4 files changed, 64 insertions, 12 deletions
diff --git a/arch/powerpc/include/asm/hw_irq.h b/arch/powerpc/include/asm/hw_irq.h
index 6627a3e1b13..09e6d5951bb 100644
--- a/arch/powerpc/include/asm/hw_irq.h
+++ b/arch/powerpc/include/asm/hw_irq.h
@@ -124,6 +124,8 @@ static inline bool arch_irq_disabled_regs(struct pt_regs *regs)
 	return !regs->softe;
 }
 
+extern bool prep_irq_for_idle(void);
+
 #else /* CONFIG_PPC64 */
 
 #define SET_MSR_EE(x)	mtmsr(x)
diff --git a/arch/powerpc/kernel/irq.c b/arch/powerpc/kernel/irq.c
index 64eec59193f..d6946e10a25 100644
--- a/arch/powerpc/kernel/irq.c
+++ b/arch/powerpc/kernel/irq.c
@@ -286,6 +286,52 @@ void notrace restore_interrupts(void)
 		__hard_irq_enable();
 }
 
+/*
+ * This is a helper to use when about to go into idle low-power
+ * when the latter has the side effect of re-enabling interrupts
+ * (such as calling H_CEDE under pHyp).
+ *
+ * You call this function with interrupts soft-disabled (this is
+ * already the case when ppc_md.power_save is called). The function
+ * will return whether to enter power save or just return.
+ *
+ * In the former case, it will have notified lockdep of interrupts
+ * being re-enabled and generally sanitized the lazy irq state,
+ * and in the latter case it will leave with interrupts hard
+ * disabled and marked as such, so the local_irq_enable() call
+ * in cpu_idle() will properly re-enable everything.
+ */
+bool prep_irq_for_idle(void)
+{
+	/*
+	 * First we need to hard disable to ensure no interrupt
+	 * occurs before we effectively enter the low power state
+	 */
+	hard_irq_disable();
+
+	/*
+	 * If anything happened while we were soft-disabled,
+	 * we return now and do not enter the low power state.
+	 */
+	if (lazy_irq_pending())
+		return false;
+
+	/* Tell lockdep we are about to re-enable */
+	trace_hardirqs_on();
+
+	/*
+	 * Mark interrupts as soft-enabled and clear the
+	 * PACA_IRQ_HARD_DIS from the pending mask since we
+	 * are about to hard enable as well as a side effect
+	 * of entering the low power state.
+	 */
+	local_paca->irq_happened &= ~PACA_IRQ_HARD_DIS;
+	local_paca->soft_enabled = 1;
+
+	/* Tell the caller to enter the low power state */
+	return true;
+}
+
 #endif /* CONFIG_PPC64 */
 
 int arch_show_interrupts(struct seq_file *p, int prec)
diff --git a/arch/powerpc/platforms/cell/pervasive.c b/arch/powerpc/platforms/cell/pervasive.c
index efdacc82957..d17e98bc0c1 100644
--- a/arch/powerpc/platforms/cell/pervasive.c
+++ b/arch/powerpc/platforms/cell/pervasive.c
@@ -42,11 +42,9 @@ static void cbe_power_save(void)
 {
 	unsigned long ctrl, thread_switch_control;
 
-	/*
-	 * We need to hard disable interrupts, the local_irq_enable() done by
-	 * our caller upon return will hard re-enable.
-	 */
-	hard_irq_disable();
+	/* Ensure our interrupt state is properly tracked */
+	if (!prep_irq_for_idle())
+		return;
 
 	ctrl = mfspr(SPRN_CTRLF);
 
@@ -81,6 +79,9 @@ static void cbe_power_save(void)
 	 */
 	ctrl &= ~(CTRL_RUNLATCH | CTRL_TE);
 	mtspr(SPRN_CTRLT, ctrl);
+
+	/* Re-enable interrupts in MSR */
+	__hard_irq_enable();
 }
 
 static int cbe_system_reset_exception(struct pt_regs *regs)
diff --git a/arch/powerpc/platforms/pseries/processor_idle.c b/arch/powerpc/platforms/pseries/processor_idle.c
index e61483e8e96..c71be66bd5d 100644
--- a/arch/powerpc/platforms/pseries/processor_idle.c
+++ b/arch/powerpc/platforms/pseries/processor_idle.c
@@ -99,15 +99,18 @@ out:
 static void check_and_cede_processor(void)
 {
 	/*
-	 * Interrupts are soft-disabled at this point,
-	 * but not hard disabled. So an interrupt might have
-	 * occurred before entering NAP, and would be potentially
-	 * lost (edge events, decrementer events, etc...) unless
-	 * we first hard disable then check.
+	 * Ensure our interrupt state is properly tracked,
+	 * also checks if no interrupt has occurred while we
+	 * were soft-disabled
 	 */
-	hard_irq_disable();
-	if (!lazy_irq_pending())
+	if (prep_irq_for_idle()) {
 		cede_processor();
+#ifdef CONFIG_TRACE_IRQFLAGS
+		/* Ensure that H_CEDE returns with IRQs on */
+		if (WARN_ON(!(mfmsr() & MSR_EE)))
+			__hard_irq_enable();
+#endif
+	}
 }
 
 static int dedicated_cede_loop(struct cpuidle_device *dev,