xref: /openbmc/linux/kernel/printk/printk_safe.c (revision 20e2fc42)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * printk_safe.c - Safe printk for printk-deadlock-prone contexts
4  */
5 
6 #include <linux/preempt.h>
7 #include <linux/spinlock.h>
8 #include <linux/debug_locks.h>
9 #include <linux/smp.h>
10 #include <linux/cpumask.h>
11 #include <linux/irq_work.h>
12 #include <linux/printk.h>
13 
14 #include "internal.h"
15 
16 /*
17  * printk() could not take logbuf_lock in NMI context. Instead,
18  * it uses an alternative implementation that temporary stores
19  * the strings into a per-CPU buffer. The content of the buffer
20  * is later flushed into the main ring buffer via IRQ work.
21  *
22  * The alternative implementation is chosen transparently
23  * by examinig current printk() context mask stored in @printk_context
24  * per-CPU variable.
25  *
26  * The implementation allows to flush the strings also from another CPU.
27  * There are situations when we want to make sure that all buffers
28  * were handled or when IRQs are blocked.
29  */
30 static int printk_safe_irq_ready __read_mostly;
31 
32 #define SAFE_LOG_BUF_LEN ((1 << CONFIG_PRINTK_SAFE_LOG_BUF_SHIFT) -	\
33 				sizeof(atomic_t) -			\
34 				sizeof(atomic_t) -			\
35 				sizeof(struct irq_work))
36 
37 struct printk_safe_seq_buf {
38 	atomic_t		len;	/* length of written data */
39 	atomic_t		message_lost;
40 	struct irq_work		work;	/* IRQ work that flushes the buffer */
41 	unsigned char		buffer[SAFE_LOG_BUF_LEN];
42 };
43 
44 static DEFINE_PER_CPU(struct printk_safe_seq_buf, safe_print_seq);
45 static DEFINE_PER_CPU(int, printk_context);
46 
47 #ifdef CONFIG_PRINTK_NMI
48 static DEFINE_PER_CPU(struct printk_safe_seq_buf, nmi_print_seq);
49 #endif
50 
51 /* Get flushed in a more safe context. */
52 static void queue_flush_work(struct printk_safe_seq_buf *s)
53 {
54 	if (printk_safe_irq_ready)
55 		irq_work_queue(&s->work);
56 }
57 
58 /*
59  * Add a message to per-CPU context-dependent buffer. NMI and printk-safe
60  * have dedicated buffers, because otherwise printk-safe preempted by
61  * NMI-printk would have overwritten the NMI messages.
62  *
63  * The messages are flushed from irq work (or from panic()), possibly,
64  * from other CPU, concurrently with printk_safe_log_store(). Should this
65  * happen, printk_safe_log_store() will notice the buffer->len mismatch
66  * and repeat the write.
67  */
68 static __printf(2, 0) int printk_safe_log_store(struct printk_safe_seq_buf *s,
69 						const char *fmt, va_list args)
70 {
71 	int add;
72 	size_t len;
73 	va_list ap;
74 
75 again:
76 	len = atomic_read(&s->len);
77 
78 	/* The trailing '\0' is not counted into len. */
79 	if (len >= sizeof(s->buffer) - 1) {
80 		atomic_inc(&s->message_lost);
81 		queue_flush_work(s);
82 		return 0;
83 	}
84 
85 	/*
86 	 * Make sure that all old data have been read before the buffer
87 	 * was reset. This is not needed when we just append data.
88 	 */
89 	if (!len)
90 		smp_rmb();
91 
92 	va_copy(ap, args);
93 	add = vscnprintf(s->buffer + len, sizeof(s->buffer) - len, fmt, ap);
94 	va_end(ap);
95 	if (!add)
96 		return 0;
97 
98 	/*
99 	 * Do it once again if the buffer has been flushed in the meantime.
100 	 * Note that atomic_cmpxchg() is an implicit memory barrier that
101 	 * makes sure that the data were written before updating s->len.
102 	 */
103 	if (atomic_cmpxchg(&s->len, len, len + add) != len)
104 		goto again;
105 
106 	queue_flush_work(s);
107 	return add;
108 }
109 
110 static inline void printk_safe_flush_line(const char *text, int len)
111 {
112 	/*
113 	 * Avoid any console drivers calls from here, because we may be
114 	 * in NMI or printk_safe context (when in panic). The messages
115 	 * must go only into the ring buffer at this stage.  Consoles will
116 	 * get explicitly called later when a crashdump is not generated.
117 	 */
118 	printk_deferred("%.*s", len, text);
119 }
120 
121 /* printk part of the temporary buffer line by line */
122 static int printk_safe_flush_buffer(const char *start, size_t len)
123 {
124 	const char *c, *end;
125 	bool header;
126 
127 	c = start;
128 	end = start + len;
129 	header = true;
130 
131 	/* Print line by line. */
132 	while (c < end) {
133 		if (*c == '\n') {
134 			printk_safe_flush_line(start, c - start + 1);
135 			start = ++c;
136 			header = true;
137 			continue;
138 		}
139 
140 		/* Handle continuous lines or missing new line. */
141 		if ((c + 1 < end) && printk_get_level(c)) {
142 			if (header) {
143 				c = printk_skip_level(c);
144 				continue;
145 			}
146 
147 			printk_safe_flush_line(start, c - start);
148 			start = c++;
149 			header = true;
150 			continue;
151 		}
152 
153 		header = false;
154 		c++;
155 	}
156 
157 	/* Check if there was a partial line. Ignore pure header. */
158 	if (start < end && !header) {
159 		static const char newline[] = KERN_CONT "\n";
160 
161 		printk_safe_flush_line(start, end - start);
162 		printk_safe_flush_line(newline, strlen(newline));
163 	}
164 
165 	return len;
166 }
167 
168 static void report_message_lost(struct printk_safe_seq_buf *s)
169 {
170 	int lost = atomic_xchg(&s->message_lost, 0);
171 
172 	if (lost)
173 		printk_deferred("Lost %d message(s)!\n", lost);
174 }
175 
176 /*
177  * Flush data from the associated per-CPU buffer. The function
178  * can be called either via IRQ work or independently.
179  */
180 static void __printk_safe_flush(struct irq_work *work)
181 {
182 	static raw_spinlock_t read_lock =
183 		__RAW_SPIN_LOCK_INITIALIZER(read_lock);
184 	struct printk_safe_seq_buf *s =
185 		container_of(work, struct printk_safe_seq_buf, work);
186 	unsigned long flags;
187 	size_t len;
188 	int i;
189 
190 	/*
191 	 * The lock has two functions. First, one reader has to flush all
192 	 * available message to make the lockless synchronization with
193 	 * writers easier. Second, we do not want to mix messages from
194 	 * different CPUs. This is especially important when printing
195 	 * a backtrace.
196 	 */
197 	raw_spin_lock_irqsave(&read_lock, flags);
198 
199 	i = 0;
200 more:
201 	len = atomic_read(&s->len);
202 
203 	/*
204 	 * This is just a paranoid check that nobody has manipulated
205 	 * the buffer an unexpected way. If we printed something then
206 	 * @len must only increase. Also it should never overflow the
207 	 * buffer size.
208 	 */
209 	if ((i && i >= len) || len > sizeof(s->buffer)) {
210 		const char *msg = "printk_safe_flush: internal error\n";
211 
212 		printk_safe_flush_line(msg, strlen(msg));
213 		len = 0;
214 	}
215 
216 	if (!len)
217 		goto out; /* Someone else has already flushed the buffer. */
218 
219 	/* Make sure that data has been written up to the @len */
220 	smp_rmb();
221 	i += printk_safe_flush_buffer(s->buffer + i, len - i);
222 
223 	/*
224 	 * Check that nothing has got added in the meantime and truncate
225 	 * the buffer. Note that atomic_cmpxchg() is an implicit memory
226 	 * barrier that makes sure that the data were copied before
227 	 * updating s->len.
228 	 */
229 	if (atomic_cmpxchg(&s->len, len, 0) != len)
230 		goto more;
231 
232 out:
233 	report_message_lost(s);
234 	raw_spin_unlock_irqrestore(&read_lock, flags);
235 }
236 
237 /**
238  * printk_safe_flush - flush all per-cpu nmi buffers.
239  *
240  * The buffers are flushed automatically via IRQ work. This function
241  * is useful only when someone wants to be sure that all buffers have
242  * been flushed at some point.
243  */
244 void printk_safe_flush(void)
245 {
246 	int cpu;
247 
248 	for_each_possible_cpu(cpu) {
249 #ifdef CONFIG_PRINTK_NMI
250 		__printk_safe_flush(&per_cpu(nmi_print_seq, cpu).work);
251 #endif
252 		__printk_safe_flush(&per_cpu(safe_print_seq, cpu).work);
253 	}
254 }
255 
256 /**
257  * printk_safe_flush_on_panic - flush all per-cpu nmi buffers when the system
258  *	goes down.
259  *
260  * Similar to printk_safe_flush() but it can be called even in NMI context when
261  * the system goes down. It does the best effort to get NMI messages into
262  * the main ring buffer.
263  *
264  * Note that it could try harder when there is only one CPU online.
265  */
266 void printk_safe_flush_on_panic(void)
267 {
268 	/*
269 	 * Make sure that we could access the main ring buffer.
270 	 * Do not risk a double release when more CPUs are up.
271 	 */
272 	if (raw_spin_is_locked(&logbuf_lock)) {
273 		if (num_online_cpus() > 1)
274 			return;
275 
276 		debug_locks_off();
277 		raw_spin_lock_init(&logbuf_lock);
278 	}
279 
280 	printk_safe_flush();
281 }
282 
283 #ifdef CONFIG_PRINTK_NMI
284 /*
285  * Safe printk() for NMI context. It uses a per-CPU buffer to
286  * store the message. NMIs are not nested, so there is always only
287  * one writer running. But the buffer might get flushed from another
288  * CPU, so we need to be careful.
289  */
290 static __printf(1, 0) int vprintk_nmi(const char *fmt, va_list args)
291 {
292 	struct printk_safe_seq_buf *s = this_cpu_ptr(&nmi_print_seq);
293 
294 	return printk_safe_log_store(s, fmt, args);
295 }
296 
297 void notrace printk_nmi_enter(void)
298 {
299 	this_cpu_or(printk_context, PRINTK_NMI_CONTEXT_MASK);
300 }
301 
302 void notrace printk_nmi_exit(void)
303 {
304 	this_cpu_and(printk_context, ~PRINTK_NMI_CONTEXT_MASK);
305 }
306 
307 /*
308  * Marks a code that might produce many messages in NMI context
309  * and the risk of losing them is more critical than eventual
310  * reordering.
311  *
312  * It has effect only when called in NMI context. Then printk()
313  * will try to store the messages into the main logbuf directly
314  * and use the per-CPU buffers only as a fallback when the lock
315  * is not available.
316  */
317 void printk_nmi_direct_enter(void)
318 {
319 	if (this_cpu_read(printk_context) & PRINTK_NMI_CONTEXT_MASK)
320 		this_cpu_or(printk_context, PRINTK_NMI_DIRECT_CONTEXT_MASK);
321 }
322 
323 void printk_nmi_direct_exit(void)
324 {
325 	this_cpu_and(printk_context, ~PRINTK_NMI_DIRECT_CONTEXT_MASK);
326 }
327 
328 #else
329 
330 static __printf(1, 0) int vprintk_nmi(const char *fmt, va_list args)
331 {
332 	return 0;
333 }
334 
335 #endif /* CONFIG_PRINTK_NMI */
336 
337 /*
338  * Lock-less printk(), to avoid deadlocks should the printk() recurse
339  * into itself. It uses a per-CPU buffer to store the message, just like
340  * NMI.
341  */
342 static __printf(1, 0) int vprintk_safe(const char *fmt, va_list args)
343 {
344 	struct printk_safe_seq_buf *s = this_cpu_ptr(&safe_print_seq);
345 
346 	return printk_safe_log_store(s, fmt, args);
347 }
348 
349 /* Can be preempted by NMI. */
350 void __printk_safe_enter(void)
351 {
352 	this_cpu_inc(printk_context);
353 }
354 
355 /* Can be preempted by NMI. */
356 void __printk_safe_exit(void)
357 {
358 	this_cpu_dec(printk_context);
359 }
360 
361 __printf(1, 0) int vprintk_func(const char *fmt, va_list args)
362 {
363 	/*
364 	 * Try to use the main logbuf even in NMI. But avoid calling console
365 	 * drivers that might have their own locks.
366 	 */
367 	if ((this_cpu_read(printk_context) & PRINTK_NMI_DIRECT_CONTEXT_MASK) &&
368 	    raw_spin_trylock(&logbuf_lock)) {
369 		int len;
370 
371 		len = vprintk_store(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
372 		raw_spin_unlock(&logbuf_lock);
373 		defer_console_output();
374 		return len;
375 	}
376 
377 	/* Use extra buffer in NMI when logbuf_lock is taken or in safe mode. */
378 	if (this_cpu_read(printk_context) & PRINTK_NMI_CONTEXT_MASK)
379 		return vprintk_nmi(fmt, args);
380 
381 	/* Use extra buffer to prevent a recursion deadlock in safe mode. */
382 	if (this_cpu_read(printk_context) & PRINTK_SAFE_CONTEXT_MASK)
383 		return vprintk_safe(fmt, args);
384 
385 	/* No obstacles. */
386 	return vprintk_default(fmt, args);
387 }
388 
389 void __init printk_safe_init(void)
390 {
391 	int cpu;
392 
393 	for_each_possible_cpu(cpu) {
394 		struct printk_safe_seq_buf *s;
395 
396 		s = &per_cpu(safe_print_seq, cpu);
397 		init_irq_work(&s->work, __printk_safe_flush);
398 
399 #ifdef CONFIG_PRINTK_NMI
400 		s = &per_cpu(nmi_print_seq, cpu);
401 		init_irq_work(&s->work, __printk_safe_flush);
402 #endif
403 	}
404 
405 	/*
406 	 * In the highly unlikely event that a NMI were to trigger at
407 	 * this moment. Make sure IRQ work is set up before this
408 	 * variable is set.
409 	 */
410 	barrier();
411 	printk_safe_irq_ready = 1;
412 
413 	/* Flush pending messages that did not have scheduled IRQ works. */
414 	printk_safe_flush();
415 }
416