perf/x86/lbr: Fix shorter LBRs call stacks for the system-wide mode

In the system-wide mode, LBR callstacks are shorter in comparison to
the per-process mode.

LBR MSRs are reset during a context switch in the system-wide mode. For
the LBR call stack, the LBRs should be always saved/restored during a
context switch.

Use the space in task_struct to save/restore the LBR call stack data.

For a system-wide event, it's unnecessagy to update the
lbr_callstack_users for each threads. Add a variable in x86_pmu to
indicate whether the system-wide event is active.

Fixes: 76cb2c6 ("perf/x86/intel: Save/restore LBR stack during context switch")
Reported-by: Andi Kleen <ak@linux.intel.com>
Reported-by: Alexey Budankov <alexey.budankov@linux.intel.com>
Debugged-by: Alexey Budankov <alexey.budankov@linux.intel.com>
Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/r/20250314172700.438923-5-kan.liang@linux.intel.com

Author: Kan Liang

arch/x86/events/intel/lbr.c

@@ -422,11 +422,17 @@ static __always_inline bool lbr_is_reset_in_cstate(void *ctx)
 	return !rdlbr_from(((struct x86_perf_task_context *)ctx)->tos, NULL);
 }
 
+static inline bool has_lbr_callstack_users(void *ctx)
+{
+	return task_context_opt(ctx)->lbr_callstack_users ||
+	       x86_pmu.lbr_callstack_users;
+}
+
 static void __intel_pmu_lbr_restore(void *ctx)
 {
 	struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
 
-	if (task_context_opt(ctx)->lbr_callstack_users == 0 ||
+	if (!has_lbr_callstack_users(ctx) ||
 	    task_context_opt(ctx)->lbr_stack_state == LBR_NONE) {
 		intel_pmu_lbr_reset();
 		return;
@@ -503,7 +509,7 @@ static void __intel_pmu_lbr_save(void *ctx)
 {
 	struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
 
-	if (task_context_opt(ctx)->lbr_callstack_users == 0) {
+	if (!has_lbr_callstack_users(ctx)) {
 		task_context_opt(ctx)->lbr_stack_state = LBR_NONE;
 		return;
 	}
@@ -543,6 +549,7 @@ void intel_pmu_lbr_sched_task(struct perf_event_pmu_context *pmu_ctx,
 			      struct task_struct *task, bool sched_in)
 {
 	struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
+	struct perf_ctx_data *ctx_data;
 	void *task_ctx;
 
 	if (!cpuc->lbr_users)
@@ -553,14 +560,18 @@ void intel_pmu_lbr_sched_task(struct perf_event_pmu_context *pmu_ctx,
 	 * the task was scheduled out, restore the stack. Otherwise flush
 	 * the LBR stack.
 	 */
-	task_ctx = pmu_ctx ? pmu_ctx->task_ctx_data : NULL;
+	rcu_read_lock();
+	ctx_data = rcu_dereference(task->perf_ctx_data);
+	task_ctx = ctx_data ? ctx_data->data : NULL;
 	if (task_ctx) {
 		if (sched_in)
 			__intel_pmu_lbr_restore(task_ctx);
 		else
 			__intel_pmu_lbr_save(task_ctx);
+		rcu_read_unlock();
 		return;
 	}
+	rcu_read_unlock();
 
 	/*
 	 * Since a context switch can flip the address space and LBR entries
@@ -589,9 +600,19 @@ void intel_pmu_lbr_add(struct perf_event *event)
 
 	cpuc->br_sel = event->hw.branch_reg.reg;
 
-	if (branch_user_callstack(cpuc->br_sel) && event->pmu_ctx->task_ctx_data)
-		task_context_opt(event->pmu_ctx->task_ctx_data)->lbr_callstack_users++;
+	if (branch_user_callstack(cpuc->br_sel)) {
+		if (event->attach_state & PERF_ATTACH_TASK) {
+			struct task_struct *task = event->hw.target;
+			struct perf_ctx_data *ctx_data;
 
+			rcu_read_lock();
+			ctx_data = rcu_dereference(task->perf_ctx_data);
+			if (ctx_data)
+				task_context_opt(ctx_data->data)->lbr_callstack_users++;
+			rcu_read_unlock();
+		} else
+			x86_pmu.lbr_callstack_users++;
+	}
 	/*
 	 * Request pmu::sched_task() callback, which will fire inside the
 	 * regular perf event scheduling, so that call will:
@@ -665,9 +686,19 @@ void intel_pmu_lbr_del(struct perf_event *event)
 	if (!x86_pmu.lbr_nr)
 		return;
 
-	if (branch_user_callstack(cpuc->br_sel) &&
-	    event->pmu_ctx->task_ctx_data)
-		task_context_opt(event->pmu_ctx->task_ctx_data)->lbr_callstack_users--;
+	if (branch_user_callstack(cpuc->br_sel)) {
+		if (event->attach_state & PERF_ATTACH_TASK) {
+			struct task_struct *task = event->hw.target;
+			struct perf_ctx_data *ctx_data;
+
+			rcu_read_lock();
+			ctx_data = rcu_dereference(task->perf_ctx_data);
+			if (ctx_data)
+				task_context_opt(ctx_data->data)->lbr_callstack_users--;
+			rcu_read_unlock();
+		} else
+			x86_pmu.lbr_callstack_users--;
+	}
 
 	if (event->hw.flags & PERF_X86_EVENT_LBR_SELECT)
 		cpuc->lbr_select = 0;

arch/x86/events/perf_event.h

@@ -920,6 +920,7 @@ struct x86_pmu {
 		const int	*lbr_sel_map;	   /* lbr_select mappings */
 		int		*lbr_ctl_map;	   /* LBR_CTL mappings */
 	};
+	u64		lbr_callstack_users;	   /* lbr callstack system wide users */
 	bool		lbr_double_abort;	   /* duplicated lbr aborts */
 	bool		lbr_pt_coexist;		   /* (LBR|BTS) may coexist with PT */