. 223 */ 224 for (; stack < stack_info.end; stack++) { 225 unsigned long real_addr; 226 int reliable = 0; 227 unsigned long addr = READ_ONCE_NOCHECK(*stack); 228 unsigned long *ret_addr_p = 229 unwind_get_return_address_ptr(&state); 230 231 if (!__kernel_text_address(addr)) 232 continue; 233 234 /* 235 * Don't print regs->ip again if it was already printed 236 * by show_regs_if_on_stack(). 237 */ 238 if (regs && stack == ®s->ip) 239 goto next; 240 241 if (stack == ret_addr_p) 242 reliable = 1; 243 244 /* 245 * When function graph tracing is enabled for a 246 * function, its return address on the stack is 247 * replaced with the address of an ftrace handler 248 * (return_to_handler). In that case, before printing 249 * the "real" address, we want to print the handler 250 * address as an "unreliable" hint that function graph 251 * tracing was involved. 252 */ 253 real_addr = ftrace_graph_ret_addr(task, &graph_idx, 254 addr, stack); 255 if (real_addr != addr) 256 printk_stack_address(addr, 0, log_lvl); 257 printk_stack_address(real_addr, reliable, log_lvl); 258 259 if (!reliable) 260 continue; 261 262 next: 263 /* 264 * Get the next frame from the unwinder. No need to 265 * check for an error: if anything goes wrong, the rest 266 * of the addresses will just be printed as unreliable. 267 */ 268 unwind_next_frame(&state); 269 270 /* if the frame has entry regs, print them */ 271 regs = unwind_get_entry_regs(&state, &partial); 272 if (regs) 273 show_regs_if_on_stack(&stack_info, regs, partial); 274 } 275 276 if (stack_name) 277 printk("%s </%s>\n", log_lvl, stack_name); 278 } 279 } 280 281 void show_stack(struct task_struct *task, unsigned long *sp) 282 { 283 task = task ? : current; 284 285 /* 286 * Stack frames below this one aren't interesting. Don't show them 287 * if we're printing for %current. 288 */ 289 if (!sp && task == current) 290 sp = get_stack_pointer(current, NULL); 291 292 show_trace_log_lvl(task, NULL, sp, KERN_DEFAULT); 293 } 294 295 void show_stack_regs(struct pt_regs *regs) 296 { 297 show_trace_log_lvl(current, regs, NULL, KERN_DEFAULT); 298 } 299 300 static arch_spinlock_t die_lock = __ARCH_SPIN_LOCK_UNLOCKED; 301 static int die_owner = -1; 302 static unsigned int die_nest_count; 303 304 unsigned long oops_begin(void) 305 { 306 int cpu; 307 unsigned long flags; 308 309 oops_enter(); 310 311 /* racy, but better than risking deadlock. */ 312 raw_local_irq_save(flags); 313 cpu = smp_processor_id(); 314 if (!arch_spin_trylock(&die_lock)) { 315 if (cpu == die_owner) 316 /* nested oops. should stop eventually */; 317 else 318 arch_spin_lock(&die_lock); 319 } 320 die_nest_count++; 321 die_owner = cpu; 322 console_verbose(); 323 bust_spinlocks(1); 324 return flags; 325 } 326 NOKPROBE_SYMBOL(oops_begin); 327 328 void __noreturn rewind_stack_do_exit(int signr); 329 330 void oops_end(unsigned long flags, struct pt_regs *regs, int signr) 331 { 332 if (regs && kexec_should_crash(current)) 333 crash_kexec(regs); 334 335 bust_spinlocks(0); 336 die_owner = -1; 337 add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE); 338 die_nest_count--; 339 if (!die_nest_count) 340 /* Nest count reaches zero, release the lock. */ 341 arch_spin_unlock(&die_lock); 342 raw_local_irq_restore(flags); 343 oops_exit(); 344 345 /* Executive summary in case the oops scrolled away */ 346 __show_regs(&exec_summary_regs, true); 347 348 if (!signr) 349 return; 350 if (in_interrupt()) 351 panic("Fatal exception in interrupt"); 352 if (panic_on_oops) 353 panic("Fatal exception"); 354 355 /* 356 * We're not going to return, but we might be on an IST stack or 357 * have very little stack space left. Rewind the stack and kill 358 * the task. 359 */ 360 rewind_stack_do_exit(signr); 361 } 362 NOKPROBE_SYMBOL(oops_end); 363 364 int __die(const char *str, struct pt_regs *regs, long err) 365 { 366 /* Save the regs of the first oops for the executive summary later. */ 367 if (!die_counter) 368 exec_summary_regs = *regs; 369 370 printk(KERN_DEFAULT 371 "%s: %04lx [#%d]%s%s%s%s%s\n", str, err & 0xffff, ++die_counter, 372 IS_ENABLED(CONFIG_PREEMPT) ? " PREEMPT" : "", 373 IS_ENABLED(CONFIG_SMP) ? " SMP" : "", 374 debug_pagealloc_enabled() ? " DEBUG_PAGEALLOC" : "", 375 IS_ENABLED(CONFIG_KASAN) ? " KASAN" : "", 376 IS_ENABLED(CONFIG_PAGE_TABLE_ISOLATION) ? 377 (boot_cpu_has(X86_FEATURE_PTI) ? " PTI" : " NOPTI") : ""); 378 379 show_regs(regs); 380 print_modules(); 381 382 if (notify_die(DIE_OOPS, str, regs, err, 383 current->thread.trap_nr, SIGSEGV) == NOTIFY_STOP) 384 return 1; 385 386 return 0; 387 } 388 NOKPROBE_SYMBOL(__die); 389 390 /* 391 * This is gone through when something in the kernel has done something bad 392 * and is about to be terminated: 393 */ 394 void die(const char *str, struct pt_regs *regs, long err) 395 { 396 unsigned long flags = oops_begin(); 397 int sig = SIGSEGV; 398 399 if (__die(str, regs, err)) 400 sig = 0; 401 oops_end(flags, regs, sig); 402 } 403 404 void show_regs(struct pt_regs *regs) 405 { 406 bool all = true; 407 408 show_regs_print_info(KERN_DEFAULT); 409 410 if (IS_ENABLED(CONFIG_X86_32)) 411 all = !user_mode(regs); 412 413 __show_regs(regs, all); 414 415 /* 416 * When in-kernel, we also print out the stack at the time of the fault.. 417 */ 418 if (!user_mode(regs)) 419 show_trace_log_lvl(current, regs, NULL, KERN_DEFAULT); 420 } 421