[RFC 9/9] tty: implement BTM as mutex instead of BKL

[Arnd Bergmann <arnd@arndb.de>]

The TTY layer now has its own ways to deal with recursive
locking and release-on-sleep for the tty_lock() calls,
meaning that it's safe to replace the Big Kernel Lock
with a subsystem specific Big TTY Mutex (BTM).

This patch for now makes the new behaviour an optional
experimental feature that can be enabled for testing
purposes.

Using a regular mutex here will change the behaviour
when blocked on the BTM from spinning to sleeping,
but that should be visible to the user.

Using the mutex also means that all the BTM is now
covered by lockdep.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
---
 drivers/char/Makefile     |    1 +
 include/linux/init_task.h |    1 +
 include/linux/sched.h     |    1 +
 include/linux/tty.h       |   19 +++++++++++++++++++
 kernel/fork.c             |    1 +
 lib/Kconfig.debug         |   10 ++++++++++
 6 files changed, 33 insertions(+), 0 deletions(-)

diff --git a/drivers/char/Makefile b/drivers/char/Makefile
index f957edf..74ee3fa 100644
--- a/drivers/char/Makefile
+++ b/drivers/char/Makefile
@@ -9,6 +9,7 @@ FONTMAPFILE = cp437.uni
 
 obj-y	 += mem.o random.o tty_io.o n_tty.o tty_ioctl.o tty_ldisc.o tty_buffer.o tty_port.o
 
+obj-$(CONFIG_TTY_MUTEX)		+= tty_mutex.o
 obj-$(CONFIG_LEGACY_PTYS)	+= pty.o
 obj-$(CONFIG_UNIX98_PTYS)	+= pty.o
 obj-y				+= misc.o
diff --git a/include/linux/init_task.h b/include/linux/init_task.h
index b1ed1cd..3c0b4ab 100644
--- a/include/linux/init_task.h
+++ b/include/linux/init_task.h
@@ -114,6 +114,7 @@ extern struct cred init_cred;
 	.usage		= ATOMIC_INIT(2),				\
 	.flags		= PF_KTHREAD,					\
 	.lock_depth	= -1,						\
+	.tty_lock_depth	= -1,						\
 	.prio		= MAX_PRIO-20,					\
 	.static_prio	= MAX_PRIO-20,					\
 	.normal_prio	= MAX_PRIO-20,					\
diff --git a/include/linux/sched.h b/include/linux/sched.h
index dad7f66..5f03259 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -1175,6 +1175,7 @@ struct task_struct {
 	unsigned int ptrace;
 
 	int lock_depth;		/* BKL lock depth */
+	int tty_lock_depth;	/* TTY lock depth */
 
 #ifdef CONFIG_SMP
 #ifdef __ARCH_WANT_UNLOCKED_CTXSW
diff --git a/include/linux/tty.h b/include/linux/tty.h
index 60b3d69..1659ba8 100644
--- a/include/linux/tty.h
+++ b/include/linux/tty.h
@@ -572,6 +572,7 @@ extern int vt_ioctl(struct tty_struct *tty, struct file *file,
 extern long vt_compat_ioctl(struct tty_struct *tty, struct file * file,
 		     unsigned int cmd, unsigned long arg);
 
+/* tty_mutex.c */
 /* functions for preparation of BKL removal */
 
 /*
@@ -584,6 +585,22 @@ extern long vt_compat_ioctl(struct tty_struct *tty, struct file * file,
  * be shown to never get called with this held already, it should
  * use tty_lock() instead.
  */ 
+#ifdef CONFIG_TTY_MUTEX
+extern void __lockfunc tty_lock_nested(void) __acquires(tty_lock);
+extern void __lockfunc tty_lock(void) __acquires(tty_lock);
+extern void __lockfunc tty_unlock(void) __releases(tty_lock);
+#define tty_locked()		(current->tty_lock_depth >= 0)
+int __lockfunc __reacquire_tty_lock(void);
+void __lockfunc __release_tty_lock(void);
+#define release_tty_lock(tsk) do {           \
+        if (unlikely((tsk)->tty_lock_depth >= 0))   \
+                __release_tty_lock();        \
+} while (0)
+#define reacquire_tty_lock(tsk) \
+        ((tsk->tty_lock_depth >= 0) ? \
+                __reacquire_tty_lock() : 0 )
+
+#else
 static inline void __lockfunc tty_lock_nested(void) __acquires(kernel_lock)
 {
 	lock_kernel();
@@ -609,6 +626,8 @@ static inline void __release_tty_lock(void)
 #define release_tty_lock(tsk) do { } while (0)
 #define reacquire_tty_lock(tsk) do { } while (0)
 
+#endif
+
 /*
  * mutex_lock_tty - lock a mutex without holding the BTM
  *
diff --git a/kernel/fork.c b/kernel/fork.c
index 4799c5f..b66376a 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -1061,6 +1061,7 @@ static struct task_struct *copy_process(unsigned long clone_flags,
 	posix_cpu_timers_init(p);
 
 	p->lock_depth = -1;		/* -1 = no lock */
+	p->tty_lock_depth = -1;		/* -1 = no lock */
 	do_posix_clock_monotonic_gettime(&p->start_time);
 	p->real_start_time = p->start_time;
 	monotonic_to_bootbased(&p->real_start_time);
diff --git a/lib/Kconfig.debug b/lib/Kconfig.debug
index 1fafb4b..32ff32c 100644
--- a/lib/Kconfig.debug
+++ b/lib/Kconfig.debug
@@ -428,6 +428,16 @@ config RT_MUTEX_TESTER
 	help
 	  This option enables a rt-mutex tester.
 
+config TTY_MUTEX
+	bool "Use a mutex instead of BKL for TTY locking"
+	depends on EXPERIMENTAL && SMP
+	help
+	  The TTY subsystem traditionally depends on the big kernel lock
+	  for serialization. Saying Y here replaces the BKL with the Big
+	  TTY Mutex (BTM).
+	  Building a kernel without the BKL is only possible with TTY_MUTEX
+	  enabled.
+
 config DEBUG_SPINLOCK
 	bool "Spinlock and rw-lock debugging: basic checks"
 	depends on DEBUG_KERNEL
-- 
1.7.0