Commit 849f05c9 authored by John Levon's avatar John Levon Committed by Tomasz Zawadzki
Browse files

nvmf/vfio-user: create queues directly 



Currently, queue creation is delegated to nvmf_vfio_user_accept() for both I/O
and admin queues, by adding it to a ->new_qps list. But there's no good reason
for this indirection to exist, and it would make interrupt support for the
accept poller harder to implement. Instead, directly call
spdk_nvmf_tgt_new_qpair().

Signed-off-by: default avatarJohn Levon <john.levon@nutanix.com>
Change-Id: I182bfac9c621eda2f746201db7022ab6fce1bf30
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/9053


Tested-by: default avatarSPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Mellanox Build Bot
Reviewed-by: default avatarChangpeng Liu <changpeng.liu@intel.com>
Reviewed-by: default avatarBen Walker <benjamin.walker@intel.com>
parent 785246f9

lib/nvmf/vfio_user.c

+10 −19
Original line number Diff line number Diff line
@@ -203,8 +203,6 @@ struct nvmf_vfio_user_transport { 
	struct nvmf_vfio_user_transport_opts    transport_opts;

	pthread_mutex_t				lock;

	TAILQ_HEAD(, nvmf_vfio_user_endpoint)	endpoints;



	TAILQ_HEAD(, nvmf_vfio_user_qpair)	new_qps;

};



/*
@@ -566,7 +564,6 @@ nvmf_vfio_user_create(struct spdk_nvmf_transport_opts *opts) 
	}



	TAILQ_INIT(&vu_transport->endpoints);

	TAILQ_INIT(&vu_transport->new_qps);



	if (opts->transport_specific != NULL &&

	    spdk_json_decode_object_relaxed(opts->transport_specific, vfio_user_transport_opts_decoder,
@@ -1138,15 +1135,17 @@ handle_create_io_q(struct nvmf_vfio_user_ctrlr *ctrlr, 
	if (is_cq) {

		*hdbl(ctrlr, io_q) = 0;

	} else {

		/* After we've returned here, on the next time nvmf_vfio_user_accept executes it will

		 * pick up this qpair and will eventually call nvmf_vfio_user_poll_group_add which will

		 * call spdk_nvmf_request_exec_fabrics with a generated fabrics connect command. That

		 * will then call handle_queue_connect_rsp, which is where we ultimately complete

		 * this command.

		/*

		 * Create our new I/O qpair. This asynchronously invokes, on a

		 * suitable poll group, the nvmf_vfio_user_poll_group_add()

		 * callback, which will call spdk_nvmf_request_exec_fabrics()

		 * with a generated fabrics connect command. This command is

		 * then eventually completed via handle_queue_connect_rsp().

		 */

		vu_qpair = ctrlr->qp[qid];

		vu_qpair->create_io_sq_cmd = *cmd;

		TAILQ_INSERT_TAIL(&ctrlr->transport->new_qps, vu_qpair, link);

		spdk_nvmf_tgt_new_qpair(ctrlr->transport->transport.tgt,

					&vu_qpair->qpair);

		*tdbl(ctrlr, io_q) = 0;

		return 0;

	}
@@ -1924,7 +1923,7 @@ nvmf_vfio_user_create_ctrlr(struct nvmf_vfio_user_transport *transport, 
	endpoint->ctrlr = ctrlr;



	/* Notify the generic layer about the new admin queue pair */

	TAILQ_INSERT_TAIL(&ctrlr->transport->new_qps, ctrlr->qp[0], link);

	spdk_nvmf_tgt_new_qpair(transport->transport.tgt, &ctrlr->qp[0]->qpair);



out:

	if (err != 0) {
@@ -2106,7 +2105,6 @@ static uint32_t 
nvmf_vfio_user_accept(struct spdk_nvmf_transport *transport)

{

	struct nvmf_vfio_user_transport *vu_transport;

	struct nvmf_vfio_user_qpair *qp, *tmp_qp;

	struct nvmf_vfio_user_endpoint *endpoint;

	uint32_t count = 0;

	int err;
@@ -2137,12 +2135,6 @@ nvmf_vfio_user_accept(struct spdk_nvmf_transport *transport) 
		nvmf_vfio_user_create_ctrlr(vu_transport, endpoint);

	}



	TAILQ_FOREACH_SAFE(qp, &vu_transport->new_qps, link, tmp_qp) {

		count++;

		TAILQ_REMOVE(&vu_transport->new_qps, qp, link);

		spdk_nvmf_tgt_new_qpair(transport->tgt, &qp->qpair);

	}



	pthread_mutex_unlock(&vu_transport->lock);



	return count;
@@ -2318,8 +2310,7 @@ handle_queue_connect_rsp(struct nvmf_vfio_user_req *req, void *cb_arg) 
}



/*

 * Add the given qpair to the given poll group. New qpairs are added to

 * ->new_qps; they are processed via nvmf_vfio_user_accept(), calling

 * Add the given qpair to the given poll group. New qpairs are added via

 * spdk_nvmf_tgt_new_qpair(), which picks a poll group, then calls back

 * here via nvmf_transport_poll_group_add().

 */