[PATCH 06/10] nvme/nvme-fabrics: introduce nvmf_reconnect_ctrl_work API

[James Smart]

On 10/20/2021 3:38 AM, Max Gurtovoy wrote:
> Reconnect work is duplicated in RDMA and TCP transports. Move this logic
> to common code. For that, introduce a new ctrl op to setup a ctrl.
> 
> Also update the RDMA/TCP transport drivers to use this API and remove
> the duplicated code.
> 
> Reviewed-by: Israel Rukshin <israelr at nvidia.com>
> Reviewed-by: Chaitanya Kulkarni <kch at nvidia.com>
> Signed-off-by: Max Gurtovoy <mgurtovoy at nvidia.com>
> ---
>   drivers/nvme/host/fabrics.c | 24 +++++++++++
>   drivers/nvme/host/fabrics.h |  1 +
>   drivers/nvme/host/nvme.h    |  1 +
>   drivers/nvme/host/rdma.c    | 82 ++++++++++++++-----------------------
>   drivers/nvme/host/tcp.c     | 28 +------------
>   5 files changed, 58 insertions(+), 78 deletions(-)
> 
> diff --git a/drivers/nvme/host/fabrics.c b/drivers/nvme/host/fabrics.c
> index 5a770196eb60..6a2283e09164 100644
> --- a/drivers/nvme/host/fabrics.c
> +++ b/drivers/nvme/host/fabrics.c
> @@ -526,6 +526,30 @@ void nvmf_error_recovery(struct nvme_ctrl *ctrl)
>   }
>   EXPORT_SYMBOL_GPL(nvmf_error_recovery);
>   
> +void nvmf_reconnect_ctrl_work(struct work_struct *work)
> +{
> +	struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
> +			struct nvme_ctrl, connect_work);
> +
> +	++ctrl->nr_reconnects;
> +
> +	if (ctrl->ops->setup_ctrl(ctrl, false))
> +		goto requeue;
> +
> +	dev_info(ctrl->device, "Successfully reconnected (%d attempt)\n",
> +			ctrl->nr_reconnects);
> +
> +	ctrl->nr_reconnects = 0;
> +
> +	return;
> +
> +requeue:
> +	dev_info(ctrl->device, "Failed reconnect attempt %d\n",
> +		 ctrl->nr_reconnects);
> +	nvmf_reconnect_or_remove(ctrl);
> +}
> +EXPORT_SYMBOL_GPL(nvmf_reconnect_ctrl_work);
> +
>   /**
>    * nvmf_register_transport() - NVMe Fabrics Library registration function.
>    * @ops:	Transport ops instance to be registered to the
> diff --git a/drivers/nvme/host/fabrics.h b/drivers/nvme/host/fabrics.h
> index 8655eff74ed0..49c98b69647f 100644
> --- a/drivers/nvme/host/fabrics.h
> +++ b/drivers/nvme/host/fabrics.h
> @@ -191,6 +191,7 @@ bool nvmf_should_reconnect(struct nvme_ctrl *ctrl);
>   void nvmf_reconnect_or_remove(struct nvme_ctrl *ctrl);
>   void nvmf_error_recovery(struct nvme_ctrl *ctrl);
>   void nvmf_error_recovery_work(struct work_struct *work);
> +void nvmf_reconnect_ctrl_work(struct work_struct *work);
>   bool nvmf_ip_options_match(struct nvme_ctrl *ctrl,
>   		struct nvmf_ctrl_options *opts);
>   
> diff --git a/drivers/nvme/host/nvme.h b/drivers/nvme/host/nvme.h
> index 5cdf2ec45e9a..e137db2760d8 100644
> --- a/drivers/nvme/host/nvme.h
> +++ b/drivers/nvme/host/nvme.h
> @@ -497,6 +497,7 @@ struct nvme_ctrl_ops {
>   	/* Fabrics only */
>   	void (*teardown_ctrl_io_queues)(struct nvme_ctrl *ctrl, bool remove);
>   	void (*teardown_ctrl_admin_queue)(struct nvme_ctrl *ctrl, bool remove);
> +	int (*setup_ctrl)(struct nvme_ctrl *ctrl, bool new);
>   };
>   
>   /*


Add in a routine to call to check for connectivity and we're pretty 
close to something FC could adapt to.

But I'd really like to get rid of the "new" argument that gets passed 
around. We should be able to have some kind of state on the fabric 
controller that allows us to derive new or not (fc did this - see 
ioq_live).

-- james