On 11/8/19 11:38 PM, Song Liu wrote:
On Wed, Nov 6, 2019 at 2:04 AM Guoqing Jiang
<guoqing.jiang@xxxxxxxxxxxxxxx> wrote:
[...]
-static void mddev_destroy_serial_pool(struct mddev *mddev, struct md_rdev *rdev)
+static void mddev_destroy_serial_pool(struct mddev *mddev, struct md_rdev *rdev,
+ bool is_force)
I guess mddev_destroy_serial_pool() doesn't need an extra argument. We
can just to clear_bit, etc.. no?
Let me explain a little bit, the function are used in below cases:
1. write-behind mode is enabled and write-mostly device has multi queues,
the pool should be destroyed if the last device flagged with CollisionCheck
is removed from array, or the write-mostly flag will be cleared from the
last
device.
2. if we want to enable/disable serialization for normal raid1 in later
patch,
which means the pool could be created/destroyed whether bitmap is existed
or not, and all member device should be flagged with CollisionCheck.
The new argument is introduced to distinguish the two cases.
I understand the difference here. And I agree that we need a new argument
for mddev_create_serial_pool(). The question is, it is possible NOT to add the
extra argument to mddev_destory_serial_pool().
I'd prefer to add it, so both create pool and destroy pool have the same
argument.
Though I can check rdev is NULL or not to differentiate where the two
functions
are called, but it is a little weird and not intuitive.
Thanks,
Guoqing
