[PATCH 0/2] blk-wbt: fix two wbt enable problems

[PATCH 0/2] blk-wbt: fix two wbt enable problems

[Zhang Yi]

Commit a79050434b45 ("blk-rq-qos: refactor out common elements of
blk-wbt") introduce two wbt enable problems. The first one will cause
false positive check by rwb_enabled() and lead to IO hung, the second
one is wbt_enable_default() could not re-enable wbt after switch
elevator back from bfq to other one.

Zhang Yi (2):
  blk-wbt: introduce a new disable state to prevent false positive by
    rwb_enabled()
  blk-wbt: make sure throttle is enabled properly

 block/blk-wbt.c | 11 ++++++++---
 block/blk-wbt.h |  1 +
 2 files changed, 9 insertions(+), 3 deletions(-)

-- 
2.31.1

[PATCH 1/2] blk-wbt: introduce a new disable state to prevent false positive by rwb_enabled()

[Zhang Yi]

Now that we disable wbt by simply zero out rwb->wb_normal in
wbt_disable_default() when switch elevator to bfq, but it's not safe
because it will become false positive if we change queue depth. If it
become false positive between wbt_wait() and wbt_track() when submit
write request, it will lead to drop rqw->inflight to -1 in wbt_done(),
which will end up trigger IO hung. Fix this issue by introduce a new
state which mean the wbt was disabled.

Fixes: a79050434b45 ("blk-rq-qos: refactor out common elements of blk-wbt")
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
---
 block/blk-wbt.c | 5 +++--
 block/blk-wbt.h | 1 +
 2 files changed, 4 insertions(+), 2 deletions(-)

diff --git a/block/blk-wbt.c b/block/blk-wbt.c
index 42aed0160f86..b098ac6a84f0 100644
--- a/block/blk-wbt.c
+++ b/block/blk-wbt.c
@@ -77,7 +77,8 @@ enum {
 
 static inline bool rwb_enabled(struct rq_wb *rwb)
 {
-	return rwb && rwb->wb_normal != 0;
+	return rwb && rwb->enable_state != WBT_STATE_OFF_DEFAULT &&
+		      rwb->wb_normal != 0;
 }
 
 static void wb_timestamp(struct rq_wb *rwb, unsigned long *var)
@@ -702,7 +703,7 @@ void wbt_disable_default(struct request_queue *q)
 	rwb = RQWB(rqos);
 	if (rwb->enable_state == WBT_STATE_ON_DEFAULT) {
 		blk_stat_deactivate(rwb->cb);
-		rwb->wb_normal = 0;
+		rwb->enable_state = WBT_STATE_OFF_DEFAULT;
 	}
 }
 EXPORT_SYMBOL_GPL(wbt_disable_default);
diff --git a/block/blk-wbt.h b/block/blk-wbt.h
index 16bdc85b8df9..2eb01becde8c 100644
--- a/block/blk-wbt.h
+++ b/block/blk-wbt.h
@@ -34,6 +34,7 @@ enum {
 enum {
 	WBT_STATE_ON_DEFAULT	= 1,
 	WBT_STATE_ON_MANUAL	= 2,
+	WBT_STATE_OFF_DEFAULT
 };
 
 struct rq_wb {
-- 
2.31.1

[PATCH 2/2] blk-wbt: make sure throttle is enabled properly

[Zhang Yi]

After commit a79050434b45 ("blk-rq-qos: refactor out common elements of
blk-wbt"), if throttle was disabled by wbt_disable_default(), we could
not enable again, fix this by set enable_state back to
WBT_STATE_ON_DEFAULT.

Fixes: a79050434b45 ("blk-rq-qos: refactor out common elements of blk-wbt")
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
---
 block/blk-wbt.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/block/blk-wbt.c b/block/blk-wbt.c
index b098ac6a84f0..f5e5ac915bf7 100644
--- a/block/blk-wbt.c
+++ b/block/blk-wbt.c
@@ -637,9 +637,13 @@ void wbt_set_write_cache(struct request_queue *q, bool write_cache_on)
 void wbt_enable_default(struct request_queue *q)
 {
 	struct rq_qos *rqos = wbt_rq_qos(q);
+
 	/* Throttling already enabled? */
-	if (rqos)
+	if (rqos) {
+		if (RQWB(rqos)->enable_state == WBT_STATE_OFF_DEFAULT)
+			RQWB(rqos)->enable_state = WBT_STATE_ON_DEFAULT;
 		return;
+	}
 
 	/* Queue not registered? Maybe shutting down... */
 	if (!blk_queue_registered(q))
-- 
2.31.1

Re: [PATCH 0/2] blk-wbt: fix two wbt enable problems

[Jens Axboe]

On 6/19/21 3:36 AM, Zhang Yi wrote:
> Commit a79050434b45 ("blk-rq-qos: refactor out common elements of
> blk-wbt") introduce two wbt enable problems. The first one will cause
> false positive check by rwb_enabled() and lead to IO hung, the second
> one is wbt_enable_default() could not re-enable wbt after switch
> elevator back from bfq to other one.

Applied, thanks.

-- 
Jens Axboe