perf core: Separate accounting of contexts and real addresses in a stack trace

The perf_sample->ip_callchain->nr value includes all the entries in the
ip_callchain->ip[] array, real addresses and PERF_CONTEXT_{KERNEL,USER,etc},
while what the user expects is that what is in the kernel.perf_event_max_stack
sysctl or in the upcoming per event perf_event_attr.sample_max_stack knob be
honoured in terms of IP addresses in the stack trace.

So allocate a bunch of extra entries for contexts, and do the accounting
via perf_callchain_entry_ctx struct members.

A new sysctl, kernel.perf_event_max_contexts_per_stack is also
introduced for investigating possible bugs in the callchain
implementation by some arch.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Masami Hiramatsu <mhiramat@kernel.org>
Cc: Milian Wolff <milian.wolff@kdab.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: Wang Nan <wangnan0@huawei.com>
Cc: Zefan Li <lizefan@huawei.com>
Link: http://lkml.kernel.org/n/tip-3b4wnqk340c4sg4gwkfdi9yk@git.kernel.orgSigned-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

Documentation/sysctl/kernel.txt

@@ -61,6 +61,7 @@ show up in /proc/sys/kernel:
 - perf_cpu_time_max_percent
 - perf_event_paranoid
 - perf_event_max_stack
+- perf_event_max_contexts_per_stack
 - pid_max
 - powersave-nap               [ PPC only ]
 - printk
@@ -668,6 +669,19 @@ The default value is 127.
 
 ==============================================================
 
+perf_event_max_contexts_per_stack:
+
+Controls maximum number of stack frame context entries for
+(attr.sample_type & PERF_SAMPLE_CALLCHAIN) configured events, for
+instance, when using 'perf record -g' or 'perf trace --call-graph fp'.
+
+This can only be done when no events are in use that have callchains
+enabled, otherwise writing to this file will return -EBUSY.
+
+The default value is 8.
+
+==============================================================
+
 pid_max:
 
 PID allocation wrap value.  When the kernel's next PID value

include/linux/perf_event.h

@@ -65,6 +65,8 @@ struct perf_callchain_entry_ctx {
 	struct perf_callchain_entry *entry;
 	u32			    max_stack;
 	u32			    nr;
+	short			    contexts;
+	bool			    contexts_maxed;
 };
 
 struct perf_raw_record {
@@ -1078,12 +1080,24 @@ extern int get_callchain_buffers(void);
 extern void put_callchain_buffers(void);
 
 extern int sysctl_perf_event_max_stack;
+extern int sysctl_perf_event_max_contexts_per_stack;
 
-#define perf_callchain_store_context(ctx, context) perf_callchain_store(ctx, context)
+static inline int perf_callchain_store_context(struct perf_callchain_entry_ctx *ctx, u64 ip)
+{
+	if (ctx->contexts < sysctl_perf_event_max_contexts_per_stack) {
+		struct perf_callchain_entry *entry = ctx->entry;
+		entry->ip[entry->nr++] = ip;
+		++ctx->contexts;
+		return 0;
+	} else {
+		ctx->contexts_maxed = true;
+		return -1; /* no more room, stop walking the stack */
+	}
+}
 
 static inline int perf_callchain_store(struct perf_callchain_entry_ctx *ctx, u64 ip)
 {
-	if (ctx->nr < ctx->max_stack) {
+	if (ctx->nr < ctx->max_stack && !ctx->contexts_maxed) {
 		struct perf_callchain_entry *entry = ctx->entry;
 		entry->ip[entry->nr++] = ip;
 		++ctx->nr;

include/uapi/linux/perf_event.h

@@ -862,6 +862,7 @@ enum perf_event_type {
 };
 
 #define PERF_MAX_STACK_DEPTH		127
+#define PERF_MAX_CONTEXTS_PER_STACK	  8
 
 enum perf_callchain_context {
 	PERF_CONTEXT_HV			= (__u64)-32,

kernel/events/callchain.c

@@ -19,11 +19,13 @@ struct callchain_cpus_entries {
 };
 
 int sysctl_perf_event_max_stack __read_mostly = PERF_MAX_STACK_DEPTH;
+int sysctl_perf_event_max_contexts_per_stack __read_mostly = PERF_MAX_CONTEXTS_PER_STACK;
 
 static inline size_t perf_callchain_entry__sizeof(void)
 {
 	return (sizeof(struct perf_callchain_entry) +
-		sizeof(__u64) * sysctl_perf_event_max_stack);
+		sizeof(__u64) * (sysctl_perf_event_max_stack +
+				 sysctl_perf_event_max_contexts_per_stack));
 }
 
 static DEFINE_PER_CPU(int, callchain_recursion[PERF_NR_CONTEXTS]);
@@ -197,6 +199,8 @@ get_perf_callchain(struct pt_regs *regs, u32 init_nr, bool kernel, bool user,
 	ctx.entry     = entry;
 	ctx.max_stack = max_stack;
 	ctx.nr	      = entry->nr = init_nr;
+	ctx.contexts       = 0;
+	ctx.contexts_maxed = false;
 
 	if (kernel && !user_mode(regs)) {
 		if (add_mark)
@@ -228,6 +232,10 @@ get_perf_callchain(struct pt_regs *regs, u32 init_nr, bool kernel, bool user,
 	return entry;
 }
 
+/*
+ * Used for sysctl_perf_event_max_stack and
+ * sysctl_perf_event_max_contexts_per_stack.
+ */
 int perf_event_max_stack_handler(struct ctl_table *table, int write,
 				 void __user *buffer, size_t *lenp, loff_t *ppos)
 {

kernel/sysctl.c

@@ -1156,6 +1156,15 @@ static struct ctl_table kern_table[] = {
 		.extra1		= &zero,
 		.extra2		= &six_hundred_forty_kb,
 	},
+	{
+		.procname	= "perf_event_max_contexts_per_stack",
+		.data		= &sysctl_perf_event_max_contexts_per_stack,
+		.maxlen		= sizeof(sysctl_perf_event_max_contexts_per_stack),
+		.mode		= 0644,
+		.proc_handler	= perf_event_max_stack_handler,
+		.extra1		= &zero,
+		.extra2		= &one_thousand,
+	},
 #endif
 #ifdef CONFIG_KMEMCHECK
 	{