~ljy/RK3588_XEN.git

2024-11-01 2f529f9b558ca1c1bd74be7437a84e4711743404

commit \| author \| age
a07526	1	/*
H	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>
2f529f	10	#include <linux/irq_pipeline.h>
a07526	11	#include <linux/kdebug.h>
H	12	#include <linux/module.h>
	13	#include <linux/ptrace.h>
	14	#include <linux/sched/debug.h>
	15	#include <linux/sched/task_stack.h>
	16	#include <linux/ftrace.h>
	17	#include <linux/kexec.h>
	18	#include <linux/bug.h>
	19	#include <linux/nmi.h>
	20	#include <linux/sysfs.h>
	21	#include <linux/kasan.h>
	22
	23	#include <asm/cpu_entry_area.h>
	24	#include <asm/stacktrace.h>
	25	#include <asm/unwind.h>
	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 noinstr 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	/* Called from get_stack_info_noinstr - so must be noinstr too */
	51	bool noinstr in_entry_stack(unsigned long stack, struct stack_info info)
	52	{
	53	struct entry_stack *ss = cpu_entry_stack(smp_processor_id());
	54
	55	void *begin = ss;
	56	void *end = ss + 1;
	57
	58	if ((void )stack < begin \|\| (void )stack >= end)
	59	return false;
	60
	61	info->type = STACK_TYPE_ENTRY;
	62	info->begin = begin;
	63	info->end = end;
	64	info->next_sp = NULL;
	65
	66	return true;
	67	}
	68
	69	static void printk_stack_address(unsigned long address, int reliable,
	70	const char *log_lvl)
	71	{
	72	touch_nmi_watchdog();
	73	printk("%s %s%pB\n", log_lvl, reliable ? "" : "? ", (void *)address);
	74	}
	75
	76	static int copy_code(struct pt_regs regs, u8 buf, unsigned long src,
	77	unsigned int nbytes)
	78	{
	79	if (!user_mode(regs))
	80	return copy_from_kernel_nofault(buf, (u8 *)src, nbytes);
	81
	82	/* The user space code from other tasks cannot be accessed. */
	83	if (regs != task_pt_regs(current))
	84	return -EPERM;
	85	/*
	86	* Make sure userspace isn't trying to trick us into dumping kernel
	87	* memory by pointing the userspace instruction pointer at it.
	88	*/
	89	if (__chk_range_not_ok(src, nbytes, TASK_SIZE_MAX))
	90	return -EINVAL;
	91
	92	/*
	93	* Even if named copy_from_user_nmi() this can be invoked from
	94	* other contexts and will not try to resolve a pagefault, which is
	95	* the correct thing to do here as this code can be called from any
	96	* context.
	97	*/
	98	return copy_from_user_nmi(buf, (void __user *)src, nbytes);
	99	}
	100
	101	/*
	102	* There are a couple of reasons for the 2/3rd prologue, courtesy of Linus:
	103	*
	104	* In case where we don't have the exact kernel image (which, if we did, we can
	105	* simply disassemble and navigate to the RIP), the purpose of the bigger
	106	* prologue is to have more context and to be able to correlate the code from
	107	* the different toolchains better.
	108	*
	109	* In addition, it helps in recreating the register allocation of the failing
	110	* kernel and thus make sense of the register dump.
	111	*
	112	* What is more, the additional complication of a variable length insn arch like
	113	* x86 warrants having longer byte sequence before rIP so that the disassembler
	114	* can "sync" up properly and find instruction boundaries when decoding the
	115	* opcode bytes.
	116	*
	117	* Thus, the 2/3rds prologue and 64 byte OPCODE_BUFSIZE is just a random
	118	* guesstimate in attempt to achieve all of the above.
	119	*/
	120	void show_opcodes(struct pt_regs regs, const char loglvl)
	121	{
	122	#define PROLOGUE_SIZE 42
	123	#define EPILOGUE_SIZE 21
	124	#define OPCODE_BUFSIZE (PROLOGUE_SIZE + 1 + EPILOGUE_SIZE)
	125	u8 opcodes[OPCODE_BUFSIZE];
	126	unsigned long prologue = regs->ip - PROLOGUE_SIZE;
	127
	128	switch (copy_code(regs, opcodes, prologue, sizeof(opcodes))) {
	129	case 0:
	130	printk("%sCode: %" __stringify(PROLOGUE_SIZE) "ph <%02x> %"
	131	__stringify(EPILOGUE_SIZE) "ph\n", loglvl, opcodes,
	132	opcodes[PROLOGUE_SIZE], opcodes + PROLOGUE_SIZE + 1);
	133	break;
	134	case -EPERM:
	135	/* No access to the user space stack of other tasks. Ignore. */
	136	break;
	137	default:
	138	printk("%sCode: Unable to access opcode bytes at RIP 0x%lx.\n",
	139	loglvl, prologue);
	140	break;
	141	}
	142	}
	143
	144	void show_ip(struct pt_regs regs, const char loglvl)
	145	{
	146	#ifdef CONFIG_X86_32
	147	printk("%sEIP: %pS\n", loglvl, (void *)regs->ip);
	148	#else
	149	printk("%sRIP: %04x:%pS\n", loglvl, (int)regs->cs, (void *)regs->ip);
	150	#endif
	151	show_opcodes(regs, loglvl);
	152	}
	153
	154	void show_iret_regs(struct pt_regs regs, const char log_lvl)
	155	{
	156	show_ip(regs, log_lvl);
	157	printk("%sRSP: %04x:%016lx EFLAGS: %08lx", log_lvl, (int)regs->ss,
	158	regs->sp, regs->flags);
	159	}
	160
	161	static void show_regs_if_on_stack(struct stack_info info, struct pt_regs regs,
	162	bool partial, const char *log_lvl)
	163	{
	164	/*
	165	* These on_stack() checks aren't strictly necessary: the unwind code
	166	* has already validated the 'regs' pointer. The checks are done for
	167	* ordering reasons: if the registers are on the next stack, we don't
	168	* want to print them out yet. Otherwise they'll be shown as part of
	169	* the wrong stack. Later, when show_trace_log_lvl() switches to the
	170	* next stack, this function will be called again with the same regs so
	171	* they can be printed in the right context.
	172	*/
	173	if (!partial && on_stack(info, regs, sizeof(*regs))) {
	174	__show_regs(regs, SHOW_REGS_SHORT, log_lvl);
	175
	176	} else if (partial && on_stack(info, (void *)regs + IRET_FRAME_OFFSET,
	177	IRET_FRAME_SIZE)) {
	178	/*
	179	* When an interrupt or exception occurs in entry code, the
	180	* full pt_regs might not have been saved yet. In that case
	181	* just print the iret frame.
	182	*/
	183	show_iret_regs(regs, log_lvl);
	184	}
	185	}
	186
	187	void show_trace_log_lvl(struct task_struct task, struct pt_regs regs,
	188	unsigned long stack, const char log_lvl)
	189	{
	190	struct unwind_state state;
	191	struct stack_info stack_info = {0};
	192	unsigned long visit_mask = 0;
	193	int graph_idx = 0;
	194	bool partial = false;
	195
	196	printk("%sCall Trace:\n", log_lvl);
	197
	198	unwind_start(&state, task, regs, stack);
	199	stack = stack ? : get_stack_pointer(task, regs);
	200	regs = unwind_get_entry_regs(&state, &partial);
	201
	202	/*
	203	* Iterate through the stacks, starting with the current stack pointer.
	204	* Each stack has a pointer to the next one.
	205	*
	206	* x86-64 can have several stacks:
	207	* - task stack
	208	* - interrupt stack
	209	* - HW exception stacks (double fault, nmi, debug, mce)
	210	* - entry stack
	211	*
	212	* x86-32 can have up to four stacks:
	213	* - task stack
	214	* - softirq stack
	215	* - hardirq stack
	216	* - entry stack
	217	*/
	218	for ( ; stack; stack = PTR_ALIGN(stack_info.next_sp, sizeof(long))) {
	219	const char *stack_name;
	220
	221	if (get_stack_info(stack, task, &stack_info, &visit_mask)) {
	222	/*
	223	* We weren't on a valid stack. It's possible that
	224	* we overflowed a valid stack into a guard page.
	225	* See if the next page up is valid so that we can
	226	* generate some kind of backtrace if this happens.
	227	*/
	228	stack = (unsigned long *)PAGE_ALIGN((unsigned long)stack);
	229	if (get_stack_info(stack, task, &stack_info, &visit_mask))
	230	break;
	231	}
	232
	233	stack_name = stack_type_name(stack_info.type);
	234	if (stack_name)
	235	printk("%s <%s>\n", log_lvl, stack_name);
	236
	237	if (regs)
	238	show_regs_if_on_stack(&stack_info, regs, partial, log_lvl);
	239
	240	/*
	241	* Scan the stack, printing any text addresses we find. At the
	242	* same time, follow proper stack frames with the unwinder.
	243	*
	244	* Addresses found during the scan which are not reported by
	245	* the unwinder are considered to be additional clues which are
	246	* sometimes useful for debugging and are prefixed with '?'.
	247	* This also serves as a failsafe option in case the unwinder
	248	* goes off in the weeds.
	249	*/
	250	for (; stack < stack_info.end; stack++) {
	251	unsigned long real_addr;
	252	int reliable = 0;
	253	unsigned long addr = READ_ONCE_NOCHECK(*stack);
	254	unsigned long *ret_addr_p =
	255	unwind_get_return_address_ptr(&state);
	256
	257	if (!__kernel_text_address(addr))
	258	continue;
	259
	260	/*
	261	* Don't print regs->ip again if it was already printed
	262	* by show_regs_if_on_stack().
	263	*/
	264	if (regs && stack == &regs->ip)
	265	goto next;
	266
	267	if (stack == ret_addr_p)
	268	reliable = 1;
	269
	270	/*
	271	* When function graph tracing is enabled for a
	272	* function, its return address on the stack is
	273	* replaced with the address of an ftrace handler
	274	* (return_to_handler). In that case, before printing
	275	* the "real" address, we want to print the handler
	276	* address as an "unreliable" hint that function graph
	277	* tracing was involved.
	278	*/
	279	real_addr = ftrace_graph_ret_addr(task, &graph_idx,
	280	addr, stack);
	281	if (real_addr != addr)
	282	printk_stack_address(addr, 0, log_lvl);
	283	printk_stack_address(real_addr, reliable, log_lvl);
	284
	285	if (!reliable)
	286	continue;
	287
	288	next:
	289	/*
	290	* Get the next frame from the unwinder. No need to
	291	* check for an error: if anything goes wrong, the rest
	292	* of the addresses will just be printed as unreliable.
	293	*/
	294	unwind_next_frame(&state);
	295
	296	/* if the frame has entry regs, print them */
	297	regs = unwind_get_entry_regs(&state, &partial);
	298	if (regs)
	299	show_regs_if_on_stack(&stack_info, regs, partial, log_lvl);
	300	}
	301
	302	if (stack_name)
	303	printk("%s </%s>\n", log_lvl, stack_name);
	304	}
	305	}
	306
	307	void show_stack(struct task_struct task, unsigned long sp,
	308	const char *loglvl)
	309	{
	310	task = task ? : current;
	311
	312	/*
	313	* Stack frames below this one aren't interesting. Don't show them
	314	* if we're printing for %current.
	315	*/
	316	if (!sp && task == current)
	317	sp = get_stack_pointer(current, NULL);
	318
	319	show_trace_log_lvl(task, NULL, sp, loglvl);
	320	}
	321
	322	void show_stack_regs(struct pt_regs *regs)
	323	{
	324	show_trace_log_lvl(current, regs, NULL, KERN_DEFAULT);
	325	}
	326
	327	static arch_spinlock_t die_lock = __ARCH_SPIN_LOCK_UNLOCKED;
	328	static int die_owner = -1;
	329	static unsigned int die_nest_count;
	330
	331	unsigned long oops_begin(void)
	332	{
	333	int cpu;
	334	unsigned long flags;
	335
	336	oops_enter();
	337
	338	/* racy, but better than risking deadlock. */
2f529f	339	flags = hard_local_irq_save();
a07526	340	cpu = smp_processor_id();
H	341	if (!arch_spin_trylock(&die_lock)) {
	342	if (cpu == die_owner)
	343	/* nested oops. should stop eventually */;
	344	else
	345	arch_spin_lock(&die_lock);
	346	}
	347	die_nest_count++;
	348	die_owner = cpu;
	349	console_verbose();
	350	bust_spinlocks(1);
	351	return flags;
	352	}
	353	NOKPROBE_SYMBOL(oops_begin);
	354
	355	void __noreturn rewind_stack_do_exit(int signr);
	356
	357	void oops_end(unsigned long flags, struct pt_regs *regs, int signr)
	358	{
	359	if (regs && kexec_should_crash(current))
	360	crash_kexec(regs);
	361
	362	bust_spinlocks(0);
	363	die_owner = -1;
	364	add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
	365	die_nest_count--;
	366	if (!die_nest_count)
	367	/* Nest count reaches zero, release the lock. */
	368	arch_spin_unlock(&die_lock);
2f529f	369	hard_local_irq_restore(flags);
a07526	370	oops_exit();
H	371
	372	/* Executive summary in case the oops scrolled away */
	373	__show_regs(&exec_summary_regs, SHOW_REGS_ALL, KERN_DEFAULT);
	374
	375	if (!signr)
	376	return;
	377	if (in_interrupt())
	378	panic("Fatal exception in interrupt");
	379	if (panic_on_oops)
	380	panic("Fatal exception");
	381
	382	/*
	383	* We're not going to return, but we might be on an IST stack or
	384	* have very little stack space left. Rewind the stack and kill
	385	* the task.
	386	* Before we rewind the stack, we have to tell KASAN that we're going to
	387	* reuse the task stack and that existing poisons are invalid.
	388	*/
	389	kasan_unpoison_task_stack(current);
	390	rewind_stack_do_exit(signr);
	391	}
	392	NOKPROBE_SYMBOL(oops_end);
	393
	394	static void __die_header(const char str, struct pt_regs regs, long err)
	395	{
	396	const char *pr = "";
	397
2f529f	398	irq_pipeline_oops();
H	399
a07526	400	/* Save the regs of the first oops for the executive summary later. */
H	401	if (!die_counter)
	402	exec_summary_regs = *regs;
	403
	404	if (IS_ENABLED(CONFIG_PREEMPTION))
	405	pr = IS_ENABLED(CONFIG_PREEMPT_RT) ? " PREEMPT_RT" : " PREEMPT";
	406
	407	printk(KERN_DEFAULT
2f529f	408	"%s: %04lx [#%d]%s%s%s%s%s%s\n", str, err & 0xffff, ++die_counter,
a07526	409	pr,
H	410	IS_ENABLED(CONFIG_SMP) ? " SMP" : "",
	411	debug_pagealloc_enabled() ? " DEBUG_PAGEALLOC" : "",
	412	IS_ENABLED(CONFIG_KASAN) ? " KASAN" : "",
	413	IS_ENABLED(CONFIG_PAGE_TABLE_ISOLATION) ?
2f529f	414	(boot_cpu_has(X86_FEATURE_PTI) ? " PTI" : " NOPTI") : "",
H	415	irqs_pipelined() ? " IRQ_PIPELINE" : "");
a07526	416	}
H	417	NOKPROBE_SYMBOL(__die_header);
	418
	419	static int __die_body(const char str, struct pt_regs regs, long err)
	420	{
	421	show_regs(regs);
	422	print_modules();
	423
	424	if (notify_die(DIE_OOPS, str, regs, err,
	425	current->thread.trap_nr, SIGSEGV) == NOTIFY_STOP)
	426	return 1;
	427
	428	return 0;
	429	}
	430	NOKPROBE_SYMBOL(__die_body);
	431
	432	int __die(const char str, struct pt_regs regs, long err)
	433	{
	434	__die_header(str, regs, err);
	435	return __die_body(str, regs, err);
	436	}
	437	NOKPROBE_SYMBOL(__die);
	438
	439	/*
	440	* This is gone through when something in the kernel has done something bad
	441	* and is about to be terminated:
	442	*/
	443	void die(const char str, struct pt_regs regs, long err)
	444	{
	445	unsigned long flags = oops_begin();
	446	int sig = SIGSEGV;
	447
	448	if (__die(str, regs, err))
	449	sig = 0;
	450	oops_end(flags, regs, sig);
	451	}
	452
	453	void die_addr(const char str, struct pt_regs regs, long err, long gp_addr)
	454	{
	455	unsigned long flags = oops_begin();
	456	int sig = SIGSEGV;
	457
	458	__die_header(str, regs, err);
	459	if (gp_addr)
	460	kasan_non_canonical_hook(gp_addr);
	461	if (__die_body(str, regs, err))
	462	sig = 0;
	463	oops_end(flags, regs, sig);
	464	}
	465
	466	void show_regs(struct pt_regs *regs)
	467	{
	468	enum show_regs_mode print_kernel_regs;
	469
	470	show_regs_print_info(KERN_DEFAULT);
	471
	472	print_kernel_regs = user_mode(regs) ? SHOW_REGS_USER : SHOW_REGS_ALL;
	473	__show_regs(regs, print_kernel_regs, KERN_DEFAULT);
	474
	475	/*
	476	* When in-kernel, we also print out the stack at the time of the fault..
	477	*/
	478	if (!user_mode(regs))
	479	show_trace_log_lvl(current, regs, NULL, KERN_DEFAULT);
	480	}