Subject: x86: Support for lazy preemption
From: Thomas Gleixner <tglx@linutronix.de>
Date: Thu Nov  1 11:03:47 2012 +0100

From: Thomas Gleixner <tglx@linutronix.de>

Implement the x86 pieces for lazy preempt.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>


---
 arch/x86/Kconfig                   |    1 +
 arch/x86/include/asm/preempt.h     |   33 ++++++++++++++++++++++++++++++++-
 arch/x86/include/asm/thread_info.h |    7 +++++++
 include/linux/entry-common.h       |    2 +-
 kernel/entry/common.c              |    2 +-
 5 files changed, 42 insertions(+), 3 deletions(-)
---
@ arch/x86/Kconfig:253 @ config X86
 	select HAVE_PCI
 	select HAVE_PERF_REGS
 	select HAVE_PERF_USER_STACK_DUMP
+	select HAVE_PREEMPT_LAZY
 	select MMU_GATHER_RCU_TABLE_FREE	if PARAVIRT
 	select MMU_GATHER_MERGE_VMAS
 	select HAVE_POSIX_CPU_TIMERS_TASK_WORK
--- a/arch/x86/include/asm/preempt.h
+++ b/arch/x86/include/asm/preempt.h
@ arch/x86/Kconfig:93 @ static __always_inline void __preempt_co
  * a decrement which hits zero means we have no preempt_count and should
  * reschedule.
  */
-static __always_inline bool __preempt_count_dec_and_test(void)
+static __always_inline bool ____preempt_count_dec_and_test(void)
 {
 	return GEN_UNARY_RMWcc("decl", __preempt_count, e, __percpu_arg([var]));
 }
 
+static __always_inline bool __preempt_count_dec_and_test(void)
+{
+	if (____preempt_count_dec_and_test())
+		return true;
+#ifdef CONFIG_PREEMPT_LAZY
+	if (preempt_count())
+		return false;
+	if (current_thread_info()->preempt_lazy_count)
+		return false;
+	return test_thread_flag(TIF_NEED_RESCHED_LAZY);
+#else
+	return false;
+#endif
+}
+
 /*
  * Returns true when we need to resched and can (barring IRQ state).
  */
 static __always_inline bool should_resched(int preempt_offset)
 {
+#ifdef CONFIG_PREEMPT_LAZY
+	u32 tmp;
+	tmp = raw_cpu_read_4(__preempt_count);
+	if (tmp == preempt_offset)
+		return true;
+
+	/* preempt count == 0 ? */
+	tmp &= ~PREEMPT_NEED_RESCHED;
+	if (tmp != preempt_offset)
+		return false;
+	/* XXX PREEMPT_LOCK_OFFSET */
+	if (current_thread_info()->preempt_lazy_count)
+		return false;
+	return test_thread_flag(TIF_NEED_RESCHED_LAZY);
+#else
 	return unlikely(raw_cpu_read_4(__preempt_count) == preempt_offset);
+#endif
 }
 
 #ifdef CONFIG_PREEMPTION
--- a/arch/x86/include/asm/thread_info.h
+++ b/arch/x86/include/asm/thread_info.h
@ arch/x86/Kconfig:60 @ struct thread_info {
 	unsigned long		flags;		/* low level flags */
 	unsigned long		syscall_work;	/* SYSCALL_WORK_ flags */
 	u32			status;		/* thread synchronous flags */
+	int			preempt_lazy_count;	/* 0 => lazy preemptable
+							  <0 => BUG */
 #ifdef CONFIG_SMP
 	u32			cpu;		/* current CPU */
 #endif
@ arch/x86/Kconfig:70 @ struct thread_info {
 #define INIT_THREAD_INFO(tsk)			\
 {						\
 	.flags		= 0,			\
+	.preempt_lazy_count	= 0,		\
 }
 
 #else /* !__ASSEMBLY__ */
@ arch/x86/Kconfig:98 @ struct thread_info {
 #define TIF_NOCPUID		15	/* CPUID is not accessible in userland */
 #define TIF_NOTSC		16	/* TSC is not accessible in userland */
 #define TIF_NOTIFY_SIGNAL	17	/* signal notifications exist */
+#define TIF_NEED_RESCHED_LAZY	19	/* lazy rescheduling necessary */
 #define TIF_MEMDIE		20	/* is terminating due to OOM killer */
 #define TIF_POLLING_NRFLAG	21	/* idle is polling for TIF_NEED_RESCHED */
 #define TIF_IO_BITMAP		22	/* uses I/O bitmap */
@ arch/x86/Kconfig:122 @ struct thread_info {
 #define _TIF_NOCPUID		(1 << TIF_NOCPUID)
 #define _TIF_NOTSC		(1 << TIF_NOTSC)
 #define _TIF_NOTIFY_SIGNAL	(1 << TIF_NOTIFY_SIGNAL)
+#define _TIF_NEED_RESCHED_LAZY	(1 << TIF_NEED_RESCHED_LAZY)
 #define _TIF_POLLING_NRFLAG	(1 << TIF_POLLING_NRFLAG)
 #define _TIF_IO_BITMAP		(1 << TIF_IO_BITMAP)
 #define _TIF_SPEC_FORCE_UPDATE	(1 << TIF_SPEC_FORCE_UPDATE)
@ arch/x86/Kconfig:154 @ struct thread_info {
 
 #define _TIF_WORK_CTXSW_NEXT	(_TIF_WORK_CTXSW)
 
+#define _TIF_NEED_RESCHED_MASK	(_TIF_NEED_RESCHED | _TIF_NEED_RESCHED_LAZY)
+
 #define STACK_WARN		(THREAD_SIZE/8)
 
 /*
--- a/include/linux/entry-common.h
+++ b/include/linux/entry-common.h
@ arch/x86/Kconfig:62 @
 
 #define EXIT_TO_USER_MODE_WORK						\
 	(_TIF_SIGPENDING | _TIF_NOTIFY_RESUME | _TIF_UPROBE |		\
-	 _TIF_NEED_RESCHED | _TIF_PATCH_PENDING | _TIF_NOTIFY_SIGNAL |	\
+	 _TIF_NEED_RESCHED_MASK | _TIF_PATCH_PENDING | _TIF_NOTIFY_SIGNAL |	\
 	 ARCH_EXIT_TO_USER_MODE_WORK)
 
 /**
--- a/kernel/entry/common.c
+++ b/kernel/entry/common.c
@ arch/x86/Kconfig:158 @ static unsigned long exit_to_user_mode_l
 
 		local_irq_enable_exit_to_user(ti_work);
 
-		if (ti_work & _TIF_NEED_RESCHED)
+		if (ti_work & _TIF_NEED_RESCHED_MASK)
 			schedule();
 
 		if (ti_work & _TIF_UPROBE)