On 9/19/22 9:08 AM, Yu Kuai wrote:

Hi,

在 2022/09/18 19:36, Guoqing Jiang 写道:

quoted

On 9/16/22 7:34 PM, Yu Kuai wrote:

quoted

From: Yu Kuai <redacted>

Currently, wait_barrier() will hold 'resync_lock' to read 
'conf->barrier',
and io can't be dispatched until 'barrier' is dropped.

Since holding the 'barrier' is not common, convert 'resync_lock' to use
seqlock so that holding lock can be avoided in fast path.

Signed-off-by: Yu Kuai <redacted>
---
  drivers/md/raid10.c | 87 
++++++++++++++++++++++++++++++---------------
  drivers/md/raid10.h |  2 +-
  2 files changed, 59 insertions(+), 30 deletions(-)

diff --git a/drivers/md/raid10.c b/drivers/md/raid10.c
index 9a28abd19709..2daa7d57034c 100644
--- a/drivers/md/raid10.c
+++ b/drivers/md/raid10.c

@@ -79,6 +79,21 @@ static void end_reshape(struct r10conf *conf);

  #include "raid1-10.c"
+#define NULL_CMD
+#define cmd_before(conf, cmd) \
+    do { \
+        write_sequnlock_irq(&(conf)->resync_lock); \
+        cmd; \
+    } while (0)
+#define cmd_after(conf) write_seqlock_irq(&(conf)->resync_lock)

The two is not paired well given only cmd_before needs the 'cmd'.

May be should I just remove cmd_after?

I'd remove it but just my personal flavor.

quoted

quoted

+
+#define wait_event_barrier_cmd(conf, cond, cmd) \
+    wait_event_cmd((conf)->wait_barrier, cond, cmd_before(conf, 
cmd), \
+               cmd_after(conf))
+
+#define wait_event_barrier(conf, cond) \
+    wait_event_barrier_cmd(conf, cond, NULL_CMD)

What is the issue without define NULL_CMD?

Checkpatch will complain this:

ERROR: space prohibited before that close parenthesis ')'
#38: FILE: drivers/md/raid10.c:94:
+       wait_event_barrier_cmd(conf, cond, )

total: 1 errors, 0 warnings, 169 lines checked

Hmm, freeze_array has a different usage for it, so two cmds before sleep
and one cmd after sleep, perhaps it is the best way for now.

Thanks,
Guoqing