From: Chunguang Xu <[email protected]>
We found a issue on production environment while using NVMe
over RDMA, admin_q reconnect failed forever while remote
target and network is ok. After dig into it, we found it
may caused by a ABBA deadlock due to tag allocation. In my
case, the tag was hold by a keep alive request waiting
inside admin_q, as we quiesced admin_q while reset ctrl,
so the request maked as idle and will not process before
reset success. As fabric_q shares tagset with admin_q,
while reconnect remote target, we need a tag for connect
command, but the only one reserved tag was held by keep
alive command which waiting inside admin_q. As a result,
we failed to reconnect admin_q forever. In order to fix
this issue, I think we should keep two reserved tags for
admin queue.
Fixes: ed01fee283a0 ("nvme-fabrics: only reserve a single tag")
Signed-off-by: Chunguang Xu <[email protected]>
---
drivers/nvme/host/core.c | 4 ++--
drivers/nvme/host/fabrics.h | 10 ++++------
2 files changed, 6 insertions(+), 8 deletions(-)
diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
index 0a96362912ce..3d394a075d20 100644
--- a/drivers/nvme/host/core.c
+++ b/drivers/nvme/host/core.c
@@ -4359,7 +4359,7 @@ int nvme_alloc_admin_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
set->ops = ops;
set->queue_depth = NVME_AQ_MQ_TAG_DEPTH;
if (ctrl->ops->flags & NVME_F_FABRICS)
- set->reserved_tags = NVMF_RESERVED_TAGS;
+ set->reserved_tags = NVMF_ADMIN_RESERVED_TAGS;
set->numa_node = ctrl->numa_node;
set->flags = BLK_MQ_F_NO_SCHED;
if (ctrl->ops->flags & NVME_F_BLOCKING)
@@ -4428,7 +4428,7 @@ int nvme_alloc_io_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
if (ctrl->quirks & NVME_QUIRK_SHARED_TAGS)
set->reserved_tags = NVME_AQ_DEPTH;
else if (ctrl->ops->flags & NVME_F_FABRICS)
- set->reserved_tags = NVMF_RESERVED_TAGS;
+ set->reserved_tags = NVMF_IO_RESERVED_TAGS;
set->numa_node = ctrl->numa_node;
set->flags = BLK_MQ_F_SHOULD_MERGE;
if (ctrl->ops->flags & NVME_F_BLOCKING)
diff --git a/drivers/nvme/host/fabrics.h b/drivers/nvme/host/fabrics.h
index 06cc54851b1b..a4def76a182d 100644
--- a/drivers/nvme/host/fabrics.h
+++ b/drivers/nvme/host/fabrics.h
@@ -18,12 +18,10 @@
/* default is -1: the fail fast mechanism is disabled */
#define NVMF_DEF_FAIL_FAST_TMO -1
-/*
- * Reserved one command for internal usage. This command is used for sending
- * the connect command, as well as for the keep alive command on the admin
- * queue once live.
- */
-#define NVMF_RESERVED_TAGS 1
+/* Reserved for connect */
+#define NVMF_IO_RESERVED_TAGS 1
+/* Reserved for connect and keep alive */
+#define NVMF_ADMIN_RESERVED_TAGS 2
/*
* Define a host as seen by the target. We allocate one at boot, but also
--
2.25.1
On 07/03/2024 13:06, brookxu.cn wrote:
> From: Chunguang Xu <[email protected]>
>
> We found a issue on production environment while using NVMe
> over RDMA, admin_q reconnect failed forever while remote
> target and network is ok. After dig into it, we found it
> may caused by a ABBA deadlock due to tag allocation. In my
> case, the tag was hold by a keep alive request waiting
> inside admin_q, as we quiesced admin_q while reset ctrl,
> so the request maked as idle and will not process before
> reset success. As fabric_q shares tagset with admin_q,
> while reconnect remote target, we need a tag for connect
> command, but the only one reserved tag was held by keep
> alive command which waiting inside admin_q. As a result,
> we failed to reconnect admin_q forever. In order to fix
> this issue, I think we should keep two reserved tags for
> admin queue.
>
> Fixes: ed01fee283a0 ("nvme-fabrics: only reserve a single tag")
> Signed-off-by: Chunguang Xu <[email protected]>
Reviewed-by: Sagi Grimberg <[email protected]>
On 3/7/24 03:06, brookxu.cn wrote:
> From: Chunguang Xu <[email protected]>
>
> We found a issue on production environment while using NVMe
> over RDMA, admin_q reconnect failed forever while remote
> target and network is ok. After dig into it, we found it
> may caused by a ABBA deadlock due to tag allocation. In my
> case, the tag was hold by a keep alive request waiting
> inside admin_q, as we quiesced admin_q while reset ctrl,
> so the request maked as idle and will not process before
> reset success. As fabric_q shares tagset with admin_q,
> while reconnect remote target, we need a tag for connect
> command, but the only one reserved tag was held by keep
> alive command which waiting inside admin_q. As a result,
> we failed to reconnect admin_q forever. In order to fix
> this issue, I think we should keep two reserved tags for
> admin queue.
plz consider rearranged line length, no change in wording to use the
full length :-
We found a issue on production environment while using NVMe over RDMA,
admin_q reconnect failed forever while remote target and network is ok.
After dig into it, we found it may caused by a ABBA deadlock due to tag
allocation. In my case, the tag was hold by a keep alive request
waiting inside admin_q, as we quiesced admin_q while reset ctrl, so the
request maked as idle and will not process before reset success. As
fabric_q shares tagset with admin_q, while reconnect remote target, we
need a tag for connect command, but the only one reserved tag was held
by keep alive command which waiting inside admin_q. As a result, we
failed to reconnect admin_q forever. In order to fix this issue, I think
we should keep two reserved tags for admin queue.
Rest of the patch looks good and follows the discussion on V1.
Reviewed-by: Chaitanya Kulkarni <[email protected]>
-ck