ftrace: Add recording of functions that caused recursion

This adds CONFIG_FTRACE_RECORD_RECURSION that will record to a file
"recursed_functions" all the functions that caused recursion while a
callback to the function tracer was running.

Link: https://lkml.kernel.org/r/20201106023548.102375687@goodmis.org

Cc: Masami Hiramatsu <mhiramat@kernel.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Guo Ren <guoren@kernel.org>
Cc: "James E.J. Bottomley" <James.Bottomley@HansenPartnership.com>
Cc: Helge Deller <deller@gmx.de>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Christian Borntraeger <borntraeger@de.ibm.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Borislav Petkov <bp@alien8.de>
Cc: x86@kernel.org
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Anton Vorontsov <anton@enomsg.org>
Cc: Colin Cross <ccross@android.com>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Jiri Kosina <jikos@kernel.org>
Cc: Miroslav Benes <mbenes@suse.cz>
Cc: Petr Mladek <pmladek@suse.com>
Cc: Joe Lawrence <joe.lawrence@redhat.com>
Cc: Kamalesh Babulal <kamalesh@linux.vnet.ibm.com>
Cc: Mauro Carvalho Chehab <mchehab+huawei@kernel.org>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: linux-doc@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Cc: linux-csky@vger.kernel.org
Cc: linux-parisc@vger.kernel.org
Cc: linuxppc-dev@lists.ozlabs.org
Cc: linux-s390@vger.kernel.org
Cc: live-patching@vger.kernel.org
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>

Documentation/trace/ftrace-uses.rst

@@ -118,7 +118,7 @@ can help in this regard. If you start your code with:
 
 	int bit;
 
-	bit = ftrace_test_recursion_trylock();
+	bit = ftrace_test_recursion_trylock(ip, parent_ip);
 	if (bit < 0)
 		return;
 
@@ -130,7 +130,9 @@ The code in between will be safe to use, even if it ends up calling a
 function that the callback is tracing. Note, on success,
 ftrace_test_recursion_trylock() will disable preemption, and the
 ftrace_test_recursion_unlock() will enable it again (if it was previously
-enabled).
+enabled). The instruction pointer (ip) and its parent (parent_ip) is passed to
+ftrace_test_recursion_trylock() to record where the recursion happened
+(if CONFIG_FTRACE_RECORD_RECURSION is set).
 
 Alternatively, if the FTRACE_OPS_FL_RECURSION flag is set on the ftrace_ops
 (as explained below), then a helper trampoline will be used to test

arch/csky/kernel/probes/ftrace.c

@@ -18,7 +18,7 @@ void kprobe_ftrace_handler(unsigned long ip, unsigned long parent_ip,
 	struct kprobe *p;
 	struct kprobe_ctlblk *kcb;
 
-	bit = ftrace_test_recursion_trylock();
+	bit = ftrace_test_recursion_trylock(ip, parent_ip);
 	if (bit < 0)
 		return;
 

arch/parisc/kernel/ftrace.c

@@ -210,7 +210,7 @@ void kprobe_ftrace_handler(unsigned long ip, unsigned long parent_ip,
 	struct kprobe *p;
 	int bit;
 
-	bit = ftrace_test_recursion_trylock();
+	bit = ftrace_test_recursion_trylock(ip, parent_ip);
 	if (bit < 0)
 		return;
 

arch/powerpc/kernel/kprobes-ftrace.c

@@ -20,7 +20,7 @@ void kprobe_ftrace_handler(unsigned long nip, unsigned long parent_nip,
 	struct kprobe_ctlblk *kcb;
 	int bit;
 
-	bit = ftrace_test_recursion_trylock();
+	bit = ftrace_test_recursion_trylock(nip, parent_nip);
 	if (bit < 0)
 		return;
 

arch/s390/kernel/ftrace.c

@@ -204,7 +204,7 @@ void kprobe_ftrace_handler(unsigned long ip, unsigned long parent_ip,
 	struct kprobe *p;
 	int bit;
 
-	bit = ftrace_test_recursion_trylock();
+	bit = ftrace_test_recursion_trylock(ip, parent_ip);
 	if (bit < 0)
 		return;
 

arch/x86/kernel/kprobes/ftrace.c

@@ -20,7 +20,7 @@ void kprobe_ftrace_handler(unsigned long ip, unsigned long parent_ip,
 	struct kprobe_ctlblk *kcb;
 	int bit;
 
-	bit = ftrace_test_recursion_trylock();
+	bit = ftrace_test_recursion_trylock(ip, parent_ip);
 	if (bit < 0)
 		return;
 

fs/pstore/ftrace.c

@@ -41,7 +41,7 @@ static void notrace pstore_ftrace_call(unsigned long ip,
 	if (unlikely(oops_in_progress))
 		return;
 
-	bit = ftrace_test_recursion_trylock();
+	bit = ftrace_test_recursion_trylock(ip, parent_ip);
 	if (bit < 0)
 		return;
 

include/linux/trace_recursion.h

@@ -91,6 +91,9 @@ enum {
 	 * not be correct. Allow for a single recursion to cover this case.
 	 */
 	TRACE_TRANSITION_BIT,
+
+	/* Used to prevent recursion recording from recursing. */
+	TRACE_RECORD_RECURSION_BIT,
 };
 
 #define trace_recursion_set(bit)	do { (current)->trace_recursion |= (1<<(bit)); } while (0)
@@ -142,7 +145,22 @@ static __always_inline int trace_get_context_bit(void)
 			pc & HARDIRQ_MASK ? TRACE_CTX_IRQ : TRACE_CTX_SOFTIRQ;
 }
 
-static __always_inline int trace_test_and_set_recursion(int start, int max)
+#ifdef CONFIG_FTRACE_RECORD_RECURSION
+extern void ftrace_record_recursion(unsigned long ip, unsigned long parent_ip);
+# define do_ftrace_record_recursion(ip, pip)				\
+	do {								\
+		if (!trace_recursion_test(TRACE_RECORD_RECURSION_BIT)) { \
+			trace_recursion_set(TRACE_RECORD_RECURSION_BIT); \
+			ftrace_record_recursion(ip, pip);		\
+			trace_recursion_clear(TRACE_RECORD_RECURSION_BIT); \
+		}							\
+	} while (0)
+#else
+# define do_ftrace_record_recursion(ip, pip)	do { } while (0)
+#endif
+
+static __always_inline int trace_test_and_set_recursion(unsigned long ip, unsigned long pip,
+							int start, int max)
 {
 	unsigned int val = current->trace_recursion;
 	int bit;
@@ -158,8 +176,10 @@ static __always_inline int trace_test_and_set_recursion(int start, int max)
 		 * a switch between contexts. Allow for a single recursion.
 		 */
 		bit = TRACE_TRANSITION_BIT;
-		if (trace_recursion_test(bit))
+		if (trace_recursion_test(bit)) {
+			do_ftrace_record_recursion(ip, pip);
 			return -1;
+		}
 		trace_recursion_set(bit);
 		barrier();
 		return bit + 1;
@@ -199,9 +219,10 @@ static __always_inline void trace_clear_recursion(int bit)
  * Returns: -1 if a recursion happened.
  *           >= 0 if no recursion
  */
-static __always_inline int ftrace_test_recursion_trylock(void)
+static __always_inline int ftrace_test_recursion_trylock(unsigned long ip,
+							 unsigned long parent_ip)
 {
-	return trace_test_and_set_recursion(TRACE_FTRACE_START, TRACE_FTRACE_MAX);
+	return trace_test_and_set_recursion(ip, parent_ip, TRACE_FTRACE_START, TRACE_FTRACE_MAX);
 }
 
 /**

kernel/livepatch/patch.c

@@ -49,7 +49,7 @@ static void notrace klp_ftrace_handler(unsigned long ip,
 
 	ops = container_of(fops, struct klp_ops, fops);
 
-	bit = ftrace_test_recursion_trylock();
+	bit = ftrace_test_recursion_trylock(ip, parent_ip);
 	if (WARN_ON_ONCE(bit < 0))
 		return;
 	/*

kernel/trace/Kconfig

@@ -727,6 +727,31 @@ config TRACE_EVAL_MAP_FILE
 
 	If unsure, say N.
 
+config FTRACE_RECORD_RECURSION
+	bool "Record functions that recurse in function tracing"
+	depends on FUNCTION_TRACER
+	help
+	  All callbacks that attach to the function tracing have some sort
+	  of protection against recursion. Even though the protection exists,
+	  it adds overhead. This option will create a file in the tracefs
+	  file system called "recursed_functions" that will list the functions
+	  that triggered a recursion.
+
+	  This will add more overhead to cases that have recursion.
+
+	  If unsure, say N
+
+config FTRACE_RECORD_RECURSION_SIZE
+	int "Max number of recursed functions to record"
+	default	128
+	depends on FTRACE_RECORD_RECURSION
+	help
+	  This defines the limit of number of functions that can be
+	  listed in the "recursed_functions" file, that lists all
+	  the functions that caused a recursion to happen.
+	  This file can be reset, but the limit can not change in
+	  size at runtime.
+
 config GCOV_PROFILE_FTRACE
 	bool "Enable GCOV profiling on ftrace subsystem"
 	depends on GCOV_KERNEL

kernel/trace/Makefile

@@ -92,6 +92,7 @@ obj-$(CONFIG_DYNAMIC_EVENTS) += trace_dynevent.o
 obj-$(CONFIG_PROBE_EVENTS) += trace_probe.o
 obj-$(CONFIG_UPROBE_EVENTS) += trace_uprobe.o
 obj-$(CONFIG_BOOTTIME_TRACING) += trace_boot.o
+obj-$(CONFIG_FTRACE_RECORD_RECURSION) += trace_recursion_record.o
 
 obj-$(CONFIG_TRACEPOINT_BENCHMARK) += trace_benchmark.o
 

kernel/trace/ftrace.c

@@ -6918,7 +6918,7 @@ __ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
 	struct ftrace_ops *op;
 	int bit;
 
-	bit = trace_test_and_set_recursion(TRACE_LIST_START, TRACE_LIST_MAX);
+	bit = trace_test_and_set_recursion(ip, parent_ip, TRACE_LIST_START, TRACE_LIST_MAX);
 	if (bit < 0)
 		return;
 
@@ -6993,7 +6993,7 @@ static void ftrace_ops_assist_func(unsigned long ip, unsigned long parent_ip,
 {
 	int bit;
 
-	bit = trace_test_and_set_recursion(TRACE_LIST_START, TRACE_LIST_MAX);
+	bit = trace_test_and_set_recursion(ip, parent_ip, TRACE_LIST_START, TRACE_LIST_MAX);
 	if (bit < 0)
 		return;
 

kernel/trace/trace_event_perf.c

@@ -447,7 +447,7 @@ perf_ftrace_function_call(unsigned long ip, unsigned long parent_ip,
 	if ((unsigned long)ops->private != smp_processor_id())
 		return;
 
-	bit = ftrace_test_recursion_trylock();
+	bit = ftrace_test_recursion_trylock(ip, parent_ip);
 	if (bit < 0)
 		return;
 

kernel/trace/trace_functions.c

@@ -141,7 +141,7 @@ function_trace_call(unsigned long ip, unsigned long parent_ip,
 	if (unlikely(!tr->function_enabled))
 		return;
 
-	bit = ftrace_test_recursion_trylock();
+	bit = ftrace_test_recursion_trylock(ip, parent_ip);
 	if (bit < 0)
 		return;
 

kernel/trace/trace_output.c

@@ -353,8 +353,8 @@ static inline const char *kretprobed(const char *name)
 }
 #endif /* CONFIG_KRETPROBES */
 
-static void
-seq_print_sym(struct trace_seq *s, unsigned long address, bool offset)
+void
+trace_seq_print_sym(struct trace_seq *s, unsigned long address, bool offset)
 {
 #ifdef CONFIG_KALLSYMS
 	char str[KSYM_SYMBOL_LEN];
@@ -420,7 +420,7 @@ seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
 		goto out;
 	}
 
-	seq_print_sym(s, ip, sym_flags & TRACE_ITER_SYM_OFFSET);
+	trace_seq_print_sym(s, ip, sym_flags & TRACE_ITER_SYM_OFFSET);
 
 	if (sym_flags & TRACE_ITER_SYM_ADDR)
 		trace_seq_printf(s, " <" IP_FMT ">", ip);

kernel/trace/trace_output.h

@@ -16,6 +16,7 @@ extern int
 seq_print_ip_sym(struct trace_seq *s, unsigned long ip,
 		unsigned long sym_flags);
 
+extern void trace_seq_print_sym(struct trace_seq *s, unsigned long address, bool offset);
 extern int trace_print_context(struct trace_iterator *iter);
 extern int trace_print_lat_context(struct trace_iterator *iter);
 

kernel/trace/trace_recursion_record.c

+// SPDX-License-Identifier: GPL-2.0
+
+#include <linux/seq_file.h>
+#include <linux/kallsyms.h>
+#include <linux/module.h>
+#include <linux/ftrace.h>
+#include <linux/fs.h>
+
+#include "trace_output.h"
+
+struct recursed_functions {
+	unsigned long		ip;
+	unsigned long		parent_ip;
+};
+
+static struct recursed_functions recursed_functions[CONFIG_FTRACE_RECORD_RECURSION_SIZE];
+static atomic_t nr_records;
+
+/*
+ * Cache the last found function. Yes, updates to this is racey, but
+ * so is memory cache ;-)
+ */
+static unsigned long cached_function;
+
+void ftrace_record_recursion(unsigned long ip, unsigned long parent_ip)
+{
+	int index = 0;
+	int i;
+	unsigned long old;
+
+ again:
+	/* First check the last one recorded */
+	if (ip == cached_function)
+		return;
+
+	i = atomic_read(&nr_records);
+	/* nr_records is -1 when clearing records */
+	smp_mb__after_atomic();
+	if (i < 0)
+		return;
+
+	/*
+	 * If there's two writers and this writer comes in second,
+	 * the cmpxchg() below to update the ip will fail. Then this
+	 * writer will try again. It is possible that index will now
+	 * be greater than nr_records. This is because the writer
+	 * that succeeded has not updated the nr_records yet.
+	 * This writer could keep trying again until the other writer
+	 * updates nr_records. But if the other writer takes an
+	 * interrupt, and that interrupt locks up that CPU, we do
+	 * not want this CPU to lock up due to the recursion protection,
+	 * and have a bug report showing this CPU as the cause of
+	 * locking up the computer. To not lose this record, this
+	 * writer will simply use the next position to update the
+	 * recursed_functions, and it will update the nr_records
+	 * accordingly.
+	 */
+	if (index < i)
+		index = i;
+	if (index >= CONFIG_FTRACE_RECORD_RECURSION_SIZE)
+		return;
+
+	for (i = index - 1; i >= 0; i--) {
+		if (recursed_functions[i].ip == ip) {
+			cached_function = ip;
+			return;
+		}
+	}
+
+	cached_function = ip;
+
+	/*
+	 * We only want to add a function if it hasn't been added before.
+	 * Add to the current location before incrementing the count.
+	 * If it fails to add, then increment the index (save in i)
+	 * and try again.
+	 */
+	old = cmpxchg(&recursed_functions[index].ip, 0, ip);
+	if (old != 0) {
+		/* Did something else already added this for us? */
+		if (old == ip)
+			return;
+		/* Try the next location (use i for the next index) */
+		index++;
+		goto again;
+	}
+
+	recursed_functions[index].parent_ip = parent_ip;
+
+	/*
+	 * It's still possible that we could race with the clearing
+	 *    CPU0                                    CPU1
+	 *    ----                                    ----
+	 *                                       ip = func
+	 *  nr_records = -1;
+	 *  recursed_functions[0] = 0;
+	 *                                       i = -1
+	 *                                       if (i < 0)
+	 *  nr_records = 0;
+	 *  (new recursion detected)
+	 *      recursed_functions[0] = func
+	 *                                            cmpxchg(recursed_functions[0],
+	 *                                                    func, 0)
+	 *
+	 * But the worse that could happen is that we get a zero in
+	 * the recursed_functions array, and it's likely that "func" will
+	 * be recorded again.
+	 */
+	i = atomic_read(&nr_records);
+	smp_mb__after_atomic();
+	if (i < 0)
+		cmpxchg(&recursed_functions[index].ip, ip, 0);
+	else if (i <= index)
+		atomic_cmpxchg(&nr_records, i, index + 1);
+}
+EXPORT_SYMBOL_GPL(ftrace_record_recursion);
+
+static DEFINE_MUTEX(recursed_function_lock);
+static struct trace_seq *tseq;
+
+static void *recursed_function_seq_start(struct seq_file *m, loff_t *pos)
+{
+	void *ret = NULL;
+	int index;
+
+	mutex_lock(&recursed_function_lock);
+	index = atomic_read(&nr_records);
+	if (*pos < index) {
+		ret = &recursed_functions[*pos];
+	}
+
+	tseq = kzalloc(sizeof(*tseq), GFP_KERNEL);
+	if (!tseq)
+		return ERR_PTR(-ENOMEM);
+
+	trace_seq_init(tseq);
+
+	return ret;
+}
+
+static void *recursed_function_seq_next(struct seq_file *m, void *v, loff_t *pos)
+{
+	int index;
+	int p;
+
+	index = atomic_read(&nr_records);
+	p = ++(*pos);
+
+	return p < index ? &recursed_functions[p] : NULL;
+}
+
+static void recursed_function_seq_stop(struct seq_file *m, void *v)
+{
+	kfree(tseq);
+	mutex_unlock(&recursed_function_lock);
+}
+
+static int recursed_function_seq_show(struct seq_file *m, void *v)
+{
+	struct recursed_functions *record = v;
+	int ret = 0;
+
+	if (record) {
+		trace_seq_print_sym(tseq, record->parent_ip, true);
+		trace_seq_puts(tseq, ":\t");
+		trace_seq_print_sym(tseq, record->ip, true);
+		trace_seq_putc(tseq, '\n');
+		ret = trace_print_seq(m, tseq);
+	}
+
+	return ret;
+}
+
+static const struct seq_operations recursed_function_seq_ops = {
+	.start  = recursed_function_seq_start,
+	.next   = recursed_function_seq_next,
+	.stop   = recursed_function_seq_stop,
+	.show   = recursed_function_seq_show
+};
+
+static int recursed_function_open(struct inode *inode, struct file *file)
+{
+	int ret = 0;
+
+	mutex_lock(&recursed_function_lock);
+	/* If this file was opened for write, then erase contents */
+	if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
+		/* disable updating records */
+		atomic_set(&nr_records, -1);
+		smp_mb__after_atomic();
+		memset(recursed_functions, 0, sizeof(recursed_functions));
+		smp_wmb();
+		/* enable them again */
+		atomic_set(&nr_records, 0);
+	}
+	if (file->f_mode & FMODE_READ)
+		ret = seq_open(file, &recursed_function_seq_ops);
+	mutex_unlock(&recursed_function_lock);
+
+	return ret;
+}
+
+static ssize_t recursed_function_write(struct file *file,
+				       const char __user *buffer,
+				       size_t count, loff_t *ppos)
+{
+	return count;
+}
+
+static int recursed_function_release(struct inode *inode, struct file *file)
+{
+	if (file->f_mode & FMODE_READ)
+		seq_release(inode, file);
+	return 0;
+}
+
+static const struct file_operations recursed_functions_fops = {
+	.open           = recursed_function_open,
+	.write		= recursed_function_write,
+	.read           = seq_read,
+	.llseek         = seq_lseek,
+	.release        = recursed_function_release,
+};
+
+__init static int create_recursed_functions(void)
+{
+	struct dentry *dentry;
+
+	dentry = trace_create_file("recursed_functions", 0644, NULL, NULL,
+				   &recursed_functions_fops);
+	if (!dentry)
+		pr_warn("WARNING: Failed to create recursed_functions\n");
+	return 0;
+}
+
+fs_initcall(create_recursed_functions);