10264c8c9SSteven Rostedt (VMware) /* SPDX-License-Identifier: GPL-2.0 */
20264c8c9SSteven Rostedt (VMware) #ifndef _LINUX_TRACE_RECURSION_H
30264c8c9SSteven Rostedt (VMware) #define _LINUX_TRACE_RECURSION_H
40264c8c9SSteven Rostedt (VMware) 
50264c8c9SSteven Rostedt (VMware) #include <linux/interrupt.h>
60264c8c9SSteven Rostedt (VMware) #include <linux/sched.h>
70264c8c9SSteven Rostedt (VMware) 
80264c8c9SSteven Rostedt (VMware) #ifdef CONFIG_TRACING
90264c8c9SSteven Rostedt (VMware) 
100264c8c9SSteven Rostedt (VMware) /* Only current can touch trace_recursion */
110264c8c9SSteven Rostedt (VMware) 
120264c8c9SSteven Rostedt (VMware) /*
130264c8c9SSteven Rostedt (VMware)  * For function tracing recursion:
140264c8c9SSteven Rostedt (VMware)  *  The order of these bits are important.
150264c8c9SSteven Rostedt (VMware)  *
160264c8c9SSteven Rostedt (VMware)  *  When function tracing occurs, the following steps are made:
170264c8c9SSteven Rostedt (VMware)  *   If arch does not support a ftrace feature:
180264c8c9SSteven Rostedt (VMware)  *    call internal function (uses INTERNAL bits) which calls...
190264c8c9SSteven Rostedt (VMware)  *   The function callback, which can use the FTRACE bits to
200264c8c9SSteven Rostedt (VMware)  *    check for recursion.
210264c8c9SSteven Rostedt (VMware)  */
220264c8c9SSteven Rostedt (VMware) enum {
230264c8c9SSteven Rostedt (VMware) 	/* Function recursion bits */
240264c8c9SSteven Rostedt (VMware) 	TRACE_FTRACE_BIT,
250264c8c9SSteven Rostedt (VMware) 	TRACE_FTRACE_NMI_BIT,
260264c8c9SSteven Rostedt (VMware) 	TRACE_FTRACE_IRQ_BIT,
270264c8c9SSteven Rostedt (VMware) 	TRACE_FTRACE_SIRQ_BIT,
28ed65df63SSteven Rostedt (VMware) 	TRACE_FTRACE_TRANSITION_BIT,
290264c8c9SSteven Rostedt (VMware) 
30ed65df63SSteven Rostedt (VMware) 	/* Internal use recursion bits */
310264c8c9SSteven Rostedt (VMware) 	TRACE_INTERNAL_BIT,
320264c8c9SSteven Rostedt (VMware) 	TRACE_INTERNAL_NMI_BIT,
330264c8c9SSteven Rostedt (VMware) 	TRACE_INTERNAL_IRQ_BIT,
340264c8c9SSteven Rostedt (VMware) 	TRACE_INTERNAL_SIRQ_BIT,
35ed65df63SSteven Rostedt (VMware) 	TRACE_INTERNAL_TRANSITION_BIT,
360264c8c9SSteven Rostedt (VMware) 
370264c8c9SSteven Rostedt (VMware) 	TRACE_BRANCH_BIT,
380264c8c9SSteven Rostedt (VMware) /*
390264c8c9SSteven Rostedt (VMware)  * Abuse of the trace_recursion.
400264c8c9SSteven Rostedt (VMware)  * As we need a way to maintain state if we are tracing the function
410264c8c9SSteven Rostedt (VMware)  * graph in irq because we want to trace a particular function that
420264c8c9SSteven Rostedt (VMware)  * was called in irq context but we have irq tracing off. Since this
430264c8c9SSteven Rostedt (VMware)  * can only be modified by current, we can reuse trace_recursion.
440264c8c9SSteven Rostedt (VMware)  */
450264c8c9SSteven Rostedt (VMware) 	TRACE_IRQ_BIT,
460264c8c9SSteven Rostedt (VMware) 
470264c8c9SSteven Rostedt (VMware) 	/* Set if the function is in the set_graph_function file */
480264c8c9SSteven Rostedt (VMware) 	TRACE_GRAPH_BIT,
490264c8c9SSteven Rostedt (VMware) 
500264c8c9SSteven Rostedt (VMware) 	/*
510264c8c9SSteven Rostedt (VMware) 	 * In the very unlikely case that an interrupt came in
520264c8c9SSteven Rostedt (VMware) 	 * at a start of graph tracing, and we want to trace
530264c8c9SSteven Rostedt (VMware) 	 * the function in that interrupt, the depth can be greater
540264c8c9SSteven Rostedt (VMware) 	 * than zero, because of the preempted start of a previous
550264c8c9SSteven Rostedt (VMware) 	 * trace. In an even more unlikely case, depth could be 2
560264c8c9SSteven Rostedt (VMware) 	 * if a softirq interrupted the start of graph tracing,
570264c8c9SSteven Rostedt (VMware) 	 * followed by an interrupt preempting a start of graph
580264c8c9SSteven Rostedt (VMware) 	 * tracing in the softirq, and depth can even be 3
590264c8c9SSteven Rostedt (VMware) 	 * if an NMI came in at the start of an interrupt function
600264c8c9SSteven Rostedt (VMware) 	 * that preempted a softirq start of a function that
610264c8c9SSteven Rostedt (VMware) 	 * preempted normal context!!!! Luckily, it can't be
620264c8c9SSteven Rostedt (VMware) 	 * greater than 3, so the next two bits are a mask
630264c8c9SSteven Rostedt (VMware) 	 * of what the depth is when we set TRACE_GRAPH_BIT
640264c8c9SSteven Rostedt (VMware) 	 */
650264c8c9SSteven Rostedt (VMware) 
660264c8c9SSteven Rostedt (VMware) 	TRACE_GRAPH_DEPTH_START_BIT,
670264c8c9SSteven Rostedt (VMware) 	TRACE_GRAPH_DEPTH_END_BIT,
680264c8c9SSteven Rostedt (VMware) 
690264c8c9SSteven Rostedt (VMware) 	/*
700264c8c9SSteven Rostedt (VMware) 	 * To implement set_graph_notrace, if this bit is set, we ignore
710264c8c9SSteven Rostedt (VMware) 	 * function graph tracing of called functions, until the return
720264c8c9SSteven Rostedt (VMware) 	 * function is called to clear it.
730264c8c9SSteven Rostedt (VMware) 	 */
740264c8c9SSteven Rostedt (VMware) 	TRACE_GRAPH_NOTRACE_BIT,
750264c8c9SSteven Rostedt (VMware) 
76773c1670SSteven Rostedt (VMware) 	/* Used to prevent recursion recording from recursing. */
77773c1670SSteven Rostedt (VMware) 	TRACE_RECORD_RECURSION_BIT,
780264c8c9SSteven Rostedt (VMware) };
790264c8c9SSteven Rostedt (VMware) 
800264c8c9SSteven Rostedt (VMware) #define trace_recursion_set(bit)	do { (current)->trace_recursion |= (1<<(bit)); } while (0)
810264c8c9SSteven Rostedt (VMware) #define trace_recursion_clear(bit)	do { (current)->trace_recursion &= ~(1<<(bit)); } while (0)
820264c8c9SSteven Rostedt (VMware) #define trace_recursion_test(bit)	((current)->trace_recursion & (1<<(bit)))
830264c8c9SSteven Rostedt (VMware) 
840264c8c9SSteven Rostedt (VMware) #define trace_recursion_depth() \
850264c8c9SSteven Rostedt (VMware) 	(((current)->trace_recursion >> TRACE_GRAPH_DEPTH_START_BIT) & 3)
860264c8c9SSteven Rostedt (VMware) #define trace_recursion_set_depth(depth) \
870264c8c9SSteven Rostedt (VMware) 	do {								\
880264c8c9SSteven Rostedt (VMware) 		current->trace_recursion &=				\
890264c8c9SSteven Rostedt (VMware) 			~(3 << TRACE_GRAPH_DEPTH_START_BIT);		\
900264c8c9SSteven Rostedt (VMware) 		current->trace_recursion |=				\
910264c8c9SSteven Rostedt (VMware) 			((depth) & 3) << TRACE_GRAPH_DEPTH_START_BIT;	\
920264c8c9SSteven Rostedt (VMware) 	} while (0)
930264c8c9SSteven Rostedt (VMware) 
940264c8c9SSteven Rostedt (VMware) #define TRACE_CONTEXT_BITS	4
950264c8c9SSteven Rostedt (VMware) 
960264c8c9SSteven Rostedt (VMware) #define TRACE_FTRACE_START	TRACE_FTRACE_BIT
970264c8c9SSteven Rostedt (VMware) 
980264c8c9SSteven Rostedt (VMware) #define TRACE_LIST_START	TRACE_INTERNAL_BIT
990264c8c9SSteven Rostedt (VMware) 
100ed65df63SSteven Rostedt (VMware) #define TRACE_CONTEXT_MASK	((1 << (TRACE_LIST_START + TRACE_CONTEXT_BITS)) - 1)
1010264c8c9SSteven Rostedt (VMware) 
102da5afbebSSteven Rostedt (VMware) /*
103da5afbebSSteven Rostedt (VMware)  * Used for setting context
104da5afbebSSteven Rostedt (VMware)  *  NMI     = 0
105da5afbebSSteven Rostedt (VMware)  *  IRQ     = 1
106da5afbebSSteven Rostedt (VMware)  *  SOFTIRQ = 2
107da5afbebSSteven Rostedt (VMware)  *  NORMAL  = 3
108da5afbebSSteven Rostedt (VMware)  */
109da5afbebSSteven Rostedt (VMware) enum {
110da5afbebSSteven Rostedt (VMware) 	TRACE_CTX_NMI,
111da5afbebSSteven Rostedt (VMware) 	TRACE_CTX_IRQ,
112da5afbebSSteven Rostedt (VMware) 	TRACE_CTX_SOFTIRQ,
113da5afbebSSteven Rostedt (VMware) 	TRACE_CTX_NORMAL,
114ed65df63SSteven Rostedt (VMware) 	TRACE_CTX_TRANSITION,
115da5afbebSSteven Rostedt (VMware) };
116da5afbebSSteven Rostedt (VMware) 
1170264c8c9SSteven Rostedt (VMware) static __always_inline int trace_get_context_bit(void)
1180264c8c9SSteven Rostedt (VMware) {
11991ebe8bcSSteven Rostedt (VMware) 	unsigned char bit = interrupt_context_level();
1200264c8c9SSteven Rostedt (VMware) 
1219b84fadcSSteven Rostedt (VMware) 	return TRACE_CTX_NORMAL - bit;
1220264c8c9SSteven Rostedt (VMware) }
1230264c8c9SSteven Rostedt (VMware) 
124773c1670SSteven Rostedt (VMware) #ifdef CONFIG_FTRACE_RECORD_RECURSION
125773c1670SSteven Rostedt (VMware) extern void ftrace_record_recursion(unsigned long ip, unsigned long parent_ip);
126773c1670SSteven Rostedt (VMware) # define do_ftrace_record_recursion(ip, pip)				\
127773c1670SSteven Rostedt (VMware) 	do {								\
128773c1670SSteven Rostedt (VMware) 		if (!trace_recursion_test(TRACE_RECORD_RECURSION_BIT)) { \
129773c1670SSteven Rostedt (VMware) 			trace_recursion_set(TRACE_RECORD_RECURSION_BIT); \
130773c1670SSteven Rostedt (VMware) 			ftrace_record_recursion(ip, pip);		\
131773c1670SSteven Rostedt (VMware) 			trace_recursion_clear(TRACE_RECORD_RECURSION_BIT); \
132773c1670SSteven Rostedt (VMware) 		}							\
133773c1670SSteven Rostedt (VMware) 	} while (0)
134773c1670SSteven Rostedt (VMware) #else
135773c1670SSteven Rostedt (VMware) # define do_ftrace_record_recursion(ip, pip)	do { } while (0)
136773c1670SSteven Rostedt (VMware) #endif
137773c1670SSteven Rostedt (VMware) 
138*d099dbfdSPeter Zijlstra #ifdef CONFIG_ARCH_WANTS_NO_INSTR
139*d099dbfdSPeter Zijlstra # define trace_warn_on_no_rcu(ip)					\
140*d099dbfdSPeter Zijlstra 	({								\
141*d099dbfdSPeter Zijlstra 		bool __ret = !rcu_is_watching();			\
142*d099dbfdSPeter Zijlstra 		if (__ret && !trace_recursion_test(TRACE_RECORD_RECURSION_BIT)) { \
143*d099dbfdSPeter Zijlstra 			trace_recursion_set(TRACE_RECORD_RECURSION_BIT); \
144*d099dbfdSPeter Zijlstra 			WARN_ONCE(true, "RCU not on for: %pS\n", (void *)ip); \
145*d099dbfdSPeter Zijlstra 			trace_recursion_clear(TRACE_RECORD_RECURSION_BIT); \
146*d099dbfdSPeter Zijlstra 		}							\
147*d099dbfdSPeter Zijlstra 		__ret;							\
148*d099dbfdSPeter Zijlstra 	})
149*d099dbfdSPeter Zijlstra #else
150*d099dbfdSPeter Zijlstra # define trace_warn_on_no_rcu(ip)	false
151*d099dbfdSPeter Zijlstra #endif
152*d099dbfdSPeter Zijlstra 
153ce5e4803S王贇 /*
154ce5e4803S王贇  * Preemption is promised to be disabled when return bit >= 0.
155ce5e4803S王贇  */
156773c1670SSteven Rostedt (VMware) static __always_inline int trace_test_and_set_recursion(unsigned long ip, unsigned long pip,
157ed65df63SSteven Rostedt (VMware) 							int start)
1580264c8c9SSteven Rostedt (VMware) {
1597b68621fSSteven Rostedt (VMware) 	unsigned int val = READ_ONCE(current->trace_recursion);
1600264c8c9SSteven Rostedt (VMware) 	int bit;
1610264c8c9SSteven Rostedt (VMware) 
162*d099dbfdSPeter Zijlstra 	if (trace_warn_on_no_rcu(ip))
163*d099dbfdSPeter Zijlstra 		return -1;
164*d099dbfdSPeter Zijlstra 
1650264c8c9SSteven Rostedt (VMware) 	bit = trace_get_context_bit() + start;
1660264c8c9SSteven Rostedt (VMware) 	if (unlikely(val & (1 << bit))) {
1670264c8c9SSteven Rostedt (VMware) 		/*
168bce5c81cSSteven Rostedt (VMware) 		 * If an interrupt occurs during a trace, and another trace
169bce5c81cSSteven Rostedt (VMware) 		 * happens in that interrupt but before the preempt_count is
170bce5c81cSSteven Rostedt (VMware) 		 * updated to reflect the new interrupt context, then this
171bce5c81cSSteven Rostedt (VMware) 		 * will think a recursion occurred, and the event will be dropped.
172bce5c81cSSteven Rostedt (VMware) 		 * Let a single instance happen via the TRANSITION_BIT to
173bce5c81cSSteven Rostedt (VMware) 		 * not drop those events.
1740264c8c9SSteven Rostedt (VMware) 		 */
175ed65df63SSteven Rostedt (VMware) 		bit = TRACE_CTX_TRANSITION + start;
1767b68621fSSteven Rostedt (VMware) 		if (val & (1 << bit)) {
177773c1670SSteven Rostedt (VMware) 			do_ftrace_record_recursion(ip, pip);
1780264c8c9SSteven Rostedt (VMware) 			return -1;
179773c1670SSteven Rostedt (VMware) 		}
1807b68621fSSteven Rostedt (VMware) 	}
1810264c8c9SSteven Rostedt (VMware) 
1820264c8c9SSteven Rostedt (VMware) 	val |= 1 << bit;
1830264c8c9SSteven Rostedt (VMware) 	current->trace_recursion = val;
1840264c8c9SSteven Rostedt (VMware) 	barrier();
1850264c8c9SSteven Rostedt (VMware) 
186ce5e4803S王贇 	preempt_disable_notrace();
187ce5e4803S王贇 
188ed65df63SSteven Rostedt (VMware) 	return bit;
1890264c8c9SSteven Rostedt (VMware) }
1900264c8c9SSteven Rostedt (VMware) 
191ce5e4803S王贇 /*
192ce5e4803S王贇  * Preemption will be enabled (if it was previously enabled).
193ce5e4803S王贇  */
1940264c8c9SSteven Rostedt (VMware) static __always_inline void trace_clear_recursion(int bit)
1950264c8c9SSteven Rostedt (VMware) {
196ce5e4803S王贇 	preempt_enable_notrace();
1970264c8c9SSteven Rostedt (VMware) 	barrier();
1987b68621fSSteven Rostedt (VMware) 	trace_recursion_clear(bit);
1990264c8c9SSteven Rostedt (VMware) }
2000264c8c9SSteven Rostedt (VMware) 
2016e4eb9cbSSteven Rostedt (VMware) /**
2026e4eb9cbSSteven Rostedt (VMware)  * ftrace_test_recursion_trylock - tests for recursion in same context
2036e4eb9cbSSteven Rostedt (VMware)  *
2046e4eb9cbSSteven Rostedt (VMware)  * Use this for ftrace callbacks. This will detect if the function
2056e4eb9cbSSteven Rostedt (VMware)  * tracing recursed in the same context (normal vs interrupt),
2066e4eb9cbSSteven Rostedt (VMware)  *
2076e4eb9cbSSteven Rostedt (VMware)  * Returns: -1 if a recursion happened.
208ce5e4803S王贇  *           >= 0 if no recursion.
2096e4eb9cbSSteven Rostedt (VMware)  */
210773c1670SSteven Rostedt (VMware) static __always_inline int ftrace_test_recursion_trylock(unsigned long ip,
211773c1670SSteven Rostedt (VMware) 							 unsigned long parent_ip)
2126e4eb9cbSSteven Rostedt (VMware) {
213ed65df63SSteven Rostedt (VMware) 	return trace_test_and_set_recursion(ip, parent_ip, TRACE_FTRACE_START);
2146e4eb9cbSSteven Rostedt (VMware) }
2156e4eb9cbSSteven Rostedt (VMware) 
2166e4eb9cbSSteven Rostedt (VMware) /**
2176e4eb9cbSSteven Rostedt (VMware)  * ftrace_test_recursion_unlock - called when function callback is complete
2186e4eb9cbSSteven Rostedt (VMware)  * @bit: The return of a successful ftrace_test_recursion_trylock()
2196e4eb9cbSSteven Rostedt (VMware)  *
2206e4eb9cbSSteven Rostedt (VMware)  * This is used at the end of a ftrace callback.
2216e4eb9cbSSteven Rostedt (VMware)  */
2226e4eb9cbSSteven Rostedt (VMware) static __always_inline void ftrace_test_recursion_unlock(int bit)
2236e4eb9cbSSteven Rostedt (VMware) {
2246e4eb9cbSSteven Rostedt (VMware) 	trace_clear_recursion(bit);
2256e4eb9cbSSteven Rostedt (VMware) }
2266e4eb9cbSSteven Rostedt (VMware) 
2270264c8c9SSteven Rostedt (VMware) #endif /* CONFIG_TRACING */
2280264c8c9SSteven Rostedt (VMware) #endif /* _LINUX_TRACE_RECURSION_H */
229