On 2018/06/19 20:10, Tetsuo Handa wrote:
> On 2018/06/16 4:40, Tetsuo Handa wrote:
>> Hmm, there might be other locations calling percpu_rwsem_release() ?
>
> There are other locations calling percpu_rwsem_release(), but quite few.
>
> include/linux/fs.h:1494:#define __sb_writers_release(sb, lev) \
> include/linux/fs.h-1495-
> percpu_rwsem_release(&(sb)->s_writers.rw_sem[(lev)-1], 1, _THIS_IP_)
>
> fs/btrfs/transaction.c:1821: __sb_writers_release(fs_info->sb,
> SB_FREEZE_FS);
> fs/aio.c:1566: __sb_writers_release(file_inode(file)->i_sb,
> SB_FREEZE_WRITE);
> fs/xfs/xfs_aops.c:211: __sb_writers_release(ioend->io_inode->i_sb,
> SB_FREEZE_FS);
>
>
>
> I'd like to check what atomic_long_read(&sem->rw_sem.count) says
> when hung task is reported.
>
syzbot reproduced this problem with the patch applied.
percpu_rw_semaphore(00000000082ac9da)
->rw_sem.count=0xfffffffe00000001
->rss.gp_state=2
->rss.gp_count=1
->rss.cb_state=0
->rss.gp_type=1
->readers_block=1
->read_count=0
->list_empty(rw_sem.wait_list)=0
->writer.task= (null)
The output says that percpu_down_read() was blocked because
somebody has called percpu_down_write().
DEFINE_STATIC_PERCPU_RWSEM(sem);
percpu_down_write(&sem);
percpu_down_read(&sem);
percpu_up_read(&sem);
percpu_up_write(&sem);
The next step is to find who is calling percpu_down_write(). How do we want to
do this?
We don't want to annoy normal linux-next.git testers. Below one?
---
include/linux/percpu-rwsem.h | 4 ++++
lib/Kconfig.debug | 6 ++++++
2 files changed, 10 insertions(+)
diff --git a/include/linux/percpu-rwsem.h b/include/linux/percpu-rwsem.h
index 79b99d6..26e87c3 100644
--- a/include/linux/percpu-rwsem.h
+++ b/include/linux/percpu-rwsem.h
@@ -130,7 +130,9 @@ extern int __percpu_init_rwsem(struct percpu_rw_semaphore *,
static inline void percpu_rwsem_release(struct percpu_rw_semaphore *sem,
bool read, unsigned long ip)
{
+#ifndef CONFIG_DEBUG_AID_FOR_SYZBOT
lock_release(&sem->rw_sem.dep_map, 1, ip);
+#endif
#ifdef CONFIG_RWSEM_SPIN_ON_OWNER
if (!read)
sem->rw_sem.owner = RWSEM_OWNER_UNKNOWN;
@@ -140,7 +142,9 @@ static inline void percpu_rwsem_release(struct
percpu_rw_semaphore *sem,
static inline void percpu_rwsem_acquire(struct percpu_rw_semaphore *sem,
bool read, unsigned long ip)
{
+#ifndef CONFIG_DEBUG_AID_FOR_SYZBOT
lock_acquire(&sem->rw_sem.dep_map, 0, 1, read, 1, NULL, ip);
+#endif
#ifdef CONFIG_RWSEM_SPIN_ON_OWNER
if (!read)
sem->rw_sem.owner = current;
diff --git a/lib/Kconfig.debug b/lib/Kconfig.debug
index c731ff9..f0d02e8 100644
--- a/lib/Kconfig.debug
+++ b/lib/Kconfig.debug
@@ -1181,6 +1181,12 @@ config DEBUG_LOCK_ALLOC
spin_lock_init()/mutex_init()/etc., or whether there is any lock
held during task exit.
+config DEBUG_AID_FOR_SYZBOT
+ bool "Additional debug options for syzbot"
+ default n
+ help
+ This option is intended for testing by syzbot.
+
config LOCKDEP
bool
depends on DEBUG_KERNEL && LOCK_DEBUGGING_SUPPORT
--
Hmm, given that neither xfs nor btrfs is used, is it aio code?
