reserved tag allocation fixups
Subsystems (3)
Christoph Hellwig
maintainerHi all,
this series is based on the report form Hannes, and first reduces the reserved tags on the admin queue from 2 to 1 and then uses a nowait allocation for the keep alive request to avoid a deadlock during unlikely error scenarios.
_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
Christoph Hellwig
maintainerPATCH 1/3 nvme-fabrics: only reserve a single tag
Fabrics drivers currently reserve two tags on the admin queue. But given that the connect command is only run on a freshly created queue or after all commands have been force aborted we only need to reserve a single tag.
--- drivers/nvme/host/fabrics.h | 7 +++++++ drivers/nvme/host/fc.c | 4 ++-- drivers/nvme/host/rdma.c | 4 ++-- drivers/nvme/host/tcp.c | 4 ++-- drivers/nvme/target/loop.c | 4 ++-- 5 files changed, 15 insertions(+), 8 deletions(-)
drivers/nvme/host/fabrics.h +7 -0
@@ -18,6 +18,13 @@/* 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 +/* * Define a host as seen by the target. We allocate one at boot, but also * allow the override it when creating controllers. This is both to provide
drivers/nvme/host/fc.c +2 -2
@@ -2862,7 +2862,7 @@ nvme_fc_create_io_queues(struct nvme_fc_ctrl *ctrl)memset (&ctrl->tag_set,0 ,sizeof (ctrl->tag_set)); ctrl->tag_set.ops = &nvme_fc_mq_ops; ctrl->tag_set.queue_depth = ctrl->ctrl.opts->queue_size; - ctrl->tag_set.reserved_tags =1 ;/* fabric connect */ + ctrl->tag_set.reserved_tags =NVMF_RESERVED_TAGS ; ctrl->tag_set.numa_node = ctrl->ctrl.numa_node; ctrl->tag_set.flags =BLK_MQ_F_SHOULD_MERGE ; ctrl->tag_set.cmd_size =
@@ -3484,7 +3484,7 @@ nvme_fc_init_ctrl(struct device *dev, struct nvmf_ctrl_options *opts,memset (&ctrl->admin_tag_set,0 ,sizeof (ctrl->admin_tag_set)); ctrl->admin_tag_set.ops = &nvme_fc_admin_mq_ops; ctrl->admin_tag_set.queue_depth =NVME_AQ_MQ_TAG_DEPTH ; - ctrl->admin_tag_set.reserved_tags =2 ;/* fabric connect + Keep-Alive */ + ctrl->admin_tag_set.reserved_tags =NVMF_RESERVED_TAGS ; ctrl->admin_tag_set.numa_node = ctrl->ctrl.numa_node; ctrl->admin_tag_set.cmd_size = struct_size((struct nvme_fcp_op_w_sgl *)NULL, priv,
drivers/nvme/host/rdma.c +2 -2
@@ -798,7 +798,7 @@ static struct blk_mq_tag_set *nvme_rdma_alloc_tagset(struct nvme_ctrl *nctrl,memset (set,0 ,sizeof (*set)); set->ops = &nvme_rdma_admin_mq_ops; set->queue_depth =NVME_AQ_MQ_TAG_DEPTH ; - set->reserved_tags =2 ;/* connect + keep-alive */ + set->reserved_tags =NVMF_RESERVED_TAGS ; set->numa_node = nctrl->numa_node; set->cmd_size =sizeof (struct nvme_rdma_request ) +NVME_RDMA_DATA_SGL_SIZE ;
@@ -811,7 +811,7 @@ static struct blk_mq_tag_set *nvme_rdma_alloc_tagset(struct nvme_ctrl *nctrl,memset (set,0 ,sizeof (*set)); set->ops = &nvme_rdma_mq_ops; set->queue_depth = nctrl->sqsize +1 ; - set->reserved_tags =1 ;/* fabric connect */ + set->reserved_tags =NVMF_RESERVED_TAGS ; set->numa_node = nctrl->numa_node; set->flags =BLK_MQ_F_SHOULD_MERGE ; set->cmd_size = sizeof(struct nvme_rdma_request) +
drivers/nvme/host/tcp.c +2 -2
@@ -1575,7 +1575,7 @@ static struct blk_mq_tag_set *nvme_tcp_alloc_tagset(struct nvme_ctrl *nctrl,memset (set,0 ,sizeof (*set)); set->ops = &nvme_tcp_admin_mq_ops; set->queue_depth =NVME_AQ_MQ_TAG_DEPTH ; - set->reserved_tags =2 ;/* connect + keep-alive */ + set->reserved_tags =NVMF_RESERVED_TAGS ; set->numa_node = nctrl->numa_node; set->flags =BLK_MQ_F_BLOCKING ; set->cmd_size =sizeof (struct nvme_tcp_request );
@@ -1587,7 +1587,7 @@ static struct blk_mq_tag_set *nvme_tcp_alloc_tagset(struct nvme_ctrl *nctrl,memset (set,0 ,sizeof (*set)); set->ops = &nvme_tcp_mq_ops; set->queue_depth = nctrl->sqsize +1 ; - set->reserved_tags =1 ;/* fabric connect */ + set->reserved_tags =NVMF_RESERVED_TAGS ; set->numa_node = nctrl->numa_node; set->flags =BLK_MQ_F_SHOULD_MERGE |BLK_MQ_F_BLOCKING ; set->cmd_size =sizeof (struct nvme_tcp_request );
drivers/nvme/target/loop.c +2 -2
@@ -349,7 +349,7 @@ static int nvme_loop_configure_admin_queue(struct nvme_loop_ctrl *ctrl)memset (&ctrl->admin_tag_set,0 ,sizeof (ctrl->admin_tag_set)); ctrl->admin_tag_set.ops = &nvme_loop_admin_mq_ops; ctrl->admin_tag_set.queue_depth =NVME_AQ_MQ_TAG_DEPTH ; - ctrl->admin_tag_set.reserved_tags =2 ;/* connect + keep-alive */ + ctrl->admin_tag_set.reserved_tags =NVMF_RESERVED_TAGS ; ctrl->admin_tag_set.numa_node = ctrl->ctrl.numa_node; ctrl->admin_tag_set.cmd_size =sizeof (struct nvme_loop_iod ) +NVME_INLINE_SG_CNT *sizeof (struct scatterlist );
@@ -520,7 +520,7 @@ static int nvme_loop_create_io_queues(struct nvme_loop_ctrl *ctrl)memset (&ctrl->tag_set,0 ,sizeof (ctrl->tag_set)); ctrl->tag_set.ops = &nvme_loop_mq_ops; ctrl->tag_set.queue_depth = ctrl->ctrl.opts->queue_size; - ctrl->tag_set.reserved_tags =1 ;/* fabric connect */ + ctrl->tag_set.reserved_tags =NVMF_RESERVED_TAGS ; ctrl->tag_set.numa_node = ctrl->ctrl.numa_node; ctrl->tag_set.flags =BLK_MQ_F_SHOULD_MERGE ; ctrl->tag_set.cmd_size = sizeof(struct nvme_loop_iod) +
Signature · 7 lines
2.29.2 _______________________________________________ Linux-nvme mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-nvme
Re: [PATCH 1/3] nvme-fabrics: only reserve a single tag
Quoted 6 lines · Christoph Hellwig
Fabrics drivers currently reserve two tags on the admin queue. But given that the connect command is only run on a freshly created queue or after all commands have been force aborted we only need to reserve a single tag.
Signed-off-by Christoph Hellwig <[email protected]>
_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
Re: [PATCH 1/3] nvme-fabrics: only reserve a single tag
In reply to Christoph HellwigQuoted 6 lines · Christoph Hellwig
Fabrics drivers currently reserve two tags on the admin queue. But given that the connect command is only run on a freshly created queue or after all commands have been force aborted we only need to reserve a single tag.
Signed-off-by Christoph Hellwig <[email protected]>
Thanks a lot for creating a macro with nice comment, looks good.
_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
Re: [PATCH 1/3] nvme-fabrics: only reserve a single tag
In reply to Christoph HellwigQuoted 5 lines · Christoph Hellwig
Fabrics drivers currently reserve two tags on the admin queue. But given that the connect command is only run on a freshly created queue or after all commands have been force aborted we only need to reserve a single tag.
Quoted 9 lines
Signed-off-by Christoph Hellwig <[email protected]>--- drivers/nvme/host/fabrics.h | 7 +++++++ drivers/nvme/host/fc.c | 4 ++-- drivers/nvme/host/rdma.c | 4 ++-- drivers/nvme/host/tcp.c | 4 ++-- drivers/nvme/target/loop.c | 4 ++-- 5 files changed, 15 insertions(+), 8 deletions(-)
Cheers,
Hannes
Signature · 9 lines
Dr. Hannes Reinecke Kernel Storage Architect [email protected] +49 911 74053 688 SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer _______________________________________________ Linux-nvme mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-nvme
Sagi Grimberg
maintainerRe: [PATCH 1/3] nvme-fabrics: only reserve a single tag
In reply to Christoph HellwigQuoted 5 lines
Fabrics drivers currently reserve two tags on the admin queue. But given that the connect command is only run on a freshly created queue or after all commands have been force aborted we only need to reserve a single tag.
Umm, I think this would be an issue for non-mpath fabrics devices...
When we teardown the controller, we iterate over all tags and cancel them, however for non-mpath devices these actually stick around until they exhaust the retrys counter (with the hope that the controller will reconnect again) unlike the mpath case where the requests are failed-over.
Now if the admin queue is absolutely full during a reset, we won't have a keep-alive command.
One possible solution is to make sure that nvme_decide_disposition to complete a request for admin commands. However note that this means that admin commands would fail immediately when the controller resets, which is a regression from the existing behavior and we may piss off users with that...
_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
Christoph Hellwig
maintainerRe: [PATCH 1/3] nvme-fabrics: only reserve a single tag
Quoted 17 lines · Sagi Grimberg
Quoted 5 lines
Fabrics drivers currently reserve two tags on the admin queue. But given that the connect command is only run on a freshly created queue or after all commands have been force aborted we only need to reserve a single tag.
Umm, I think this would be an issue for non-mpath fabrics devices...
When we teardown the controller, we iterate over all tags and cancel them, however for non-mpath devices these actually stick around until they exhaust the retrys counter (with the hope that the controller will reconnect again) unlike the mpath case where the requests are failed-over.
Now if the admin queue is absolutely full during a reset, we won't have a keep-alive command.
Quoted 6 lines
One possible solution is to make sure that
nvme_decide_dispositionto complete a request for admin commands. However note that this means that admin commands would fail immediately when the controller resets, which is a regression from the existing behavior and we may piss off users with that...
We never retry admin commands! And the reserved tags are set aside and not relevant to the "normal" commands.
_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
Re: [PATCH 1/3] nvme-fabrics: only reserve a single tag
Quoted 24 lines
Quoted 17 lines
Quoted 5 lines
Fabrics drivers currently reserve two tags on the admin queue. But given that the connect command is only run on a freshly created queue or after all commands have been force aborted we only need to reserve a single tag.
Umm, I think this would be an issue for non-mpath fabrics devices...
When we teardown the controller, we iterate over all tags and cancel them, however for non-mpath devices these actually stick around until they exhaust the retrys counter (with the hope that the controller will reconnect again) unlike the mpath case where the requests are failed-over.
Now if the admin queue is absolutely full during a reset, we won't have a keep-alive command.
Quoted 6 lines
One possible solution is to make sure that
nvme_decide_dispositionto complete a request for admin commands. However note that this means that admin commands would fail immediately when the controller resets, which is a regression from the existing behavior and we may piss off users with that...
Quoted 2 lines
We never retry admin commands! And the reserved tags are set aside and not relevant to the "normal" commands.
You're right, admin commands will be noretry commands. this looks good to me.
_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
Christoph Hellwig
maintainerPATCH 2/3 nvme: merge nvme_keep_alive into nvme_keep_alive_work
In reply to Christoph HellwigMerge nvme_keep_alive into its only caller to prepare for additional changes to this code.
--- drivers/nvme/host/core.c | 26 ++++++++------------------ 1 file changed, 8 insertions(+), 18 deletions(-)
drivers/nvme/host/core.c +8 -18
@@ -1225,28 +1225,12 @@ static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)queue_delayed_work (nvme_wq, &ctrl->ka_work, ctrl->kato *HZ ); } -static int nvme_keep_alive (struct nvme_ctrl *ctrl) -{ -struct request *rq; - - rq =nvme_alloc_request (ctrl->admin_q, &ctrl->ka_cmd, -BLK_MQ_REQ_RESERVED ); -if (IS_ERR (rq)) -return PTR_ERR (rq); - - rq->timeout = ctrl->kato *HZ ; - rq->end_io_data = ctrl; - -blk_execute_rq_nowait (NULL , rq,0 , nvme_keep_alive_end_io); - -return 0 ; -} -static void nvme_keep_alive_work (struct work_struct *work) {struct nvme_ctrl *ctrl =container_of (to_delayed_work (work), struct nvme_ctrl, ka_work);bool comp_seen = ctrl->comp_seen; +struct request *rq;if ((ctrl->ctratt &NVME_CTRL_ATTR_TBKAS ) && comp_seen) { dev_dbg(ctrl->device,
@@ -1256,12 +1240,18 @@ static void nvme_keep_alive_work(struct work_struct *work)return ; } -if (nvme_keep_alive (ctrl)) { + rq =nvme_alloc_request (ctrl->admin_q, &ctrl->ka_cmd, +BLK_MQ_REQ_RESERVED ); +if (IS_ERR (rq)) {/* allocation failure, reset the controller */ dev_err (ctrl->device,"keep-alive failed\n" );nvme_reset_ctrl (ctrl);return ; } + + rq->timeout = ctrl->kato *HZ ; + rq->end_io_data = ctrl; +blk_execute_rq_nowait (NULL , rq,0 , nvme_keep_alive_end_io); }static void nvme_start_keep_alive (struct nvme_ctrl *ctrl)
Signature · 7 lines
2.29.2 _______________________________________________ Linux-nvme mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-nvme
Re: [PATCH 2/3] nvme: merge nvme_keep_alive into nvme_keep_alive_work
Quoted 4 lines · Christoph Hellwig
Merge
nvme_keep_aliveinto its only caller to prepare for additional changes to this code.Signed-off-by Christoph Hellwig <[email protected]>
_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
Re: [PATCH 2/3] nvme: merge nvme_keep_alive into nvme_keep_alive_work
In reply to Christoph HellwigQuoted 4 lines · Christoph Hellwig
Merge
nvme_keep_aliveinto its only caller to prepare for additional changes to this code.Signed-off-by Christoph Hellwig <[email protected]>
Looks good.
_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
Re: [PATCH 2/3] nvme: merge nvme_keep_alive into nvme_keep_alive_work
In reply to Christoph HellwigQuoted 4 lines · Christoph Hellwig
Merge
nvme_keep_aliveinto its only caller to prepare for additional changes to this code.Signed-off-by Christoph Hellwig <[email protected]>
Looks good.
_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
Re: [PATCH 2/3] nvme: merge nvme_keep_alive into nvme_keep_alive_work
In reply to Christoph HellwigQuoted 8 lines · Christoph Hellwig
Merge
nvme_keep_aliveinto its only caller to prepare for additional changes to this code.Signed-off-by Christoph Hellwig <[email protected]>--- drivers/nvme/host/core.c | 26 ++++++++------------------ 1 file changed, 8 insertions(+), 18 deletions(-)
Cheers,
Hannes
Signature · 9 lines
Dr. Hannes Reinecke Kernel Storage Architect [email protected] +49 911 74053 688 SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer _______________________________________________ Linux-nvme mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-nvme
Re: [PATCH 2/3] nvme: merge nvme_keep_alive into nvme_keep_alive_work
In reply to Christoph Hellwig_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
Christoph Hellwig
maintainerPATCH 3/3 nvme: allocate the keep alive request using BLK_MQ_REQ_NOWAIT
In reply to Christoph HellwigTo avoid an error recovery deadlock where the keep alive work is waiting for a request and thus can't be flushed to make progress for tearing down the controller. Also print the error code returned from blk_mq_alloc_request to help debugging any future issues in this code.
Based on an earlier patch from Hannes Reinecke <[email protected]>.
--- drivers/nvme/host/core.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/nvme/host/core.c +2 -2
@@ -1241,10 +1241,10 @@ static void nvme_keep_alive_work(struct work_struct *work) } rq =nvme_alloc_request (ctrl->admin_q, &ctrl->ka_cmd, -BLK_MQ_REQ_RESERVED ); +BLK_MQ_REQ_RESERVED |BLK_MQ_REQ_NOWAIT );if (IS_ERR (rq)) {/* allocation failure, reset the controller */ -dev_err (ctrl->device,"keep-alive failed\n" ); +dev_err (ctrl->device,"keep-alive failed: %ld\n" ,PTR_ERR (rq));nvme_reset_ctrl (ctrl);return ; }
Signature · 7 lines
2.29.2 _______________________________________________ Linux-nvme mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-nvme
Re: [PATCH 3/3] nvme: allocate the keep alive request using BLK_MQ_REQ_NOWAIT
Quoted 8 lines · Christoph Hellwig
To avoid an error recovery deadlock where the keep alive work is waiting for a request and thus can't be flushed to make progress for tearing down the controller. Also print the error code returned from
blk_mq_alloc_requestto help debugging any future issues in this code.Based on an earlier patch from Hannes Reinecke <[email protected]>.
Signed-off-by Christoph Hellwig <[email protected]>
And for what's worth, our customer tested this. I've done the same patch during debugging. It fixes the deadlock as Hannes reported.
_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
Re: [PATCH 3/3] nvme: allocate the keep alive request using BLK_MQ_REQ_NOWAIT
In reply to Christoph HellwigQuoted 8 lines · Christoph Hellwig
To avoid an error recovery deadlock where the keep alive work is waiting for a request and thus can't be flushed to make progress for tearing down the controller. Also print the error code returned from
blk_mq_alloc_requestto help debugging any future issues in this code.Based on an earlier patch from Hannes Reinecke <[email protected]>.
Signed-off-by Christoph Hellwig <[email protected]>
Looks good.
_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
Re: [PATCH 3/3] nvme: allocate the keep alive request using BLK_MQ_REQ_NOWAIT
In reply to Christoph HellwigQuoted 20 lines · Christoph Hellwig
To avoid an error recovery deadlock where the keep alive work is waiting for a request and thus can't be flushed to make progress for tearing down the controller. Also print the error code returned from
blk_mq_alloc_requestto help debugging any future issues in this code.Based on an earlier patch from Hannes Reinecke <[email protected]>.
Signed-off-by Christoph Hellwig <[email protected]>--- drivers/nvme/host/core.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)drivers/nvme/host/core.c +0
-0@@ -1241,10 +1241,10 @@ static void nvme_keep_alive_work(struct work_struct *work) }
drivers/nvme/host/core.c static void nvme_keep_alive_work(struct work_struct *work) · line 1250
Quoted 10 lines
drivers/nvme/host/core.c +2
-2rq =nvme_alloc_request (ctrl->admin_q, &ctrl->ka_cmd, -BLK_MQ_REQ_RESERVED ); +BLK_MQ_REQ_RESERVED |BLK_MQ_REQ_NOWAIT );if (IS_ERR (rq)) {/* allocation failure, reset the controller */ -dev_err (ctrl->device,"keep-alive failed\n" ); +dev_err (ctrl->device,"keep-alive failed: %ld\n" ,PTR_ERR (rq));nvme_reset_ctrl (ctrl);return ; }
Cheers,
Hannes
Signature · 9 lines
Dr. Hannes Reinecke Kernel Storage Architect [email protected] +49 911 74053 688 SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer _______________________________________________ Linux-nvme mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-nvme
Re: [PATCH 3/3] nvme: allocate the keep alive request using BLK_MQ_REQ_NOWAIT
In reply to Christoph Hellwig_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme