misc: tegra-profiler: kprobes is no longer used

Tegra Profiler: kprobes is no longer used
Use callbacks from the scheduler instead of kprobes

Bug 1343930

Change-Id: I67589dbe972cc5c853addfb6ed82c488529eed55
Signed-off-by: Igor Nabirushkin <inabirushkin@nvidia.com>
Reviewed-on: http://git-master/r/281941
Reviewed-by: Daniel Horowitz <dhorowitz@nvidia.com>
Reviewed-by: Eric Miao <emiao@nvidia.com>
Tested-by: Eric Miao <emiao@nvidia.com>
GVS: Gerrit_Virtual_Submit
Reviewed-by: Dan Willemsen <dwillemsen@nvidia.com>

9 files changed, 79 insertions, 105 deletions

diff --git a/drivers/misc/tegra-profiler/Kconfig b/drivers/misc/tegra-profiler/Kconfig
index 312b75e9b848..d07cb3c3169b 100644
--- a/drivers/misc/tegra-profiler/Kconfig
+++ b/drivers/misc/tegra-profiler/Kconfig
@@ -1,6 +1,5 @@
 config TEGRA_PROFILER
 	bool "Enable Tegra profiler"
 	depends on ARCH_TEGRA
-	select KPROBES
 	help
 	  This option enables Tegra profiler
diff --git a/drivers/misc/tegra-profiler/hrt.c b/drivers/misc/tegra-profiler/hrt.c
index fdfbb96f5910..f0a8b277b5f7 100644
--- a/drivers/misc/tegra-profiler/hrt.c
+++ b/drivers/misc/tegra-profiler/hrt.c
@@ -374,121 +374,75 @@ static int remove_active_thread(struct quadd_cpu_context *cpu_ctx, pid_t pid)
 	return 0;
 }
 
-static int task_sched_in(struct kprobe *kp, struct pt_regs *regs)
+void __quadd_task_sched_in(struct task_struct *prev,
+			   struct task_struct *task)
 {
-	int n, prev_flag, current_flag;
-	struct task_struct *prev, *task;
-	int prev_nr_active, new_nr_active;
+	int current_flag;
 	struct quadd_cpu_context *cpu_ctx = this_cpu_ptr(hrt.cpu_ctx);
 	struct quadd_ctx *ctx = hrt.quadd_ctx;
 	struct event_data events[QUADD_MAX_COUNTERS];
 	/* static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 2); */
 
-	if (hrt.active == 0)
-		return 0;
-
-	prev = (struct task_struct *)regs->ARM_r1;
-	task = current;
+	if (likely(hrt.active == 0))
+		return;
 /*
 	if (__ratelimit(&ratelimit_state))
-		pr_info("cpu: %d, prev: %u (%u) \t--> curr: %u (%u)\n",
+		pr_info("sch_in, cpu: %d, prev: %u (%u) \t--> curr: %u (%u)\n",
 			quadd_get_processor_id(), (unsigned int)prev->pid,
 			(unsigned int)prev->tgid, (unsigned int)task->pid,
 			(unsigned int)task->tgid);
 */
-	if (!prev || !prev->real_parent || !prev->group_leader ||
-		prev->group_leader->tgid != prev->tgid) {
-		pr_err_once("Warning\n");
-		return 0;
-	}
-
-	prev_flag = is_profile_process(prev->tgid);
 	current_flag = is_profile_process(task->tgid);
 
-	if (prev_flag || current_flag) {
-		prev_nr_active = atomic_read(&cpu_ctx->nr_active);
-		qm_debug_task_sched_in(prev->pid, task->pid, prev_nr_active);
-
-		if (prev_flag) {
-			n = remove_active_thread(cpu_ctx, prev->pid);
-			atomic_sub(n, &cpu_ctx->nr_active);
-		}
-		if (current_flag) {
-			add_active_thread(cpu_ctx, task->pid, task->tgid);
-			atomic_inc(&cpu_ctx->nr_active);
-		}
-
-		new_nr_active = atomic_read(&cpu_ctx->nr_active);
-		if (prev_nr_active != new_nr_active) {
-			if (prev_nr_active == 0) {
-				if (ctx->pmu)
-					ctx->pmu->start();
+	if (current_flag) {
+		add_active_thread(cpu_ctx, task->pid, task->tgid);
+		atomic_inc(&cpu_ctx->nr_active);
 
-				if (ctx->pl310)
-					ctx->pl310->read(events);
+		if (atomic_read(&cpu_ctx->nr_active) == 1) {
+			if (ctx->pmu)
+				ctx->pmu->start();
 
-				start_hrtimer(cpu_ctx);
-				atomic_inc(&hrt.nr_active_all_core);
-			} else if (new_nr_active == 0) {
-				cancel_hrtimer(cpu_ctx);
-				atomic_dec(&hrt.nr_active_all_core);
+			if (ctx->pl310)
+				ctx->pl310->read(events);
 
-				if (ctx->pmu)
-					ctx->pmu->stop();
-			}
+			start_hrtimer(cpu_ctx);
+			atomic_inc(&hrt.nr_active_all_core);
 		}
 	}
-
-	return 0;
 }
 
-static int handler_fault(struct kprobe *kp, struct pt_regs *regs, int trapnr)
+void __quadd_task_sched_out(struct task_struct *prev,
+			    struct task_struct *next)
 {
-	pr_err_once("addr: %p, symbol: %s\n", kp->addr, kp->symbol_name);
-	return 0;
-}
-
-static int start_instr(void)
-{
-	int err;
-
-	memset(&hrt.kp_in, 0, sizeof(struct kprobe));
+	int n, prev_flag;
+	struct quadd_cpu_context *cpu_ctx = this_cpu_ptr(hrt.cpu_ctx);
+	struct quadd_ctx *ctx = hrt.quadd_ctx;
+	/* static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 2); */
 
-	hrt.kp_in.pre_handler = task_sched_in;
-	hrt.kp_in.fault_handler = handler_fault;
-	hrt.kp_in.addr = 0;
-	hrt.kp_in.symbol_name = QUADD_HRT_SCHED_IN_FUNC;
+	if (likely(hrt.active == 0))
+		return;
+/*
+	if (__ratelimit(&ratelimit_state))
+		pr_info("sch_out: cpu: %d, prev: %u (%u) \t--> next: %u (%u)\n",
+			quadd_get_processor_id(), (unsigned int)prev->pid,
+			(unsigned int)prev->tgid, (unsigned int)next->pid,
+			(unsigned int)next->tgid);
+*/
 
-	err = register_kprobe(&hrt.kp_in);
-	if (err) {
-		pr_err("register_kprobe error, symbol_name: %s\n",
-			hrt.kp_in.symbol_name);
-		return err;
-	}
-	return 0;
-}
+	prev_flag = is_profile_process(prev->tgid);
 
-static void stop_instr(void)
-{
-	unregister_kprobe(&hrt.kp_in);
-}
+	if (prev_flag) {
+		n = remove_active_thread(cpu_ctx, prev->pid);
+		atomic_sub(n, &cpu_ctx->nr_active);
 
-static int init_instr(void)
-{
-	int err;
+		if (atomic_read(&cpu_ctx->nr_active) == 0) {
+			cancel_hrtimer(cpu_ctx);
+			atomic_dec(&hrt.nr_active_all_core);
 
-	err = start_instr();
-	if (err) {
-		pr_err("Init instr failed\n");
-		return err;
+			if (ctx->pmu)
+				ctx->pmu->stop();
+		}
 	}
-	stop_instr();
-	return 0;
-}
-
-static int deinit_instr(void)
-{
-	return 0;
 }
 
 static void reset_cpu_ctx(void)
@@ -512,7 +466,6 @@ static void reset_cpu_ctx(void)
 
 int quadd_hrt_start(void)
 {
-	int err;
 	u64 period;
 	long freq;
 	struct quadd_ctx *ctx = hrt.quadd_ctx;
@@ -531,12 +484,6 @@ int quadd_hrt_start(void)
 
 	reset_cpu_ctx();
 
-	err = start_instr();
-	if (err) {
-		pr_err("error: start_instr is failed\n");
-		return err;
-	}
-
 	put_header();
 
 	if (ctx->pl310)
@@ -563,7 +510,6 @@ void quadd_hrt_stop(void)
 	quadd_ma_stop(&hrt);
 
 	hrt.active = 0;
-	stop_instr();
 
 	atomic64_set(&hrt.counter_samples, 0);
 
@@ -575,7 +521,6 @@ void quadd_hrt_deinit(void)
 	if (hrt.active)
 		quadd_hrt_stop();
 
-	deinit_instr();
 	free_percpu(hrt.cpu_ctx);
 }
 
@@ -622,8 +567,5 @@ struct quadd_hrt_ctx *quadd_hrt_init(struct quadd_ctx *ctx)
 		init_hrtimer(cpu_ctx);
 	}
 
-	if (init_instr())
-		return NULL;
-
 	return &hrt;
 }
diff --git a/drivers/misc/tegra-profiler/hrt.h b/drivers/misc/tegra-profiler/hrt.h
index f113b8846a02..7eff30a6ec85 100644
--- a/drivers/misc/tegra-profiler/hrt.h
+++ b/drivers/misc/tegra-profiler/hrt.h
@@ -23,7 +23,6 @@
 
 #include <linux/hrtimer.h>
 #include <linux/limits.h>
-#include <linux/kprobes.h>
 
 #include "backtrace.h"
 
@@ -53,9 +52,6 @@ struct quadd_hrt_ctx {
 	struct quadd_cpu_context * __percpu cpu_ctx;
 	u64 sample_period;
 
-	struct kprobe kp_in;
-	/* struct kinstr ki_out; */
-
 	struct quadd_ctx *quadd_ctx;
 
 	int active;
diff --git a/drivers/misc/tegra-profiler/ma.c b/drivers/misc/tegra-profiler/ma.c
index 2021954ace58..d561a2411c5a 100644
--- a/drivers/misc/tegra-profiler/ma.c
+++ b/drivers/misc/tegra-profiler/ma.c
@@ -19,6 +19,7 @@
 #include <linux/module.h>
 #include <linux/mm.h>
 #include <linux/timer.h>
+#include <linux/sched.h>
 
 #include <linux/tegra_profiler.h>
 
diff --git a/drivers/misc/tegra-profiler/main.c b/drivers/misc/tegra-profiler/main.c
index 823fa4db82cc..2b84dd036d81 100644
--- a/drivers/misc/tegra-profiler/main.c
+++ b/drivers/misc/tegra-profiler/main.c
@@ -19,6 +19,7 @@
 #include <linux/init.h>
 #include <linux/module.h>
 #include <linux/err.h>
+#include <linux/sched.h>
 
 #include <linux/tegra_profiler.h>
 
diff --git a/drivers/misc/tegra-profiler/mmap.c b/drivers/misc/tegra-profiler/mmap.c
index 370508b84f2b..2b73e5cd9c0a 100644
--- a/drivers/misc/tegra-profiler/mmap.c
+++ b/drivers/misc/tegra-profiler/mmap.c
@@ -22,6 +22,7 @@
 #include <linux/fs.h>
 #include <linux/slab.h>
 #include <linux/err.h>
+#include <linux/sched.h>
 
 #include <linux/tegra_profiler.h>
 
@@ -145,6 +146,7 @@ char *quadd_get_mmap(struct quadd_cpu_context *cpu_ctx,
 			break;
 		}
 	} else {
+#ifdef CONFIG_MODULES
 		struct module *mod;
 
 		preempt_disable();
@@ -159,6 +161,7 @@ char *quadd_get_mmap(struct quadd_cpu_context *cpu_ctx,
 				sample->pgoff = 0;
 			}
 		}
+#endif
 	}
 
 	if (file_name) {
diff --git a/drivers/misc/tegra-profiler/version.h b/drivers/misc/tegra-profiler/version.h
index 3c03b2978711..8d2eec0e2c2d 100644
--- a/drivers/misc/tegra-profiler/version.h
+++ b/drivers/misc/tegra-profiler/version.h
@@ -18,7 +18,7 @@
 #ifndef __QUADD_VERSION_H
 #define __QUADD_VERSION_H
 
-#define QUADD_MODULE_VERSION		"1.30"
+#define QUADD_MODULE_VERSION		"1.31"
 #define QUADD_MODULE_BRANCH		"Dev"
 
 #endif	/* __QUADD_VERSION_H */
diff --git a/include/linux/tegra_profiler.h b/include/linux/tegra_profiler.h
index e8f17ad75792..5840ba3a8824 100644
--- a/include/linux/tegra_profiler.h
+++ b/include/linux/tegra_profiler.h
@@ -305,4 +305,33 @@ struct quadd_module_version {
 
 #pragma pack(pop)
 
+#ifdef CONFIG_TEGRA_PROFILER
+extern void __quadd_task_sched_in(struct task_struct *prev,
+				  struct task_struct *task);
+extern void __quadd_task_sched_out(struct task_struct *prev,
+				   struct task_struct *next);
+
+static inline void quadd_task_sched_in(struct task_struct *prev,
+				       struct task_struct *task)
+{
+	__quadd_task_sched_in(prev, task);
+}
+
+static inline void quadd_task_sched_out(struct task_struct *prev,
+					struct task_struct *next)
+{
+	__quadd_task_sched_out(prev, next);
+}
+#else
+static inline void quadd_task_sched_in(struct task_struct *prev,
+				       struct task_struct *task)
+{
+}
+
+static inline void quadd_task_sched_out(struct task_struct *prev,
+					struct task_struct *next)
+{
+}
+#endif
+
 #endif  /* __TEGRA_PROFILER_H */
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 8fbbe9cfa1c9..872cf2e019df 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -72,6 +72,7 @@
 #include <linux/slab.h>
 #include <linux/init_task.h>
 #include <linux/binfmts.h>
+#include <linux/tegra_profiler.h>
 
 #include <asm/switch_to.h>
 #include <asm/tlb.h>
@@ -1914,6 +1915,7 @@ prepare_task_switch(struct rq *rq, struct task_struct *prev,
 {
 	sched_info_switch(prev, next);
 	perf_event_task_sched_out(prev, next);
+	quadd_task_sched_out(prev, next);
 	fire_sched_out_preempt_notifiers(prev, next);
 	prepare_lock_switch(rq, next);
 	prepare_arch_switch(next);
@@ -1960,6 +1962,7 @@ static void finish_task_switch(struct rq *rq, struct task_struct *prev)
 	local_irq_disable();
 #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */
 	perf_event_task_sched_in(prev, current);
+	quadd_task_sched_in(prev, current);
 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
 	local_irq_enable();
 #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */