diff -urp linux-4.7.1.org/arch/x86/include/asm/irqflags.h linux-4.7.1.obj/arch/x86/include/asm/irqflags.h
--- linux-4.7.1.org/arch/x86/include/asm/irqflags.h	2016-08-16 09:35:15.000000000 +0200
+++ linux-4.7.1.obj/arch/x86/include/asm/irqflags.h	2016-08-18 10:01:09.514219644 +0200
@@ -44,6 +44,16 @@ static inline void native_irq_enable(voi
 	asm volatile("sti": : :"memory");
 }
 
+/*
+ * This produces a "test; jz; sti" insn sequence.
+ * It's faster than "push reg; popf". If sti is skipped, it's much faster.
+ */
+static inline void native_cond_irq_enable(unsigned long flags)
+{
+	if (flags & X86_EFLAGS_IF)
+		native_irq_enable();
+}
+
 static inline void native_safe_halt(void)
 {
 	asm volatile("sti; hlt": : :"memory");
@@ -69,7 +79,8 @@ static inline notrace unsigned long arch
 
 static inline notrace void arch_local_irq_restore(unsigned long flags)
 {
-	native_restore_fl(flags);
+	if (flags & X86_EFLAGS_IF)
+		native_irq_enable();
 }
 
 static inline notrace void arch_local_irq_disable(void)
diff -urp linux-4.7.1.org/arch/x86/kernel/paravirt.c linux-4.7.1.obj/arch/x86/kernel/paravirt.c
--- linux-4.7.1.org/arch/x86/kernel/paravirt.c	2016-08-16 09:35:15.000000000 +0200
+++ linux-4.7.1.obj/arch/x86/kernel/paravirt.c	2016-08-18 10:01:24.797155109 +0200
@@ -313,7 +313,7 @@ struct pv_time_ops pv_time_ops = {
 
 __visible struct pv_irq_ops pv_irq_ops = {
 	.save_fl = __PV_IS_CALLEE_SAVE(native_save_fl),
-	.restore_fl = __PV_IS_CALLEE_SAVE(native_restore_fl),
+	.restore_fl = __PV_IS_CALLEE_SAVE(native_cond_irq_enable),
 	.irq_disable = __PV_IS_CALLEE_SAVE(native_irq_disable),
 	.irq_enable = __PV_IS_CALLEE_SAVE(native_irq_enable),
 	.safe_halt = native_safe_halt,
diff -urp linux-4.7.1.org/arch/x86/kernel/paravirt_patch_32.c linux-4.7.1.obj/arch/x86/kernel/paravirt_patch_32.c
--- linux-4.7.1.org/arch/x86/kernel/paravirt_patch_32.c	2016-08-16 09:35:15.000000000 +0200
+++ linux-4.7.1.obj/arch/x86/kernel/paravirt_patch_32.c	2016-08-18 04:43:19.388102755 +0200
@@ -2,7 +2,7 @@
 
 DEF_NATIVE(pv_irq_ops, irq_disable, "cli");
 DEF_NATIVE(pv_irq_ops, irq_enable, "sti");
-DEF_NATIVE(pv_irq_ops, restore_fl, "push %eax; popf");
+DEF_NATIVE(pv_irq_ops, restore_fl, "testb $((1<<9)>>8),%ah; jz 1f; sti; 1: ;");
 DEF_NATIVE(pv_irq_ops, save_fl, "pushf; pop %eax");
 DEF_NATIVE(pv_cpu_ops, iret, "iret");
 DEF_NATIVE(pv_mmu_ops, read_cr2, "mov %cr2, %eax");
diff -urp linux-4.7.1.org/arch/x86/kernel/paravirt_patch_64.c linux-4.7.1.obj/arch/x86/kernel/paravirt_patch_64.c
--- linux-4.7.1.org/arch/x86/kernel/paravirt_patch_64.c	2016-08-16 09:35:15.000000000 +0200
+++ linux-4.7.1.obj/arch/x86/kernel/paravirt_patch_64.c	2016-08-18 04:42:56.364545509 +0200
@@ -4,7 +4,7 @@
 
 DEF_NATIVE(pv_irq_ops, irq_disable, "cli");
 DEF_NATIVE(pv_irq_ops, irq_enable, "sti");
-DEF_NATIVE(pv_irq_ops, restore_fl, "pushq %rdi; popfq");
+DEF_NATIVE(pv_irq_ops, restore_fl, "testw $(1<<9),%di; jz 1f; sti; 1: ;");
 DEF_NATIVE(pv_irq_ops, save_fl, "pushfq; popq %rax");
 DEF_NATIVE(pv_mmu_ops, read_cr2, "movq %cr2, %rax");
 DEF_NATIVE(pv_mmu_ops, read_cr3, "movq %cr3, %rax");