Skip to contentSkip to conversation

nvmet-rdma: remove p2p_client initialization from fast-path

PATCH New Max Gurtovoy 4 participants 9 messages Apr 8, 2019 – Apr 9, 2019

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/2 nvmet-rdma: remove p2p_client initialization from fast-path

Initialize it during command allocation.

Cc Logan Gunthorpe <logang at deltatee.com>
Cc Stephen Bates <sbates at raithlin.com>
Signed-off-by Max Gurtovoy <maxg at mellanox.com>
---
 drivers/nvme/target/rdma.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)
drivers/nvme/target/rdma.c +1 -2
index ef893ad..b727521 100644
@@ -373,6 +373,7 @@ static int nvmet_rdma_alloc_rsp(struct nvmet_rdma_device *ndev,
 	if (ib_dma_mapping_error(ndev->device, r->send_sge.addr))
 		goto out_free_rsp;
 
+	r->req.p2p_client = &ndev->device->dev;
 	r->send_sge.length = sizeof(*r->req.rsp);
 	r->send_sge.lkey = ndev->pd->local_dma_lkey;
 
@@ -763,8 +764,6 @@ static void nvmet_rdma_handle_command(struct nvmet_rdma_queue *queue,
 		cmd->send_sge.addr, cmd->send_sge.length,
 		DMA_TO_DEVICE);
 
-	cmd->req.p2p_client = &queue->dev->device->dev;
-
 	if (!nvmet_req_init(&cmd->req, &queue->nvme_cq,
 			&queue->nvme_sq, &nvmet_rdma_ops))
 		return;
Signature · 1 line
1.8.3.1
#1 lore raw

PATCH 2/2 nvmet: rename nvme_completion instances from rsp to cqe

Use NVMe namings for improving code readability.

Signed-off-by Max Gurtovoy <maxg at mellanox.com>
---
 drivers/nvme/target/core.c        | 22 +++++++++++-----------
 drivers/nvme/target/fabrics-cmd.c | 16 ++++++++--------
 drivers/nvme/target/fc.c          |  2 +-
 drivers/nvme/target/loop.c        |  6 +++---
 drivers/nvme/target/nvmet.h       |  4 ++--
 drivers/nvme/target/rdma.c        | 18 +++++++++---------
 drivers/nvme/target/tcp.c         |  8 ++++----
 7 files changed, 38 insertions(+), 38 deletions(-)
drivers/nvme/target/core.c +11 -11
index b3e765a..6feef3f 100644
@@ -644,7 +644,7 @@ static void nvmet_update_sq_head(struct nvmet_req *req)
 		} while (cmpxchg(&req->sq->sqhd, old_sqhd, new_sqhd) !=
 					old_sqhd);
 	}
-	req->rsp->sq_head = cpu_to_le16(req->sq->sqhd & 0x0000FFFF);
+	req->cqe->sq_head = cpu_to_le16(req->sq->sqhd & 0x0000FFFF);
 }
 
 static void nvmet_set_error(struct nvmet_req *req, u16 status)
@@ -653,7 +653,7 @@ static void nvmet_set_error(struct nvmet_req *req, u16 status)
 	struct nvme_error_slot *new_error_slot;
 	unsigned long flags;
 
-	req->rsp->status = cpu_to_le16(status << 1);
+	req->cqe->status = cpu_to_le16(status << 1);
 
 	if (!ctrl || req->error_loc == NVMET_NO_ERROR_LOC)
 		return;
@@ -673,15 +673,15 @@ static void nvmet_set_error(struct nvmet_req *req, u16 status)
 	spin_unlock_irqrestore(&ctrl->error_lock, flags);
 
 	/* set the more bit for this request */
-	req->rsp->status |= cpu_to_le16(1 << 14);
+	req->cqe->status |= cpu_to_le16(1 << 14);
 }
 
 static void __nvmet_req_complete(struct nvmet_req *req, u16 status)
 {
 	if (!req->sq->sqhd_disabled)
 		nvmet_update_sq_head(req);
-	req->rsp->sq_id = cpu_to_le16(req->sq->qid);
-	req->rsp->command_id = req->cmd->common.command_id;
+	req->cqe->sq_id = cpu_to_le16(req->sq->qid);
+	req->cqe->command_id = req->cmd->common.command_id;
 
 	if (unlikely(status))
 		nvmet_set_error(req, status);
@@ -838,8 +838,8 @@ bool nvmet_req_init(struct nvmet_req *req, struct nvmet_cq *cq,
 	req->sg = NULL;
 	req->sg_cnt = 0;
 	req->transfer_len = 0;
-	req->rsp->status = 0;
-	req->rsp->sq_head = 0;
+	req->cqe->status = 0;
+	req->cqe->sq_head = 0;
 	req->ns = NULL;
 	req->error_loc = NVMET_NO_ERROR_LOC;
 	req->error_slba = 0;
@@ -1066,7 +1066,7 @@ u16 nvmet_ctrl_find_get(const char *subsysnqn, const char *hostnqn, u16 cntlid,
 	if (!subsys) {
 		pr_warn("connect request for invalid subsystem %s!\n",
 			subsysnqn);
-		req->rsp->result.u32 = IPO_IATTR_CONNECT_DATA(subsysnqn);
+		req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(subsysnqn);
 		return NVME_SC_CONNECT_INVALID_PARAM | NVME_SC_DNR;
 	}
 
@@ -1087,7 +1087,7 @@ u16 nvmet_ctrl_find_get(const char *subsysnqn, const char *hostnqn, u16 cntlid,
 
 	pr_warn("could not find controller %d for subsys %s / host %s\n",
 		cntlid, subsysnqn, hostnqn);
-	req->rsp->result.u32 = IPO_IATTR_CONNECT_DATA(cntlid);
+	req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(cntlid);
 	status = NVME_SC_CONNECT_INVALID_PARAM | NVME_SC_DNR;
 
 out:
@@ -1185,7 +1185,7 @@ u16 nvmet_alloc_ctrl(const char *subsysnqn, const char *hostnqn,
 	if (!subsys) {
 		pr_warn("connect request for invalid subsystem %s!\n",
 			subsysnqn);
-		req->rsp->result.u32 = IPO_IATTR_CONNECT_DATA(subsysnqn);
+		req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(subsysnqn);
 		goto out;
 	}
 
@@ -1194,7 +1194,7 @@ u16 nvmet_alloc_ctrl(const char *subsysnqn, const char *hostnqn,
 	if (!nvmet_host_allowed(subsys, hostnqn)) {
 		pr_info("connect by host %s for subsystem %s not allowed\n",
 			hostnqn, subsysnqn);
-		req->rsp->result.u32 = IPO_IATTR_CONNECT_DATA(hostnqn);
+		req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(hostnqn);
 		up_read(&nvmet_config_sem);
 		status = NVME_SC_CONNECT_INVALID_HOST | NVME_SC_DNR;
 		goto out_put_subsystem;
drivers/nvme/target/fabrics-cmd.c +8 -8
index 3a76ebc..3b9f79a 100644
@@ -72,7 +72,7 @@ static void nvmet_execute_prop_get(struct nvmet_req *req)
 			offsetof(struct nvmf_property_get_command, attrib);
 	}
 
-	req->rsp->result.u64 = cpu_to_le64(val);
+	req->cqe->result.u64 = cpu_to_le64(val);
 	nvmet_req_complete(req, status);
 }
 
@@ -124,7 +124,7 @@ static u16 nvmet_install_queue(struct nvmet_ctrl *ctrl, struct nvmet_req *req)
 
 	if (c->cattr & NVME_CONNECT_DISABLE_SQFLOW) {
 		req->sq->sqhd_disabled = true;
-		req->rsp->sq_head = cpu_to_le16(0xffff);
+		req->cqe->sq_head = cpu_to_le16(0xffff);
 	}
 
 	if (ctrl->ops->install_queue) {
@@ -158,7 +158,7 @@ static void nvmet_execute_admin_connect(struct nvmet_req *req)
 		goto out;
 
 	/* zero out initial completion result, assign values as needed */
-	req->rsp->result.u32 = 0;
+	req->cqe->result.u32 = 0;
 
 	if (c->recfmt != 0) {
 		pr_warn("invalid connect version (%d).\n",
@@ -172,7 +172,7 @@ static void nvmet_execute_admin_connect(struct nvmet_req *req)
 		pr_warn("connect attempt for invalid controller ID %#x\n",
 			d->cntlid);
 		status = NVME_SC_CONNECT_INVALID_PARAM | NVME_SC_DNR;
-		req->rsp->result.u32 = IPO_IATTR_CONNECT_DATA(cntlid);
+		req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(cntlid);
 		goto out;
 	}
 
@@ -195,7 +195,7 @@ static void nvmet_execute_admin_connect(struct nvmet_req *req)
 
 	pr_info("creating controller %d for subsystem %s for NQN %s.\n",
 		ctrl->cntlid, ctrl->subsys->subsysnqn, ctrl->hostnqn);
-	req->rsp->result.u16 = cpu_to_le16(ctrl->cntlid);
+	req->cqe->result.u16 = cpu_to_le16(ctrl->cntlid);
 
 out:
 	kfree(d);
@@ -222,7 +222,7 @@ static void nvmet_execute_io_connect(struct nvmet_req *req)
 		goto out;
 
 	/* zero out initial completion result, assign values as needed */
-	req->rsp->result.u32 = 0;
+	req->cqe->result.u32 = 0;
 
 	if (c->recfmt != 0) {
 		pr_warn("invalid connect version (%d).\n",
@@ -240,14 +240,14 @@ static void nvmet_execute_io_connect(struct nvmet_req *req)
 	if (unlikely(qid > ctrl->subsys->max_qid)) {
 		pr_warn("invalid queue id (%d)\n", qid);
 		status = NVME_SC_CONNECT_INVALID_PARAM | NVME_SC_DNR;
-		req->rsp->result.u32 = IPO_IATTR_CONNECT_SQE(qid);
+		req->cqe->result.u32 = IPO_IATTR_CONNECT_SQE(qid);
 		goto out_ctrl_put;
 	}
 
 	status = nvmet_install_queue(ctrl, req);
 	if (status) {
 		/* pass back cntlid that had the issue of installing queue */
-		req->rsp->result.u16 = cpu_to_le16(ctrl->cntlid);
+		req->cqe->result.u16 = cpu_to_le16(ctrl->cntlid);
 		goto out_ctrl_put;
 	}
 
drivers/nvme/target/fc.c +1 -1
index 98b7b1f..6733673 100644
@@ -2187,7 +2187,7 @@ enum {
 	}
 
 	fod->req.cmd = &fod->cmdiubuf.sqe;
-	fod->req.rsp = &fod->rspiubuf.cqe;
+	fod->req.cqe = &fod->rspiubuf.cqe;
 	fod->req.port = tgtport->pe->port;
 
 	/* clear any response payload */
drivers/nvme/target/loop.c +3 -3
index b9f623a..a3ae491 100644
@@ -18,7 +18,7 @@
 struct nvme_loop_iod {
 	struct nvme_request	nvme_req;
 	struct nvme_command	cmd;
-	struct nvme_completion	rsp;
+	struct nvme_completion	cqe;
 	struct nvmet_req	req;
 	struct nvme_loop_queue	*queue;
 	struct work_struct	work;
@@ -94,7 +94,7 @@ static void nvme_loop_queue_response(struct nvmet_req *req)
 {
 	struct nvme_loop_queue *queue =
 		container_of(req->sq, struct nvme_loop_queue, nvme_sq);
-	struct nvme_completion *cqe = req->rsp;
+	struct nvme_completion *cqe = req->cqe;
 
 	/*
 	 * AEN requests are special as they don't time out and can
@@ -207,7 +207,7 @@ static int nvme_loop_init_iod(struct nvme_loop_ctrl *ctrl,
 		struct nvme_loop_iod *iod, unsigned int queue_idx)
 {
 	iod->req.cmd = &iod->cmd;
-	iod->req.rsp = &iod->rsp;
+	iod->req.cqe = &iod->cqe;
 	iod->queue = &ctrl->queues[queue_idx];
 	INIT_WORK(&iod->work, nvme_loop_execute_work);
 	return 0;
drivers/nvme/target/nvmet.h +2 -2
index 51e49ef..53d5a44 100644
@@ -284,7 +284,7 @@ struct nvmet_fabrics_ops {
 
 struct nvmet_req {
 	struct nvme_command	*cmd;
-	struct nvme_completion	*rsp;
+	struct nvme_completion	*cqe;
 	struct nvmet_sq		*sq;
 	struct nvmet_cq		*cq;
 	struct nvmet_ns		*ns;
@@ -322,7 +322,7 @@ struct nvmet_req {
 
 static inline void nvmet_set_result(struct nvmet_req *req, u32 result)
 {
-	req->rsp->result.u32 = cpu_to_le32(result);
+	req->cqe->result.u32 = cpu_to_le32(result);
 }
 
 /*
drivers/nvme/target/rdma.c +9 -9
index b727521..36d906a 100644
@@ -160,7 +160,7 @@ static inline bool nvmet_rdma_need_data_out(struct nvmet_rdma_rsp *rsp)
 {
 	return !nvme_is_write(rsp->req.cmd) &&
 		rsp->req.transfer_len &&
-		!rsp->req.rsp->status &&
+		!rsp->req.cqe->status &&
 		!(rsp->flags & NVMET_RDMA_REQ_INLINE_DATA);
 }
 
@@ -364,17 +364,17 @@ static int nvmet_rdma_alloc_rsp(struct nvmet_rdma_device *ndev,
 		struct nvmet_rdma_rsp *r)
 {
 	/* NVMe CQE / RDMA SEND */
-	r->req.rsp = kmalloc(sizeof(*r->req.rsp), GFP_KERNEL);
-	if (!r->req.rsp)
+	r->req.cqe = kmalloc(sizeof(*r->req.cqe), GFP_KERNEL);
+	if (!r->req.cqe)
 		goto out;
 
-	r->send_sge.addr = ib_dma_map_single(ndev->device, r->req.rsp,
-			sizeof(*r->req.rsp), DMA_TO_DEVICE);
+	r->send_sge.addr = ib_dma_map_single(ndev->device, r->req.cqe,
+			sizeof(*r->req.cqe), DMA_TO_DEVICE);
 	if (ib_dma_mapping_error(ndev->device, r->send_sge.addr))
 		goto out_free_rsp;
 
 	r->req.p2p_client = &ndev->device->dev;
-	r->send_sge.length = sizeof(*r->req.rsp);
+	r->send_sge.length = sizeof(*r->req.cqe);
 	r->send_sge.lkey = ndev->pd->local_dma_lkey;
 
 	r->send_cqe.done = nvmet_rdma_send_done;
@@ -389,7 +389,7 @@ static int nvmet_rdma_alloc_rsp(struct nvmet_rdma_device *ndev,
 	return 0;
 
 out_free_rsp:
-	kfree(r->req.rsp);
+	kfree(r->req.cqe);
 out:
 	return -ENOMEM;
 }
@@ -398,8 +398,8 @@ static void nvmet_rdma_free_rsp(struct nvmet_rdma_device *ndev,
 		struct nvmet_rdma_rsp *r)
 {
 	ib_dma_unmap_single(ndev->device, r->send_sge.addr,
-				sizeof(*r->req.rsp), DMA_TO_DEVICE);
-	kfree(r->req.rsp);
+				sizeof(*r->req.cqe), DMA_TO_DEVICE);
+	kfree(r->req.cqe);
 }
 
 static int
drivers/nvme/target/tcp.c +4 -4
index ad0df78..0d7df13 100644
@@ -161,14 +161,14 @@ static inline bool nvmet_tcp_has_data_in(struct nvmet_tcp_cmd *cmd)
 
 static inline bool nvmet_tcp_need_data_in(struct nvmet_tcp_cmd *cmd)
 {
-	return nvmet_tcp_has_data_in(cmd) && !cmd->req.rsp->status;
+	return nvmet_tcp_has_data_in(cmd) && !cmd->req.cqe->status;
 }
 
 static inline bool nvmet_tcp_need_data_out(struct nvmet_tcp_cmd *cmd)
 {
 	return !nvme_is_write(cmd->req.cmd) &&
 		cmd->req.transfer_len > 0 &&
-		!cmd->req.rsp->status;
+		!cmd->req.cqe->status;
 }
 
 static inline bool nvmet_tcp_has_inline_data(struct nvmet_tcp_cmd *cmd)
@@ -377,7 +377,7 @@ static void nvmet_setup_c2h_data_pdu(struct nvmet_tcp_cmd *cmd)
 	pdu->hdr.plen =
 		cpu_to_le32(pdu->hdr.hlen + hdgst +
 				cmd->req.transfer_len + ddgst);
-	pdu->command_id = cmd->req.rsp->command_id;
+	pdu->command_id = cmd->req.cqe->command_id;
 	pdu->data_length = cpu_to_le32(cmd->req.transfer_len);
 	pdu->data_offset = cpu_to_le32(cmd->wbytes_done);
 
@@ -1206,7 +1206,7 @@ static int nvmet_tcp_alloc_cmd(struct nvmet_tcp_queue *queue,
 			sizeof(*c->rsp_pdu) + hdgst, GFP_KERNEL | __GFP_ZERO);
 	if (!c->rsp_pdu)
 		goto out_free_cmd;
-	c->req.rsp = &c->rsp_pdu->cqe;
+	c->req.cqe = &c->rsp_pdu->cqe;
 
 	c->data_pdu = page_frag_alloc(&queue->pf_cache,
 			sizeof(*c->data_pdu) + hdgst, GFP_KERNEL | __GFP_ZERO);
Signature · 1 line
1.8.3.1
#2 lore raw

PATCH 2/2 nvmet: rename nvme_completion instances from rsp to cqe

Looks good.

Reviewed-by Chaitanya Kulkarni <chaitanya.kulkarni at wdc.com>

?On 4/8/19, 8:41 AM, "Linux-nvme on behalf of Max Gurtovoy" <[email protected] on behalf of [email protected]> wrote:

    Use NVMe namings for improving code readability.
    
    Signed-off-by: Max Gurtovoy <maxg at mellanox.com>
    ---
     drivers/nvme/target/core.c        | 22 +++++++++++-----------
     drivers/nvme/target/fabrics-cmd.c | 16 ++++++++--------
     drivers/nvme/target/fc.c          |  2 +-
     drivers/nvme/target/loop.c        |  6 +++---
     drivers/nvme/target/nvmet.h       |  4 ++--
     drivers/nvme/target/rdma.c        | 18 +++++++++---------
     drivers/nvme/target/tcp.c         |  8 ++++----
     7 files changed, 38 insertions(+), 38 deletions(-)
    
    diff --git a/drivers/nvme/target/core.c b/drivers/nvme/target/core.c
    index b3e765a..6feef3f 100644
    --- a/drivers/nvme/target/core.c
    +++ b/drivers/nvme/target/core.c
    @@ -644,7 +644,7 @@ static void nvmet_update_sq_head(struct nvmet_req *req)
     		} while (cmpxchg(&req->sq->sqhd, old_sqhd, new_sqhd) !=
     					old_sqhd);
     	}
    -	req->rsp->sq_head = cpu_to_le16(req->sq->sqhd & 0x0000FFFF);
    +	req->cqe->sq_head = cpu_to_le16(req->sq->sqhd & 0x0000FFFF);
     }
     
     static void nvmet_set_error(struct nvmet_req *req, u16 status)
    @@ -653,7 +653,7 @@ static void nvmet_set_error(struct nvmet_req *req, u16 status)
     	struct nvme_error_slot *new_error_slot;
     	unsigned long flags;
     
    -	req->rsp->status = cpu_to_le16(status << 1);
    +	req->cqe->status = cpu_to_le16(status << 1);
     
     	if (!ctrl || req->error_loc == NVMET_NO_ERROR_LOC)
     		return;
    @@ -673,15 +673,15 @@ static void nvmet_set_error(struct nvmet_req *req, u16 status)
     	spin_unlock_irqrestore(&ctrl->error_lock, flags);
     
     	/* set the more bit for this request */
    -	req->rsp->status |= cpu_to_le16(1 << 14);
    +	req->cqe->status |= cpu_to_le16(1 << 14);
     }
     
     static void __nvmet_req_complete(struct nvmet_req *req, u16 status)
     {
     	if (!req->sq->sqhd_disabled)
     		nvmet_update_sq_head(req);
    -	req->rsp->sq_id = cpu_to_le16(req->sq->qid);
    -	req->rsp->command_id = req->cmd->common.command_id;
    +	req->cqe->sq_id = cpu_to_le16(req->sq->qid);
    +	req->cqe->command_id = req->cmd->common.command_id;
     
     	if (unlikely(status))
     		nvmet_set_error(req, status);
    @@ -838,8 +838,8 @@ bool nvmet_req_init(struct nvmet_req *req, struct nvmet_cq *cq,
     	req->sg = NULL;
     	req->sg_cnt = 0;
     	req->transfer_len = 0;
    -	req->rsp->status = 0;
    -	req->rsp->sq_head = 0;
    +	req->cqe->status = 0;
    +	req->cqe->sq_head = 0;
     	req->ns = NULL;
     	req->error_loc = NVMET_NO_ERROR_LOC;
     	req->error_slba = 0;
    @@ -1066,7 +1066,7 @@ u16 nvmet_ctrl_find_get(const char *subsysnqn, const char *hostnqn, u16 cntlid,
     	if (!subsys) {
     		pr_warn("connect request for invalid subsystem %s!\n",
     			subsysnqn);
    -		req->rsp->result.u32 = IPO_IATTR_CONNECT_DATA(subsysnqn);
    +		req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(subsysnqn);
     		return NVME_SC_CONNECT_INVALID_PARAM | NVME_SC_DNR;
     	}
     
    @@ -1087,7 +1087,7 @@ u16 nvmet_ctrl_find_get(const char *subsysnqn, const char *hostnqn, u16 cntlid,
     
     	pr_warn("could not find controller %d for subsys %s / host %s\n",
     		cntlid, subsysnqn, hostnqn);
    -	req->rsp->result.u32 = IPO_IATTR_CONNECT_DATA(cntlid);
    +	req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(cntlid);
     	status = NVME_SC_CONNECT_INVALID_PARAM | NVME_SC_DNR;
     
     out:
    @@ -1185,7 +1185,7 @@ u16 nvmet_alloc_ctrl(const char *subsysnqn, const char *hostnqn,
     	if (!subsys) {
     		pr_warn("connect request for invalid subsystem %s!\n",
     			subsysnqn);
    -		req->rsp->result.u32 = IPO_IATTR_CONNECT_DATA(subsysnqn);
    +		req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(subsysnqn);
     		goto out;
     	}
     
    @@ -1194,7 +1194,7 @@ u16 nvmet_alloc_ctrl(const char *subsysnqn, const char *hostnqn,
     	if (!nvmet_host_allowed(subsys, hostnqn)) {
     		pr_info("connect by host %s for subsystem %s not allowed\n",
     			hostnqn, subsysnqn);
    -		req->rsp->result.u32 = IPO_IATTR_CONNECT_DATA(hostnqn);
    +		req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(hostnqn);
     		up_read(&nvmet_config_sem);
     		status = NVME_SC_CONNECT_INVALID_HOST | NVME_SC_DNR;
     		goto out_put_subsystem;
    diff --git a/drivers/nvme/target/fabrics-cmd.c b/drivers/nvme/target/fabrics-cmd.c
    index 3a76ebc..3b9f79a 100644
    --- a/drivers/nvme/target/fabrics-cmd.c
    +++ b/drivers/nvme/target/fabrics-cmd.c
    @@ -72,7 +72,7 @@ static void nvmet_execute_prop_get(struct nvmet_req *req)
     			offsetof(struct nvmf_property_get_command, attrib);
     	}
     
    -	req->rsp->result.u64 = cpu_to_le64(val);
    +	req->cqe->result.u64 = cpu_to_le64(val);
     	nvmet_req_complete(req, status);
     }
     
    @@ -124,7 +124,7 @@ static u16 nvmet_install_queue(struct nvmet_ctrl *ctrl, struct nvmet_req *req)
     
     	if (c->cattr & NVME_CONNECT_DISABLE_SQFLOW) {
     		req->sq->sqhd_disabled = true;
    -		req->rsp->sq_head = cpu_to_le16(0xffff);
    +		req->cqe->sq_head = cpu_to_le16(0xffff);
     	}
     
     	if (ctrl->ops->install_queue) {
    @@ -158,7 +158,7 @@ static void nvmet_execute_admin_connect(struct nvmet_req *req)
     		goto out;
     
     	/* zero out initial completion result, assign values as needed */
    -	req->rsp->result.u32 = 0;
    +	req->cqe->result.u32 = 0;
     
     	if (c->recfmt != 0) {
     		pr_warn("invalid connect version (%d).\n",
    @@ -172,7 +172,7 @@ static void nvmet_execute_admin_connect(struct nvmet_req *req)
     		pr_warn("connect attempt for invalid controller ID %#x\n",
     			d->cntlid);
     		status = NVME_SC_CONNECT_INVALID_PARAM | NVME_SC_DNR;
    -		req->rsp->result.u32 = IPO_IATTR_CONNECT_DATA(cntlid);
    +		req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(cntlid);
     		goto out;
     	}
     
    @@ -195,7 +195,7 @@ static void nvmet_execute_admin_connect(struct nvmet_req *req)
     
     	pr_info("creating controller %d for subsystem %s for NQN %s.\n",
     		ctrl->cntlid, ctrl->subsys->subsysnqn, ctrl->hostnqn);
    -	req->rsp->result.u16 = cpu_to_le16(ctrl->cntlid);
    +	req->cqe->result.u16 = cpu_to_le16(ctrl->cntlid);
     
     out:
     	kfree(d);
    @@ -222,7 +222,7 @@ static void nvmet_execute_io_connect(struct nvmet_req *req)
     		goto out;
     
     	/* zero out initial completion result, assign values as needed */
    -	req->rsp->result.u32 = 0;
    +	req->cqe->result.u32 = 0;
     
     	if (c->recfmt != 0) {
     		pr_warn("invalid connect version (%d).\n",
    @@ -240,14 +240,14 @@ static void nvmet_execute_io_connect(struct nvmet_req *req)
     	if (unlikely(qid > ctrl->subsys->max_qid)) {
     		pr_warn("invalid queue id (%d)\n", qid);
     		status = NVME_SC_CONNECT_INVALID_PARAM | NVME_SC_DNR;
    -		req->rsp->result.u32 = IPO_IATTR_CONNECT_SQE(qid);
    +		req->cqe->result.u32 = IPO_IATTR_CONNECT_SQE(qid);
     		goto out_ctrl_put;
     	}
     
     	status = nvmet_install_queue(ctrl, req);
     	if (status) {
     		/* pass back cntlid that had the issue of installing queue */
    -		req->rsp->result.u16 = cpu_to_le16(ctrl->cntlid);
    +		req->cqe->result.u16 = cpu_to_le16(ctrl->cntlid);
     		goto out_ctrl_put;
     	}
     
    diff --git a/drivers/nvme/target/fc.c b/drivers/nvme/target/fc.c
    index 98b7b1f..6733673 100644
    --- a/drivers/nvme/target/fc.c
    +++ b/drivers/nvme/target/fc.c
    @@ -2187,7 +2187,7 @@ enum {
     	}
     
     	fod->req.cmd = &fod->cmdiubuf.sqe;
    -	fod->req.rsp = &fod->rspiubuf.cqe;
    +	fod->req.cqe = &fod->rspiubuf.cqe;
     	fod->req.port = tgtport->pe->port;
     
     	/* clear any response payload */
    diff --git a/drivers/nvme/target/loop.c b/drivers/nvme/target/loop.c
    index b9f623a..a3ae491 100644
    --- a/drivers/nvme/target/loop.c
    +++ b/drivers/nvme/target/loop.c
    @@ -18,7 +18,7 @@
     struct nvme_loop_iod {
     	struct nvme_request	nvme_req;
     	struct nvme_command	cmd;
    -	struct nvme_completion	rsp;
    +	struct nvme_completion	cqe;
     	struct nvmet_req	req;
     	struct nvme_loop_queue	*queue;
     	struct work_struct	work;
    @@ -94,7 +94,7 @@ static void nvme_loop_queue_response(struct nvmet_req *req)
     {
     	struct nvme_loop_queue *queue =
     		container_of(req->sq, struct nvme_loop_queue, nvme_sq);
    -	struct nvme_completion *cqe = req->rsp;
    +	struct nvme_completion *cqe = req->cqe;
     
     	/*
     	 * AEN requests are special as they don't time out and can
    @@ -207,7 +207,7 @@ static int nvme_loop_init_iod(struct nvme_loop_ctrl *ctrl,
     		struct nvme_loop_iod *iod, unsigned int queue_idx)
     {
     	iod->req.cmd = &iod->cmd;
    -	iod->req.rsp = &iod->rsp;
    +	iod->req.cqe = &iod->cqe;
     	iod->queue = &ctrl->queues[queue_idx];
     	INIT_WORK(&iod->work, nvme_loop_execute_work);
     	return 0;
    diff --git a/drivers/nvme/target/nvmet.h b/drivers/nvme/target/nvmet.h
    index 51e49ef..53d5a44 100644
    --- a/drivers/nvme/target/nvmet.h
    +++ b/drivers/nvme/target/nvmet.h
    @@ -284,7 +284,7 @@ struct nvmet_fabrics_ops {
     
     struct nvmet_req {
     	struct nvme_command	*cmd;
    -	struct nvme_completion	*rsp;
    +	struct nvme_completion	*cqe;
     	struct nvmet_sq		*sq;
     	struct nvmet_cq		*cq;
     	struct nvmet_ns		*ns;
    @@ -322,7 +322,7 @@ struct nvmet_req {
     
     static inline void nvmet_set_result(struct nvmet_req *req, u32 result)
     {
    -	req->rsp->result.u32 = cpu_to_le32(result);
    +	req->cqe->result.u32 = cpu_to_le32(result);
     }
     
     /*
    diff --git a/drivers/nvme/target/rdma.c b/drivers/nvme/target/rdma.c
    index b727521..36d906a 100644
    --- a/drivers/nvme/target/rdma.c
    +++ b/drivers/nvme/target/rdma.c
    @@ -160,7 +160,7 @@ static inline bool nvmet_rdma_need_data_out(struct nvmet_rdma_rsp *rsp)
     {
     	return !nvme_is_write(rsp->req.cmd) &&
     		rsp->req.transfer_len &&
    -		!rsp->req.rsp->status &&
    +		!rsp->req.cqe->status &&
     		!(rsp->flags & NVMET_RDMA_REQ_INLINE_DATA);
     }
     
    @@ -364,17 +364,17 @@ static int nvmet_rdma_alloc_rsp(struct nvmet_rdma_device *ndev,
     		struct nvmet_rdma_rsp *r)
     {
     	/* NVMe CQE / RDMA SEND */
    -	r->req.rsp = kmalloc(sizeof(*r->req.rsp), GFP_KERNEL);
    -	if (!r->req.rsp)
    +	r->req.cqe = kmalloc(sizeof(*r->req.cqe), GFP_KERNEL);
    +	if (!r->req.cqe)
     		goto out;
     
    -	r->send_sge.addr = ib_dma_map_single(ndev->device, r->req.rsp,
    -			sizeof(*r->req.rsp), DMA_TO_DEVICE);
    +	r->send_sge.addr = ib_dma_map_single(ndev->device, r->req.cqe,
    +			sizeof(*r->req.cqe), DMA_TO_DEVICE);
     	if (ib_dma_mapping_error(ndev->device, r->send_sge.addr))
     		goto out_free_rsp;
     
     	r->req.p2p_client = &ndev->device->dev;
    -	r->send_sge.length = sizeof(*r->req.rsp);
    +	r->send_sge.length = sizeof(*r->req.cqe);
     	r->send_sge.lkey = ndev->pd->local_dma_lkey;
     
     	r->send_cqe.done = nvmet_rdma_send_done;
    @@ -389,7 +389,7 @@ static int nvmet_rdma_alloc_rsp(struct nvmet_rdma_device *ndev,
     	return 0;
     
     out_free_rsp:
    -	kfree(r->req.rsp);
    +	kfree(r->req.cqe);
     out:
     	return -ENOMEM;
     }
    @@ -398,8 +398,8 @@ static void nvmet_rdma_free_rsp(struct nvmet_rdma_device *ndev,
     		struct nvmet_rdma_rsp *r)
     {
     	ib_dma_unmap_single(ndev->device, r->send_sge.addr,
    -				sizeof(*r->req.rsp), DMA_TO_DEVICE);
    -	kfree(r->req.rsp);
    +				sizeof(*r->req.cqe), DMA_TO_DEVICE);
    +	kfree(r->req.cqe);
     }
     
     static int
    diff --git a/drivers/nvme/target/tcp.c b/drivers/nvme/target/tcp.c
    index ad0df78..0d7df13 100644
    --- a/drivers/nvme/target/tcp.c
    +++ b/drivers/nvme/target/tcp.c
    @@ -161,14 +161,14 @@ static inline bool nvmet_tcp_has_data_in(struct nvmet_tcp_cmd *cmd)
     
     static inline bool nvmet_tcp_need_data_in(struct nvmet_tcp_cmd *cmd)
     {
    -	return nvmet_tcp_has_data_in(cmd) && !cmd->req.rsp->status;
    +	return nvmet_tcp_has_data_in(cmd) && !cmd->req.cqe->status;
     }
     
     static inline bool nvmet_tcp_need_data_out(struct nvmet_tcp_cmd *cmd)
     {
     	return !nvme_is_write(cmd->req.cmd) &&
     		cmd->req.transfer_len > 0 &&
    -		!cmd->req.rsp->status;
    +		!cmd->req.cqe->status;
     }
     
     static inline bool nvmet_tcp_has_inline_data(struct nvmet_tcp_cmd *cmd)
    @@ -377,7 +377,7 @@ static void nvmet_setup_c2h_data_pdu(struct nvmet_tcp_cmd *cmd)
     	pdu->hdr.plen =
     		cpu_to_le32(pdu->hdr.hlen + hdgst +
     				cmd->req.transfer_len + ddgst);
    -	pdu->command_id = cmd->req.rsp->command_id;
    +	pdu->command_id = cmd->req.cqe->command_id;
     	pdu->data_length = cpu_to_le32(cmd->req.transfer_len);
     	pdu->data_offset = cpu_to_le32(cmd->wbytes_done);
     
    @@ -1206,7 +1206,7 @@ static int nvmet_tcp_alloc_cmd(struct nvmet_tcp_queue *queue,
     			sizeof(*c->rsp_pdu) + hdgst, GFP_KERNEL | __GFP_ZERO);
     	if (!c->rsp_pdu)
     		goto out_free_cmd;
    -	c->req.rsp = &c->rsp_pdu->cqe;
    +	c->req.cqe = &c->rsp_pdu->cqe;
     
     	c->data_pdu = page_frag_alloc(&queue->pf_cache,
     			sizeof(*c->data_pdu) + hdgst, GFP_KERNEL | __GFP_ZERO);
    -- 
    1.8.3.1
    
    
    _______________________________________________
    Linux-nvme mailing list
    Linux-nvme at lists.infradead.org
    http://lists.infradead.org/mailman/listinfo/linux-nvme
#4 lore raw

PATCH 1/2 nvmet-rdma: remove p2p_client initialization from fast-path

In reply to Max Gurtovoy
Quoted 1 line · Max Gurtovoy

Initialize it during command allocation.

It would be nice to hear some justification for this change in the commit message, but the change itself appears to be fine to me:

Reviewed-by Logan Gunthorpe <logang at deltatee.com>
Quoted 29 lines
Cc Logan Gunthorpe <logang at deltatee.com>
Cc Stephen Bates <sbates at raithlin.com>
Signed-off-by Max Gurtovoy <maxg at mellanox.com>
---
 drivers/nvme/target/rdma.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)
drivers/nvme/target/rdma.c +1 -2
index ef893ad..b727521 100644
@@ -373,6 +373,7 @@ static int nvmet_rdma_alloc_rsp(struct nvmet_rdma_device *ndev,
 	if (ib_dma_mapping_error(ndev->device, r->send_sge.addr))
 		goto out_free_rsp;
 
+	r->req.p2p_client = &ndev->device->dev;
 	r->send_sge.length = sizeof(*r->req.rsp);
 	r->send_sge.lkey = ndev->pd->local_dma_lkey;
 
@@ -763,8 +764,6 @@ static void nvmet_rdma_handle_command(struct nvmet_rdma_queue *queue,
 		cmd->send_sge.addr, cmd->send_sge.length,
 		DMA_TO_DEVICE);
 
-	cmd->req.p2p_client = &queue->dev->device->dev;
-
 	if (!nvmet_req_init(&cmd->req, &queue->nvme_cq,
 			&queue->nvme_sq, &nvmet_rdma_ops))
 		return;
#5 lore raw

PATCH 1/2 nvmet-rdma: remove p2p_client initialization from fast-path

Quoted 6 lines · Logan Gunthorpe
Quoted 1 line · Max Gurtovoy

Initialize it during command allocation.

It would be nice to hear some justification for this change in the commit message, but the change itself appears to be fine to me:

well, it's written in the subject and it's pretty small change so there is not much to say about it.

Just set this ptr once instead of every IO.

Quoted 32 lines
Reviewed-by Logan Gunthorpe <logang at deltatee.com>
Quoted 29 lines
Cc Logan Gunthorpe <logang at deltatee.com>
Cc Stephen Bates <sbates at raithlin.com>
Signed-off-by Max Gurtovoy <maxg at mellanox.com>
---
  drivers/nvme/target/rdma.c | 3 +--
  1 file changed, 1 insertion(+), 2 deletions(-)
drivers/nvme/target/rdma.c +1 -2
index ef893ad..b727521 100644
@@ -373,6 +373,7 @@ static int nvmet_rdma_alloc_rsp(struct nvmet_rdma_device *ndev,
  	if (ib_dma_mapping_error(ndev->device, r->send_sge.addr))
  		goto out_free_rsp;
 
+	r->req.p2p_client = &ndev->device->dev;
  	r->send_sge.length = sizeof(*r->req.rsp);
  	r->send_sge.lkey = ndev->pd->local_dma_lkey;
 
@@ -763,8 +764,6 @@ static void nvmet_rdma_handle_command(struct nvmet_rdma_queue *queue,
  		cmd->send_sge.addr, cmd->send_sge.length,
  		DMA_TO_DEVICE);
 
-	cmd->req.p2p_client = &queue->dev->device->dev;
-
  	if (!nvmet_req_init(&cmd->req, &queue->nvme_cq,
  			&queue->nvme_sq, &nvmet_rdma_ops))
  		return;
#6 lore raw

PATCH 1/2 nvmet-rdma: remove p2p_client initialization from fast-path

Quoted 6 lines · Max Gurtovoy
Quoted 4 lines · Logan Gunthorpe
Quoted 1 line · Max Gurtovoy

Initialize it during command allocation.

Quoted 7 lines
Quoted 2 lines

It would be nice to hear some justification for this change in the commit message, but the change itself appears to be fine to me:

well, it's written in the subject and it's pretty small change so there is not much to say about it.

Just set this ptr once instead of every IO.

But is this just an optimization-without-measuring from code review? Or did I actually cause a measurable performance regression? What actually prompted the change?

Just to be clear: I'm fine with the change and am happy it's being done; I'd just prefer to see more detail in the commit message.

Thanks,

Logan

#7 lore raw

PATCH 1/2 nvmet-rdma: remove p2p_client initialization from fast-path

Quoted 7 lines · Logan Gunthorpe
Quoted 5 lines · Max Gurtovoy
Quoted 4 lines · Logan Gunthorpe
Quoted 1 line · Max Gurtovoy

Initialize it during command allocation.

It would be nice to hear some justification for this change in the commit message, but the change itself appears to be fine to me:

Quoted 7 lines
Quoted 4 lines

well, it's written in the subject and it's pretty small change so there is not much to say about it.

Just set this ptr once instead of every IO.

But is this just an optimization-without-measuring from code review? Or did I actually cause a measurable performance regression? What actually prompted the change?

The first one :)

I don't think that one assignment can cause a regression but I also think the we shouldn't do redundant operations on the fast path.

These can accumulate in the future..

Quoted 3 lines

Just to be clear: I'm fine with the change and am happy it's being done; I'd just prefer to see more detail in the commit message.

I guess it's possible.

Lets get a review for the second patch and I'll send V2.

Quoted 4 lines

Thanks,

Logan