Commit e7832cb4 authored by Sagi Grimberg's avatar Sagi Grimberg
Browse files

nvme: make fabrics command run on a separate request queue 



We have a fundamental issue that fabric commands use the admin_q.
The reason is, that admin-connect, register reads and writes and
admin commands cannot be guaranteed ordering while we are running
controller resets.

For example, when we reset a controller we perform:
1. disable the controller
2. teardown the admin queue
3. re-establish the admin queue
4. enable the controller

In order to perform (3), we need to unquiesce the admin queue, however
we may have some admin commands that are already pending on the
quiesced admin_q and will immediate execute when we unquiesce it before
we execute (4). The host must not send admin commands to the controller
before enabling the controller.

To fix this, we have the fabric commands (admin connect and property
get/set, but not I/O queue connect) use a separate fabrics_q and make
sure to quiesce the admin_q before we disable the controller, and
unquiesce it only after we enable the controller.

This fixes the error prints from nvmet in a controller reset storm test:
kernel: nvmet: got cmd 6 while CC.EN == 0 on qid = 0
Which indicate that the host is sending an admin command when the
controller is not enabled.

Reviewed-by: default avatarJames Smart <james.smart@broadcom.com>
Reviewed-by: default avatarChristoph Hellwig <hch@lst.de>
Signed-off-by: default avatarSagi Grimberg <sagi@grimberg.me>
parent d38e9f04

drivers/nvme/host/fabrics.c

+4 −4
Original line number Diff line number Diff line
@@ -150,7 +150,7 @@ int nvmf_reg_read32(struct nvme_ctrl *ctrl, u32 off, u32 *val) 
	cmd.prop_get.fctype = nvme_fabrics_type_property_get;

	cmd.prop_get.offset = cpu_to_le32(off);



	ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, &res, NULL, 0, 0,

	ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res, NULL, 0, 0,

			NVME_QID_ANY, 0, 0, false);



	if (ret >= 0)
@@ -197,7 +197,7 @@ int nvmf_reg_read64(struct nvme_ctrl *ctrl, u32 off, u64 *val) 
	cmd.prop_get.attrib = 1;

	cmd.prop_get.offset = cpu_to_le32(off);



	ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, &res, NULL, 0, 0,

	ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res, NULL, 0, 0,

			NVME_QID_ANY, 0, 0, false);



	if (ret >= 0)
@@ -243,7 +243,7 @@ int nvmf_reg_write32(struct nvme_ctrl *ctrl, u32 off, u32 val) 
	cmd.prop_set.offset = cpu_to_le32(off);

	cmd.prop_set.value = cpu_to_le64(val);



	ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, NULL, 0, 0,

	ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, NULL, NULL, 0, 0,

			NVME_QID_ANY, 0, 0, false);

	if (unlikely(ret))

		dev_err(ctrl->device,
@@ -396,7 +396,7 @@ int nvmf_connect_admin_queue(struct nvme_ctrl *ctrl) 
	strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);

	strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);



	ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, &res,

	ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res,

			data, sizeof(*data), 0, NVME_QID_ANY, 1,

			BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT, false);

	if (ret) {

drivers/nvme/host/fc.c

+12 −3
Original line number Diff line number Diff line
@@ -2006,6 +2006,7 @@ nvme_fc_ctrl_free(struct kref *ref) 


	blk_mq_unquiesce_queue(ctrl->ctrl.admin_q);

	blk_cleanup_queue(ctrl->ctrl.admin_q);

	blk_cleanup_queue(ctrl->ctrl.fabrics_q);

	blk_mq_free_tag_set(&ctrl->admin_tag_set);



	kfree(ctrl->queues);
@@ -2633,8 +2634,6 @@ nvme_fc_create_association(struct nvme_fc_ctrl *ctrl) 
	if (ret)

		goto out_delete_hw_queue;



	blk_mq_unquiesce_queue(ctrl->ctrl.admin_q);



	ret = nvmf_connect_admin_queue(&ctrl->ctrl);

	if (ret)

		goto out_disconnect_admin_queue;
@@ -2655,6 +2654,8 @@ nvme_fc_create_association(struct nvme_fc_ctrl *ctrl) 
	ctrl->ctrl.max_hw_sectors =

		(ctrl->lport->ops->max_sgl_segments - 1) << (PAGE_SHIFT - 9);



	blk_mq_unquiesce_queue(ctrl->ctrl.admin_q);



	ret = nvme_init_identify(&ctrl->ctrl);

	if (ret)

		goto out_disconnect_admin_queue;
@@ -3101,10 +3102,16 @@ nvme_fc_init_ctrl(struct device *dev, struct nvmf_ctrl_options *opts, 
		goto out_free_queues;

	ctrl->ctrl.admin_tagset = &ctrl->admin_tag_set;



	ctrl->ctrl.fabrics_q = blk_mq_init_queue(&ctrl->admin_tag_set);

	if (IS_ERR(ctrl->ctrl.fabrics_q)) {

		ret = PTR_ERR(ctrl->ctrl.fabrics_q);

		goto out_free_admin_tag_set;

	}



	ctrl->ctrl.admin_q = blk_mq_init_queue(&ctrl->admin_tag_set);

	if (IS_ERR(ctrl->ctrl.admin_q)) {

		ret = PTR_ERR(ctrl->ctrl.admin_q);

		goto out_free_admin_tag_set;

		goto out_cleanup_fabrics_q;

	}



	/*
@@ -3176,6 +3183,8 @@ fail_ctrl: 


out_cleanup_admin_q:

	blk_cleanup_queue(ctrl->ctrl.admin_q);

out_cleanup_fabrics_q:

	blk_cleanup_queue(ctrl->ctrl.fabrics_q);

out_free_admin_tag_set:

	blk_mq_free_tag_set(&ctrl->admin_tag_set);

out_free_queues:

drivers/nvme/host/nvme.h

+1 −0
Original line number Diff line number Diff line
@@ -181,6 +181,7 @@ struct nvme_ctrl { 
	const struct nvme_ctrl_ops *ops;

	struct request_queue *admin_q;

	struct request_queue *connect_q;

	struct request_queue *fabrics_q;

	struct device *dev;

	int instance;

	int numa_node;

drivers/nvme/host/rdma.c

+17 −2
Original line number Diff line number Diff line
@@ -751,6 +751,7 @@ static void nvme_rdma_destroy_admin_queue(struct nvme_rdma_ctrl *ctrl, 
{

	if (remove) {

		blk_cleanup_queue(ctrl->ctrl.admin_q);

		blk_cleanup_queue(ctrl->ctrl.fabrics_q);

		blk_mq_free_tag_set(ctrl->ctrl.admin_tagset);

	}

	if (ctrl->async_event_sqe.data) {
@@ -792,10 +793,16 @@ static int nvme_rdma_configure_admin_queue(struct nvme_rdma_ctrl *ctrl, 
			goto out_free_async_qe;

		}



		ctrl->ctrl.fabrics_q = blk_mq_init_queue(&ctrl->admin_tag_set);

		if (IS_ERR(ctrl->ctrl.fabrics_q)) {

			error = PTR_ERR(ctrl->ctrl.fabrics_q);

			goto out_free_tagset;

		}



		ctrl->ctrl.admin_q = blk_mq_init_queue(&ctrl->admin_tag_set);

		if (IS_ERR(ctrl->ctrl.admin_q)) {

			error = PTR_ERR(ctrl->ctrl.admin_q);

			goto out_free_tagset;

			goto out_cleanup_fabrics_q;

		}

	}
@@ -810,6 +817,8 @@ static int nvme_rdma_configure_admin_queue(struct nvme_rdma_ctrl *ctrl, 
	ctrl->ctrl.max_hw_sectors =

		(ctrl->max_fr_pages - 1) << (ilog2(SZ_4K) - 9);



	blk_mq_unquiesce_queue(ctrl->ctrl.admin_q);



	error = nvme_init_identify(&ctrl->ctrl);

	if (error)

		goto out_stop_queue;
@@ -821,6 +830,9 @@ out_stop_queue: 
out_cleanup_queue:

	if (new)

		blk_cleanup_queue(ctrl->ctrl.admin_q);

out_cleanup_fabrics_q:

	if (new)

		blk_cleanup_queue(ctrl->ctrl.fabrics_q);

out_free_tagset:

	if (new)

		blk_mq_free_tag_set(ctrl->ctrl.admin_tagset);
@@ -895,6 +907,7 @@ static void nvme_rdma_teardown_admin_queue(struct nvme_rdma_ctrl *ctrl, 
			nvme_cancel_request, &ctrl->ctrl);

		blk_mq_tagset_wait_completed_request(ctrl->ctrl.admin_tagset);

	}

	if (remove)

		blk_mq_unquiesce_queue(ctrl->ctrl.admin_q);

	nvme_rdma_destroy_admin_queue(ctrl, remove);

}
@@ -1046,6 +1059,7 @@ static void nvme_rdma_error_recovery_work(struct work_struct *work) 
	nvme_rdma_teardown_io_queues(ctrl, false);

	nvme_start_queues(&ctrl->ctrl);

	nvme_rdma_teardown_admin_queue(ctrl, false);

	blk_mq_unquiesce_queue(ctrl->ctrl.admin_q);



	if (!nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_CONNECTING)) {

		/* state change failure is ok if we're in DELETING state */
@@ -1858,6 +1872,7 @@ static void nvme_rdma_shutdown_ctrl(struct nvme_rdma_ctrl *ctrl, bool shutdown) 
	cancel_delayed_work_sync(&ctrl->reconnect_work);



	nvme_rdma_teardown_io_queues(ctrl, shutdown);

	blk_mq_quiesce_queue(ctrl->ctrl.admin_q);

	if (shutdown)

		nvme_shutdown_ctrl(&ctrl->ctrl);

	else

drivers/nvme/host/tcp.c

+17 −2
Original line number Diff line number Diff line
@@ -1703,6 +1703,7 @@ static void nvme_tcp_destroy_admin_queue(struct nvme_ctrl *ctrl, bool remove) 
	nvme_tcp_stop_queue(ctrl, 0);

	if (remove) {

		blk_cleanup_queue(ctrl->admin_q);

		blk_cleanup_queue(ctrl->fabrics_q);

		blk_mq_free_tag_set(ctrl->admin_tagset);

	}

	nvme_tcp_free_admin_queue(ctrl);
@@ -1723,10 +1724,16 @@ static int nvme_tcp_configure_admin_queue(struct nvme_ctrl *ctrl, bool new) 
			goto out_free_queue;

		}



		ctrl->fabrics_q = blk_mq_init_queue(ctrl->admin_tagset);

		if (IS_ERR(ctrl->fabrics_q)) {

			error = PTR_ERR(ctrl->fabrics_q);

			goto out_free_tagset;

		}



		ctrl->admin_q = blk_mq_init_queue(ctrl->admin_tagset);

		if (IS_ERR(ctrl->admin_q)) {

			error = PTR_ERR(ctrl->admin_q);

			goto out_free_tagset;

			goto out_cleanup_fabrics_q;

		}

	}
@@ -1738,6 +1745,8 @@ static int nvme_tcp_configure_admin_queue(struct nvme_ctrl *ctrl, bool new) 
	if (error)

		goto out_stop_queue;



	blk_mq_unquiesce_queue(ctrl->admin_q);



	error = nvme_init_identify(ctrl);

	if (error)

		goto out_stop_queue;
@@ -1749,6 +1758,9 @@ out_stop_queue: 
out_cleanup_queue:

	if (new)

		blk_cleanup_queue(ctrl->admin_q);

out_cleanup_fabrics_q:

	if (new)

		blk_cleanup_queue(ctrl->fabrics_q);

out_free_tagset:

	if (new)

		blk_mq_free_tag_set(ctrl->admin_tagset);
@@ -1767,6 +1779,7 @@ static void nvme_tcp_teardown_admin_queue(struct nvme_ctrl *ctrl, 
			nvme_cancel_request, ctrl);

		blk_mq_tagset_wait_completed_request(ctrl->admin_tagset);

	}

	if (remove)

		blk_mq_unquiesce_queue(ctrl->admin_q);

	nvme_tcp_destroy_admin_queue(ctrl, remove);

}
@@ -1894,6 +1907,7 @@ static void nvme_tcp_error_recovery_work(struct work_struct *work) 
	/* unquiesce to fail fast pending requests */

	nvme_start_queues(ctrl);

	nvme_tcp_teardown_admin_queue(ctrl, false);

	blk_mq_unquiesce_queue(ctrl->admin_q);



	if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_CONNECTING)) {

		/* state change failure is ok if we're in DELETING state */
@@ -1910,6 +1924,7 @@ static void nvme_tcp_teardown_ctrl(struct nvme_ctrl *ctrl, bool shutdown) 
	cancel_delayed_work_sync(&to_tcp_ctrl(ctrl)->connect_work);



	nvme_tcp_teardown_io_queues(ctrl, shutdown);

	blk_mq_quiesce_queue(ctrl->admin_q);

	if (shutdown)

		nvme_shutdown_ctrl(ctrl);

	else

CRA Git | Maintained and supported by SUSTech CRA and CCSE