nvme_fcloop: rework to remove xxx_IN_ISR feature flags
authorJames Smart <jsmart2021@gmail.com>
Thu, 30 Nov 2017 00:47:32 +0000 (16:47 -0800)
committerChristoph Hellwig <hch@lst.de>
Mon, 8 Jan 2018 10:01:54 +0000 (11:01 +0100)
The existing fcloop driver expects the target side upcalls to
the transport to context switch, thus the calls into the nvmet layer
are not done in the calling context of the host/initiator down calls.
The xxx_IN_ISR feature flags are used to select this logic.

The xxx_IN_ISR feature flags should go away in the nvmet_fc transport
as no other lldd utilizes them. Both Broadcom and Cavium lldds have their
own non-ISR deferred handlers thus the nvmet calls can be made directly.

This patch converts the paths that make the target upcalls (command
receive, abort receive) such that they schedule a work item rather
than expecting the transport to schedule the work item.

The patch also cleans up the following:
- The completion path from target to host scheduled a host work
  element called "work". Rename it "tio_done_work" for code clarity.
- The abort io path called a iniwork item to call the host side
  io done. This is no longer needed as the abort routine can make
  the same call.

Signed-off-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
drivers/nvme/target/fcloop.c

index c0080f6ab2f5b209eb8aeae003df6659f9124f11..c5015199c03144ac99b09719b4641c71a5b7f9b1 100644 (file)
@@ -249,14 +249,15 @@ struct fcloop_fcpreq {
        u16                             status;
        bool                            active;
        bool                            aborted;
-       struct work_struct              work;
+       struct work_struct              fcp_rcv_work;
+       struct work_struct              abort_rcv_work;
+       struct work_struct              tio_done_work;
        struct nvmefc_tgt_fcp_req       tgt_fcp_req;
 };
 
 struct fcloop_ini_fcpreq {
        struct nvmefc_fcp_req           *fcpreq;
        struct fcloop_fcpreq            *tfcp_req;
-       struct work_struct              iniwork;
 };
 
 static inline struct fcloop_lsreq *
@@ -347,17 +348,58 @@ fcloop_xmt_ls_rsp(struct nvmet_fc_target_port *tport,
        return 0;
 }
 
-/*
- * FCP IO operation done by initiator abort.
- * call back up initiator "done" flows.
- */
 static void
-fcloop_tgt_fcprqst_ini_done_work(struct work_struct *work)
+fcloop_fcp_recv_work(struct work_struct *work)
+{
+       struct fcloop_fcpreq *tfcp_req =
+               container_of(work, struct fcloop_fcpreq, fcp_rcv_work);
+       struct nvmefc_fcp_req *fcpreq = tfcp_req->fcpreq;
+       struct fcloop_ini_fcpreq *inireq = NULL;
+       int ret = 0;
+
+       ret = nvmet_fc_rcv_fcp_req(tfcp_req->tport->targetport,
+                               &tfcp_req->tgt_fcp_req,
+                               fcpreq->cmdaddr, fcpreq->cmdlen);
+       if (ret) {
+               inireq = fcpreq->private;
+               inireq->tfcp_req = NULL;
+
+               fcpreq->status = tfcp_req->status;
+               fcpreq->done(fcpreq);
+       }
+}
+
+static void
+fcloop_call_host_done(struct nvmefc_fcp_req *fcpreq,
+                       struct fcloop_fcpreq *tfcp_req, int status)
+{
+       struct fcloop_ini_fcpreq *inireq = NULL;
+
+       if (fcpreq) {
+               inireq = fcpreq->private;
+               inireq->tfcp_req = NULL;
+
+               fcpreq->status = status;
+               fcpreq->done(fcpreq);
+       }
+}
+
+static void
+fcloop_fcp_abort_recv_work(struct work_struct *work)
 {
-       struct fcloop_ini_fcpreq *inireq =
-               container_of(work, struct fcloop_ini_fcpreq, iniwork);
+       struct fcloop_fcpreq *tfcp_req =
+               container_of(work, struct fcloop_fcpreq, abort_rcv_work);
+       struct nvmefc_fcp_req *fcpreq = tfcp_req->fcpreq;
+
+       if (tfcp_req->tport->targetport)
+               nvmet_fc_rcv_fcp_abort(tfcp_req->tport->targetport,
+                                       &tfcp_req->tgt_fcp_req);
+
+       spin_lock(&tfcp_req->reqlock);
+       tfcp_req->fcpreq = NULL;
+       spin_unlock(&tfcp_req->reqlock);
 
-       inireq->fcpreq->done(inireq->fcpreq);
+       fcloop_call_host_done(fcpreq, tfcp_req, -ECANCELED);
 }
 
 /*
@@ -368,8 +410,7 @@ static void
 fcloop_tgt_fcprqst_done_work(struct work_struct *work)
 {
        struct fcloop_fcpreq *tfcp_req =
-               container_of(work, struct fcloop_fcpreq, work);
-       struct fcloop_tport *tport = tfcp_req->tport;
+               container_of(work, struct fcloop_fcpreq, tio_done_work);
        struct nvmefc_fcp_req *fcpreq;
 
        spin_lock(&tfcp_req->reqlock);
@@ -377,10 +418,7 @@ fcloop_tgt_fcprqst_done_work(struct work_struct *work)
        tfcp_req->fcpreq = NULL;
        spin_unlock(&tfcp_req->reqlock);
 
-       if (tport->remoteport && fcpreq) {
-               fcpreq->status = tfcp_req->status;
-               fcpreq->done(fcpreq);
-       }
+       fcloop_call_host_done(fcpreq, tfcp_req, tfcp_req->status);
 
        kfree(tfcp_req);
 }
@@ -395,7 +433,6 @@ fcloop_fcp_req(struct nvme_fc_local_port *localport,
        struct fcloop_rport *rport = remoteport->private;
        struct fcloop_ini_fcpreq *inireq = fcpreq->private;
        struct fcloop_fcpreq *tfcp_req;
-       int ret = 0;
 
        if (!rport->targetport)
                return -ECONNREFUSED;
@@ -406,16 +443,16 @@ fcloop_fcp_req(struct nvme_fc_local_port *localport,
 
        inireq->fcpreq = fcpreq;
        inireq->tfcp_req = tfcp_req;
-       INIT_WORK(&inireq->iniwork, fcloop_tgt_fcprqst_ini_done_work);
        tfcp_req->fcpreq = fcpreq;
        tfcp_req->tport = rport->targetport->private;
        spin_lock_init(&tfcp_req->reqlock);
-       INIT_WORK(&tfcp_req->work, fcloop_tgt_fcprqst_done_work);
+       INIT_WORK(&tfcp_req->fcp_rcv_work, fcloop_fcp_recv_work);
+       INIT_WORK(&tfcp_req->abort_rcv_work, fcloop_fcp_abort_recv_work);
+       INIT_WORK(&tfcp_req->tio_done_work, fcloop_tgt_fcprqst_done_work);
 
-       ret = nvmet_fc_rcv_fcp_req(rport->targetport, &tfcp_req->tgt_fcp_req,
-                                fcpreq->cmdaddr, fcpreq->cmdlen);
+       schedule_work(&tfcp_req->fcp_rcv_work);
 
-       return ret;
+       return 0;
 }
 
 static void
@@ -594,7 +631,7 @@ fcloop_fcp_req_release(struct nvmet_fc_target_port *tgtport,
 {
        struct fcloop_fcpreq *tfcp_req = tgt_fcp_req_to_fcpreq(tgt_fcpreq);
 
-       schedule_work(&tfcp_req->work);
+       schedule_work(&tfcp_req->tio_done_work);
 }
 
 static void
@@ -610,13 +647,12 @@ fcloop_fcp_abort(struct nvme_fc_local_port *localport,
                        void *hw_queue_handle,
                        struct nvmefc_fcp_req *fcpreq)
 {
-       struct fcloop_rport *rport = remoteport->private;
        struct fcloop_ini_fcpreq *inireq = fcpreq->private;
        struct fcloop_fcpreq *tfcp_req = inireq->tfcp_req;
 
        if (!tfcp_req)
                /* abort has already been called */
-               goto finish;
+               return;
 
        /* break initiator/target relationship for io */
        spin_lock(&tfcp_req->reqlock);
@@ -624,14 +660,7 @@ fcloop_fcp_abort(struct nvme_fc_local_port *localport,
        tfcp_req->fcpreq = NULL;
        spin_unlock(&tfcp_req->reqlock);
 
-       if (rport->targetport)
-               nvmet_fc_rcv_fcp_abort(rport->targetport,
-                                       &tfcp_req->tgt_fcp_req);
-
-finish:
-       /* post the aborted io completion */
-       fcpreq->status = -ECANCELED;
-       schedule_work(&inireq->iniwork);
+       WARN_ON(!schedule_work(&tfcp_req->abort_rcv_work));
 }
 
 static void
@@ -721,8 +750,7 @@ static struct nvmet_fc_target_template tgttemplate = {
        .max_dif_sgl_segments   = FCLOOP_SGL_SEGS,
        .dma_boundary           = FCLOOP_DMABOUND_4G,
        /* optional features */
-       .target_features        = NVMET_FCTGTFEAT_CMD_IN_ISR |
-                                 NVMET_FCTGTFEAT_OPDONE_IN_ISR,
+       .target_features        = 0,
        /* sizes of additional private data for data structures */
        .target_priv_sz         = sizeof(struct fcloop_tport),
 };