[PATCH 00/15] IB/hfi1: hfi1 driver patches for 4.12

[PATCH 00/15] IB/hfi1: hfi1 driver patches for 4.12

[Dennis Dalessandro]

Hi Doug,

Here is one more set of patches for the hfi1 driver for the merge window. This
is just fixes and clean ups as usual. Of note there are about 5 patches here
which deal with context initiliazation and usage. 

Patches apply cleanly on top of: 87f0faa : IB/SA: Add OPA addr header

Patches can can also be found in my GitHub repo at:
https://github.com/ddalessa/kernel/tree/for-4.12

---

Jakub Byczkowski (1):
      IB/hfi1: Fix checks for Offline transient state

Michael J. Ruhl (9):
      IB/hfi1: Return an error on memory allocation failure
      IB/hfi1: Fix a subcontext memory leak
      IB/hfi1: Name function prototype parameters
      IB/hfi1: Use filedata rather than filepointer
      IB/hfi1: Search shared contexts on the opened device, not all devices
      IB/hfi1: Correctly clear the pkey
      IB/hfi1: Clean up context initialization
      IB/hfi1: Fix an assign/ordering issue with shared context IDs
      IB/hfi1: Clean up on context initialization failure

Mike Marciniszyn (2):
      IB/hfi1,IB/rdmavt: Move r_adefered to r_lock cache line
      IB/hfi1: Fix yield logic in send engine

Sebastian Sanchez (2):
      IB/hfi1: Get rid of divide when setting the tx request header
      IB/hfi1: Remove atomic operations for SDMA_REQ_HAVE_AHG bit

Tymoteusz Kielan (1):
      IB/hfi1: Adjust default eager_buffer_size to 8MB


 drivers/infiniband/hw/hfi1/chip.c         |   47 +--
 drivers/infiniband/hw/hfi1/chip.h         |   10 -
 drivers/infiniband/hw/hfi1/driver.c       |   42 ---
 drivers/infiniband/hw/hfi1/file_ops.c     |  425 +++++++++++++++--------------
 drivers/infiniband/hw/hfi1/hfi.h          |  107 +++----
 drivers/infiniband/hw/hfi1/init.c         |   33 +-
 drivers/infiniband/hw/hfi1/intr.c         |    3 
 drivers/infiniband/hw/hfi1/qp.c           |    4 
 drivers/infiniband/hw/hfi1/rc.c           |   13 -
 drivers/infiniband/hw/hfi1/ruc.c          |   80 +++--
 drivers/infiniband/hw/hfi1/trace_ctxts.h  |   17 +
 drivers/infiniband/hw/hfi1/trace_tx.h     |   34 ++
 drivers/infiniband/hw/hfi1/user_exp_rcv.c |  182 +++++++-----
 drivers/infiniband/hw/hfi1/user_exp_rcv.h |   17 +
 drivers/infiniband/hw/hfi1/user_sdma.c    |  190 ++++++-------
 drivers/infiniband/hw/hfi1/user_sdma.h    |   18 +
 drivers/infiniband/hw/hfi1/verbs.h        |    5 
 drivers/infiniband/hw/hfi1/vnic_main.c    |    8 -
 include/rdma/rdmavt_qp.h                  |    1 
 19 files changed, 633 insertions(+), 603 deletions(-)

[PATCH 00/15] IB/hfi1: hfi1 driver patches for 4.12

[Dennis Dalessandro]

Hi Doug,

Here is one more set of patches for the hfi1 driver for the merge window. This
is just fixes and clean ups as usual. Of note there are about 5 patches here
which deal with context initiliazation and usage. 

Patches apply cleanly on top of: 87f0faa : IB/SA: Add OPA addr header

Patches can can also be found in my GitHub repo at:
https://github.com/ddalessa/kernel/tree/for-4.12

---

Jakub Byczkowski (1):
      IB/hfi1: Fix checks for Offline transient state

Michael J. Ruhl (9):
      IB/hfi1: Return an error on memory allocation failure
      IB/hfi1: Fix a subcontext memory leak
      IB/hfi1: Name function prototype parameters
      IB/hfi1: Use filedata rather than filepointer
      IB/hfi1: Search shared contexts on the opened device, not all devices
      IB/hfi1: Correctly clear the pkey
      IB/hfi1: Clean up context initialization
      IB/hfi1: Fix an assign/ordering issue with shared context IDs
      IB/hfi1: Clean up on context initialization failure

Mike Marciniszyn (2):
      IB/hfi1,IB/rdmavt: Move r_adefered to r_lock cache line
      IB/hfi1: Fix yield logic in send engine

Sebastian Sanchez (2):
      IB/hfi1: Get rid of divide when setting the tx request header
      IB/hfi1: Remove atomic operations for SDMA_REQ_HAVE_AHG bit

Tymoteusz Kielan (1):
      IB/hfi1: Adjust default eager_buffer_size to 8MB


 drivers/infiniband/hw/hfi1/chip.c         |   47 +--
 drivers/infiniband/hw/hfi1/chip.h         |   10 -
 drivers/infiniband/hw/hfi1/driver.c       |   42 ---
 drivers/infiniband/hw/hfi1/file_ops.c     |  425 +++++++++++++++--------------
 drivers/infiniband/hw/hfi1/hfi.h          |  107 +++----
 drivers/infiniband/hw/hfi1/init.c         |   33 +-
 drivers/infiniband/hw/hfi1/intr.c         |    3 
 drivers/infiniband/hw/hfi1/qp.c           |    4 
 drivers/infiniband/hw/hfi1/rc.c           |   13 -
 drivers/infiniband/hw/hfi1/ruc.c          |   80 +++--
 drivers/infiniband/hw/hfi1/trace_ctxts.h  |   17 +
 drivers/infiniband/hw/hfi1/trace_tx.h     |   34 ++
 drivers/infiniband/hw/hfi1/user_exp_rcv.c |  182 +++++++-----
 drivers/infiniband/hw/hfi1/user_exp_rcv.h |   17 +
 drivers/infiniband/hw/hfi1/user_sdma.c    |  190 ++++++-------
 drivers/infiniband/hw/hfi1/user_sdma.h    |   18 +
 drivers/infiniband/hw/hfi1/verbs.h        |    5 
 drivers/infiniband/hw/hfi1/vnic_main.c    |    8 -
 include/rdma/rdmavt_qp.h                  |    1 
 19 files changed, 633 insertions(+), 603 deletions(-)

--
-Denny

[PATCH 01/15] IB/hfi1: Fix checks for Offline transient state

[Dennis Dalessandro]

From: Jakub Byczkowski <jakub.byczkowski-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

In goto_offline() function pstate is masked by 0xff when compared
to PLS_OFFLINE state. Mask should be 0xf0, since upper 4 bits
specify the "major" state.

Reviewed-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Reviewed-by: Easwar Hariharan <easwar.hariharan-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Jakub Byczkowski <jakub.byczkowski-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/infiniband/hw/hfi1/chip.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/drivers/infiniband/hw/hfi1/chip.c b/drivers/infiniband/hw/hfi1/chip.c
index 0f6916d..c96bb3a 100644
--- a/drivers/infiniband/hw/hfi1/chip.c
+++ b/drivers/infiniband/hw/hfi1/chip.c
@@ -10233,7 +10233,7 @@ static int goto_offline(struct hfi1_pportdata *ppd, u8 rem_reason)
 	if (pstate == PLS_OFFLINE) {
 		do_transition = 0;	/* in right state */
 		do_wait = 0;		/* ...no need to wait */
-	} else if ((pstate & 0xff) == PLS_OFFLINE) {
+	} else if ((pstate & 0xf0) == PLS_OFFLINE) {
 		do_transition = 0;	/* in an offline transient state */
 		do_wait = 1;		/* ...wait for it to settle */
 	} else {

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH 02/15] IB/hfi1, IB/rdmavt: Move r_adefered to r_lock cache line

[Dennis Dalessandro]

From: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

This field is causing excessive cache line bouncing.

There are spare bytes in the r_lock cache line so the best approach
is to make an rvt QP field and remove from the hfi1 priv field.

Signed-off-by: Sebastian Sanchez <sebastian.sanchez-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/infiniband/hw/hfi1/qp.c    |    4 +---
 drivers/infiniband/hw/hfi1/rc.c    |   13 ++++---------
 drivers/infiniband/hw/hfi1/verbs.h |    1 -
 include/rdma/rdmavt_qp.h           |    1 +
 4 files changed, 6 insertions(+), 13 deletions(-)

diff --git a/drivers/infiniband/hw/hfi1/qp.c b/drivers/infiniband/hw/hfi1/qp.c
index 4573e4c..650305c 100644
--- a/drivers/infiniband/hw/hfi1/qp.c
+++ b/drivers/infiniband/hw/hfi1/qp.c
@@ -731,9 +731,7 @@ void quiesce_qp(struct rvt_qp *qp)
 
 void notify_qp_reset(struct rvt_qp *qp)
 {
-	struct hfi1_qp_priv *priv = qp->priv;
-
-	priv->r_adefered = 0;
+	qp->r_adefered = 0;
 	clear_ahg(qp);
 }
 
diff --git a/drivers/infiniband/hw/hfi1/rc.c b/drivers/infiniband/hw/hfi1/rc.c
index 75a729c..069bdaf 100644
--- a/drivers/infiniband/hw/hfi1/rc.c
+++ b/drivers/infiniband/hw/hfi1/rc.c
@@ -727,10 +727,9 @@ void hfi1_send_rc_ack(struct hfi1_ctxtdata *rcd, struct rvt_qp *qp,
 	struct ib_header hdr;
 	struct ib_other_headers *ohdr;
 	unsigned long flags;
-	struct hfi1_qp_priv *priv = qp->priv;
 
 	/* clear the defer count */
-	priv->r_adefered = 0;
+	qp->r_adefered = 0;
 
 	/* Don't send ACK or NAK if a RDMA read or atomic is pending. */
 	if (qp->s_flags & RVT_S_RESP_PENDING)
@@ -1604,9 +1603,7 @@ static inline void rc_defered_ack(struct hfi1_ctxtdata *rcd,
 
 static inline void rc_cancel_ack(struct rvt_qp *qp)
 {
-	struct hfi1_qp_priv *priv = qp->priv;
-
-	priv->r_adefered = 0;
+	qp->r_adefered = 0;
 	if (list_empty(&qp->rspwait))
 		return;
 	list_del_init(&qp->rspwait);
@@ -2314,13 +2311,11 @@ void hfi1_rc_rcv(struct hfi1_packet *packet)
 	qp->r_nak_state = 0;
 	/* Send an ACK if requested or required. */
 	if (psn & IB_BTH_REQ_ACK) {
-		struct hfi1_qp_priv *priv = qp->priv;
-
 		if (packet->numpkt == 0) {
 			rc_cancel_ack(qp);
 			goto send_ack;
 		}
-		if (priv->r_adefered >= HFI1_PSN_CREDIT) {
+		if (qp->r_adefered >= HFI1_PSN_CREDIT) {
 			rc_cancel_ack(qp);
 			goto send_ack;
 		}
@@ -2328,7 +2323,7 @@ void hfi1_rc_rcv(struct hfi1_packet *packet)
 			rc_cancel_ack(qp);
 			goto send_ack;
 		}
-		priv->r_adefered++;
+		qp->r_adefered++;
 		rc_defered_ack(rcd, qp);
 	}
 	return;
diff --git a/drivers/infiniband/hw/hfi1/verbs.h b/drivers/infiniband/hw/hfi1/verbs.h
index 52ff275..c0913c6 100644
--- a/drivers/infiniband/hw/hfi1/verbs.h
+++ b/drivers/infiniband/hw/hfi1/verbs.h
@@ -125,7 +125,6 @@ struct hfi1_qp_priv {
 	struct sdma_engine *s_sde;                /* current sde */
 	struct send_context *s_sendcontext;       /* current sendcontext */
 	u8 s_sc;		                  /* SC[0..4] for next packet */
-	u8 r_adefered;                            /* number of acks defered */
 	struct iowait s_iowait;
 	struct rvt_qp *owner;
 };
diff --git a/include/rdma/rdmavt_qp.h b/include/rdma/rdmavt_qp.h
index 1d8141a..be6472e 100644
--- a/include/rdma/rdmavt_qp.h
+++ b/include/rdma/rdmavt_qp.h
@@ -324,6 +324,7 @@ struct rvt_qp {
 	u8 r_state;             /* opcode of last packet received */
 	u8 r_flags;
 	u8 r_head_ack_queue;    /* index into s_ack_queue[] */
+	u8 r_adefered;          /* defered ack count */
 
 	struct list_head rspwait;       /* link for waiting to respond */
 

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH 03/15] IB/hfi1: Fix yield logic in send engine

[Dennis Dalessandro]

From: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

When there are many RC QPs and an RDMA READ request
is sent, timeouts occur on the requester side because
of fairness among RC QPs on their relative SDMA engine
on the responder side.  This also hits write and send, but
to a lesser extent.

Complicating the issue is that the current code checks if workqueue
is congested before scheduling other QPs, however, this
check is based on the number of active entries in the
workqueue, which was found to be too big to for
workqueue_congested() to be effective.

Fix by reducing the number of active entries as revealed by
experimentation from the default of num_sdma to
HFI1_MAX_ACTIVE_WORKQUEUE_ENTRIES.  Retry counts were monitored
to determine the correct value.

Tracing to investigate any future issues is also added.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Sebastian Sanchez <sebastian.sanchez-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/infiniband/hw/hfi1/init.c     |    3 +
 drivers/infiniband/hw/hfi1/ruc.c      |   80 +++++++++++++++++++++------------
 drivers/infiniband/hw/hfi1/trace_tx.h |   34 ++++++++++++++
 drivers/infiniband/hw/hfi1/verbs.h    |    4 ++
 4 files changed, 90 insertions(+), 31 deletions(-)

diff --git a/drivers/infiniband/hw/hfi1/init.c b/drivers/infiniband/hw/hfi1/init.c
index 4d6b9f8..71b0204 100644
--- a/drivers/infiniband/hw/hfi1/init.c
+++ b/drivers/infiniband/hw/hfi1/init.c
@@ -70,6 +70,7 @@
 #undef pr_fmt
 #define pr_fmt(fmt) DRIVER_NAME ": " fmt
 
+#define HFI1_MAX_ACTIVE_WORKQUEUE_ENTRIES 5
 /*
  * min buffers we want to have per context, after driver
  */
@@ -623,7 +624,7 @@ static int create_workqueues(struct hfi1_devdata *dd)
 				alloc_workqueue(
 				    "hfi%d_%d",
 				    WQ_SYSFS | WQ_HIGHPRI | WQ_CPU_INTENSIVE,
-				    dd->num_sdma,
+				    HFI1_MAX_ACTIVE_WORKQUEUE_ENTRIES,
 				    dd->unit, pidx);
 			if (!ppd->hfi1_wq)
 				goto wq_error;
diff --git a/drivers/infiniband/hw/hfi1/ruc.c b/drivers/infiniband/hw/hfi1/ruc.c
index 891ba0a..3a17dab 100644
--- a/drivers/infiniband/hw/hfi1/ruc.c
+++ b/drivers/infiniband/hw/hfi1/ruc.c
@@ -800,6 +800,43 @@ void hfi1_make_ruc_header(struct rvt_qp *qp, struct ib_other_headers *ohdr,
 /* when sending, force a reschedule every one of these periods */
 #define SEND_RESCHED_TIMEOUT (5 * HZ)  /* 5s in jiffies */
 
+/**
+ * schedule_send_yield - test for a yield required for QP send engine
+ * @timeout: Final time for timeout slice for jiffies
+ * @qp: a pointer to QP
+ * @ps: a pointer to a structure with commonly lookup values for
+ *      the the send engine progress
+ *
+ * This routine checks if the time slice for the QP has expired
+ * for RC QPs, if so an additional work entry is queued. At this
+ * point, other QPs have an opportunity to be scheduled. It
+ * returns true if a yield is required, otherwise, false
+ * is returned.
+ */
+static bool schedule_send_yield(struct rvt_qp *qp,
+				struct hfi1_pkt_state *ps)
+{
+	if (unlikely(time_after(jiffies, ps->timeout))) {
+		if (!ps->in_thread ||
+		    workqueue_congested(ps->cpu, ps->ppd->hfi1_wq)) {
+			spin_lock_irqsave(&qp->s_lock, ps->flags);
+			qp->s_flags &= ~RVT_S_BUSY;
+			hfi1_schedule_send(qp);
+			spin_unlock_irqrestore(&qp->s_lock, ps->flags);
+			this_cpu_inc(*ps->ppd->dd->send_schedule);
+			trace_hfi1_rc_expired_time_slice(qp, true);
+			return true;
+		}
+
+		cond_resched();
+		this_cpu_inc(*ps->ppd->dd->send_schedule);
+		ps->timeout = jiffies + ps->timeout_int;
+	}
+
+	trace_hfi1_rc_expired_time_slice(qp, false);
+	return false;
+}
+
 void hfi1_do_send_from_rvt(struct rvt_qp *qp)
 {
 	hfi1_do_send(qp, false);
@@ -827,13 +864,13 @@ void hfi1_do_send(struct rvt_qp *qp, bool in_thread)
 	struct hfi1_pkt_state ps;
 	struct hfi1_qp_priv *priv = qp->priv;
 	int (*make_req)(struct rvt_qp *qp, struct hfi1_pkt_state *ps);
-	unsigned long timeout;
-	unsigned long timeout_int;
-	int cpu;
 
 	ps.dev = to_idev(qp->ibqp.device);
 	ps.ibp = to_iport(qp->ibqp.device, qp->port_num);
 	ps.ppd = ppd_from_ibp(ps.ibp);
+	ps.in_thread = in_thread;
+
+	trace_hfi1_rc_do_send(qp, in_thread);
 
 	switch (qp->ibqp.qp_type) {
 	case IB_QPT_RC:
@@ -844,7 +881,7 @@ void hfi1_do_send(struct rvt_qp *qp, bool in_thread)
 			return;
 		}
 		make_req = hfi1_make_rc_req;
-		timeout_int = (qp->timeout_jiffies);
+		ps.timeout_int = qp->timeout_jiffies;
 		break;
 	case IB_QPT_UC:
 		if (!loopback && ((rdma_ah_get_dlid(&qp->remote_ah_attr) &
@@ -854,11 +891,11 @@ void hfi1_do_send(struct rvt_qp *qp, bool in_thread)
 			return;
 		}
 		make_req = hfi1_make_uc_req;
-		timeout_int = SEND_RESCHED_TIMEOUT;
+		ps.timeout_int = SEND_RESCHED_TIMEOUT;
 		break;
 	default:
 		make_req = hfi1_make_ud_req;
-		timeout_int = SEND_RESCHED_TIMEOUT;
+		ps.timeout_int = SEND_RESCHED_TIMEOUT;
 	}
 
 	spin_lock_irqsave(&qp->s_lock, ps.flags);
@@ -871,9 +908,11 @@ void hfi1_do_send(struct rvt_qp *qp, bool in_thread)
 
 	qp->s_flags |= RVT_S_BUSY;
 
-	timeout = jiffies + (timeout_int) / 8;
-	cpu = priv->s_sde ? priv->s_sde->cpu :
+	ps.timeout_int = ps.timeout_int / 8;
+	ps.timeout = jiffies + ps.timeout_int;
+	ps.cpu = priv->s_sde ? priv->s_sde->cpu :
 			cpumask_first(cpumask_of_node(ps.ppd->dd->node));
+
 	/* insure a pre-built packet is handled  */
 	ps.s_txreq = get_waiting_verbs_txreq(qp);
 	do {
@@ -889,28 +928,9 @@ void hfi1_do_send(struct rvt_qp *qp, bool in_thread)
 			/* Record that s_ahg is empty. */
 			qp->s_hdrwords = 0;
 			/* allow other tasks to run */
-			if (unlikely(time_after(jiffies, timeout))) {
-				if (!in_thread ||
-				    workqueue_congested(
-						cpu,
-						ps.ppd->hfi1_wq)) {
-					spin_lock_irqsave(
-						&qp->s_lock,
-						ps.flags);
-					qp->s_flags &= ~RVT_S_BUSY;
-					hfi1_schedule_send(qp);
-					spin_unlock_irqrestore(
-						&qp->s_lock,
-						ps.flags);
-					this_cpu_inc(
-						*ps.ppd->dd->send_schedule);
-					return;
-				}
-				cond_resched();
-				this_cpu_inc(
-					*ps.ppd->dd->send_schedule);
-				timeout = jiffies + (timeout_int) / 8;
-			}
+			if (schedule_send_yield(qp, &ps))
+				return;
+
 			spin_lock_irqsave(&qp->s_lock, ps.flags);
 		}
 	} while (make_req(qp, &ps));
diff --git a/drivers/infiniband/hw/hfi1/trace_tx.h b/drivers/infiniband/hw/hfi1/trace_tx.h
index 2c9ac57..c59809a 100644
--- a/drivers/infiniband/hw/hfi1/trace_tx.h
+++ b/drivers/infiniband/hw/hfi1/trace_tx.h
@@ -676,6 +676,40 @@
 	)
 );
 
+DECLARE_EVENT_CLASS(
+	hfi1_do_send_template,
+	TP_PROTO(struct rvt_qp *qp, bool flag),
+	TP_ARGS(qp, flag),
+	TP_STRUCT__entry(
+		DD_DEV_ENTRY(dd_from_ibdev(qp->ibqp.device))
+		__field(u32, qpn)
+		__field(bool, flag)
+	),
+	TP_fast_assign(
+		DD_DEV_ASSIGN(dd_from_ibdev(qp->ibqp.device))
+		__entry->qpn = qp->ibqp.qp_num;
+		__entry->flag = flag;
+	),
+	TP_printk(
+		"[%s] qpn %x flag %d",
+		__get_str(dev),
+		__entry->qpn,
+		__entry->flag
+	)
+);
+
+DEFINE_EVENT(
+	hfi1_do_send_template, hfi1_rc_do_send,
+	TP_PROTO(struct rvt_qp *qp, bool flag),
+	TP_ARGS(qp, flag)
+);
+
+DEFINE_EVENT(
+	hfi1_do_send_template, hfi1_rc_expired_time_slice,
+	TP_PROTO(struct rvt_qp *qp, bool flag),
+	TP_ARGS(qp, flag)
+);
+
 #endif /* __HFI1_TRACE_TX_H */
 
 #undef TRACE_INCLUDE_PATH
diff --git a/drivers/infiniband/hw/hfi1/verbs.h b/drivers/infiniband/hw/hfi1/verbs.h
index c0913c6..cd635d0 100644
--- a/drivers/infiniband/hw/hfi1/verbs.h
+++ b/drivers/infiniband/hw/hfi1/verbs.h
@@ -139,6 +139,10 @@ struct hfi1_pkt_state {
 	struct hfi1_pportdata *ppd;
 	struct verbs_txreq *s_txreq;
 	unsigned long flags;
+	unsigned long timeout;
+	unsigned long timeout_int;
+	int cpu;
+	bool in_thread;
 };
 
 #define HFI1_PSN_CREDIT  16

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH 04/15] IB/hfi1: Get rid of divide when setting the tx request header

[Dennis Dalessandro]

From: Sebastian Sanchez <sebastian.sanchez-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

Div instructions show costly in profiles when
the tx request header is set. Using right shift
instead of a divide operation reduces the cycles
spent in the function that sets the tx request
header as shown in the profile. Use right shift
operation instead.

Profile before change:
43.24%  009
|
|--23.41%-- user_sdma_send_pkts
|          |
|          |--99.90%-- hfi1_user_sdma_process_requestAfter:

Profile after change:
45.75%  009
|
|--14.81%-- user_sdma_send_pkts
|          |
|          |--99.95%-- hfi1_user_sdma_process_request

Reviewed-by: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Sebastian Sanchez <sebastian.sanchez-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/infiniband/hw/hfi1/user_sdma.c |   34 ++++++++++++++++----------------
 1 files changed, 17 insertions(+), 17 deletions(-)

diff --git a/drivers/infiniband/hw/hfi1/user_sdma.c b/drivers/infiniband/hw/hfi1/user_sdma.c
index 0749689..8adb6df 100644
--- a/drivers/infiniband/hw/hfi1/user_sdma.c
+++ b/drivers/infiniband/hw/hfi1/user_sdma.c
@@ -143,7 +143,9 @@
 
 /* KDETH OM multipliers and switch over point */
 #define KDETH_OM_SMALL     4
+#define KDETH_OM_SMALL_SHIFT     2
 #define KDETH_OM_LARGE     64
+#define KDETH_OM_LARGE_SHIFT     6
 #define KDETH_OM_MAX_SIZE  (1 << ((KDETH_OM_LARGE / KDETH_OM_SMALL) + 1))
 
 /* Tx request flag bits */
@@ -229,12 +231,6 @@ struct user_sdma_request {
 	 */
 	u32 tidoffset;
 	/*
-	 * KDETH.OM
-	 * Remember this because the header template always sets it
-	 * to 0.
-	 */
-	u8 omfactor;
-	/*
 	 * We copy the iovs for this request (based on
 	 * info.iovcnt). These are only the data vectors
 	 */
@@ -1323,6 +1319,7 @@ static int set_txreq_header(struct user_sdma_request *req,
 {
 	struct hfi1_user_sdma_pkt_q *pq = req->pq;
 	struct hfi1_pkt_header *hdr = &tx->hdr;
+	u8 omfactor; /* KDETH.OM */
 	u16 pbclen;
 	int ret;
 	u32 tidval = 0, lrhlen = get_lrh_len(*hdr, pad_len(datalen));
@@ -1400,8 +1397,9 @@ static int set_txreq_header(struct user_sdma_request *req,
 			}
 			tidval = req->tids[req->tididx];
 		}
-		req->omfactor = EXP_TID_GET(tidval, LEN) * PAGE_SIZE >=
-			KDETH_OM_MAX_SIZE ? KDETH_OM_LARGE : KDETH_OM_SMALL;
+		omfactor = EXP_TID_GET(tidval, LEN) * PAGE_SIZE >=
+			KDETH_OM_MAX_SIZE ? KDETH_OM_LARGE_SHIFT :
+			KDETH_OM_SMALL_SHIFT;
 		/* Set KDETH.TIDCtrl based on value for this TID. */
 		KDETH_SET(hdr->kdeth.ver_tid_offset, TIDCTRL,
 			  EXP_TID_GET(tidval, CTRL));
@@ -1416,12 +1414,12 @@ static int set_txreq_header(struct user_sdma_request *req,
 		 * transfer.
 		 */
 		SDMA_DBG(req, "TID offset %ubytes %uunits om%u",
-			 req->tidoffset, req->tidoffset / req->omfactor,
-			 req->omfactor != KDETH_OM_SMALL);
+			 req->tidoffset, req->tidoffset >> omfactor,
+			 omfactor != KDETH_OM_SMALL_SHIFT);
 		KDETH_SET(hdr->kdeth.ver_tid_offset, OFFSET,
-			  req->tidoffset / req->omfactor);
+			  req->tidoffset >> omfactor);
 		KDETH_SET(hdr->kdeth.ver_tid_offset, OM,
-			  req->omfactor != KDETH_OM_SMALL);
+			  omfactor != KDETH_OM_SMALL_SHIFT);
 	}
 done:
 	trace_hfi1_sdma_user_header(pq->dd, pq->ctxt, pq->subctxt,
@@ -1433,6 +1431,7 @@ static int set_txreq_header_ahg(struct user_sdma_request *req,
 				struct user_sdma_txreq *tx, u32 len)
 {
 	int diff = 0;
+	u8 omfactor; /* KDETH.OM */
 	struct hfi1_user_sdma_pkt_q *pq = req->pq;
 	struct hfi1_pkt_header *hdr = &req->hdr;
 	u16 pbclen = le16_to_cpu(hdr->pbc[0]);
@@ -1484,14 +1483,15 @@ static int set_txreq_header_ahg(struct user_sdma_request *req,
 			}
 			tidval = req->tids[req->tididx];
 		}
-		req->omfactor = ((EXP_TID_GET(tidval, LEN) *
+		omfactor = ((EXP_TID_GET(tidval, LEN) *
 				  PAGE_SIZE) >=
-				 KDETH_OM_MAX_SIZE) ? KDETH_OM_LARGE :
-			KDETH_OM_SMALL;
+				 KDETH_OM_MAX_SIZE) ? KDETH_OM_LARGE_SHIFT :
+				 KDETH_OM_SMALL_SHIFT;
 		/* KDETH.OM and KDETH.OFFSET (TID) */
 		AHG_HEADER_SET(req->ahg, diff, 7, 0, 16,
-			       ((!!(req->omfactor - KDETH_OM_SMALL)) << 15 |
-				((req->tidoffset / req->omfactor) & 0x7fff)));
+			       ((!!(omfactor - KDETH_OM_SMALL_SHIFT)) << 15 |
+				((req->tidoffset >> omfactor)
+				 & 0x7fff)));
 		/* KDETH.TIDCtrl, KDETH.TID, KDETH.Intr, KDETH.SH */
 		val = cpu_to_le16(((EXP_TID_GET(tidval, CTRL) & 0x3) << 10) |
 				   (EXP_TID_GET(tidval, IDX) & 0x3ff));

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH 05/15] IB/hfi1: Adjust default eager_buffer_size to 8MB

[Dennis Dalessandro]

From: Tymoteusz Kielan <tymoteusz.kielan-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

Performance analysis shows benefits for PSM2 in increasing eager buffer
size from 2MB to 8MB. The change has neutral impact on verbs.
Make change to the module parameter's default value. Allocation
ring down was verified to work with the larger buffer size.

Reviewed-by: Tadeusz Struk <tadeusz.struk-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Tymoteusz Kielan <tymoteusz.kielan-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/infiniband/hw/hfi1/init.c |    4 ++--
 1 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/infiniband/hw/hfi1/init.c b/drivers/infiniband/hw/hfi1/init.c
index 71b0204..c573f06 100644
--- a/drivers/infiniband/hw/hfi1/init.c
+++ b/drivers/infiniband/hw/hfi1/init.c
@@ -102,9 +102,9 @@
 module_param_named(rcvarr_split, hfi1_rcvarr_split, uint, S_IRUGO);
 MODULE_PARM_DESC(rcvarr_split, "Percent of context's RcvArray entries used for Eager buffers");
 
-static uint eager_buffer_size = (2 << 20); /* 2MB */
+static uint eager_buffer_size = (8 << 20); /* 8MB */
 module_param(eager_buffer_size, uint, S_IRUGO);
-MODULE_PARM_DESC(eager_buffer_size, "Size of the eager buffers, default: 2MB");
+MODULE_PARM_DESC(eager_buffer_size, "Size of the eager buffers, default: 8MB");
 
 static uint rcvhdrcnt = 2048; /* 2x the max eager buffer count */
 module_param_named(rcvhdrcnt, rcvhdrcnt, uint, S_IRUGO);

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH 06/15] IB/hfi1: Return an error on memory allocation failure

[Dennis Dalessandro]

From: Michael J. Ruhl <michael.j.ruhl@intel.com>

If the eager buffer allocation fails, it is necessary to return
an error code.

Cc: stable@vger.kernel.org
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
---
 drivers/infiniband/hw/hfi1/init.c |    1 +
 1 files changed, 1 insertions(+), 0 deletions(-)

diff --git a/drivers/infiniband/hw/hfi1/init.c b/drivers/infiniband/hw/hfi1/init.c
index c573f06..b2db776 100644
--- a/drivers/infiniband/hw/hfi1/init.c
+++ b/drivers/infiniband/hw/hfi1/init.c
@@ -1778,6 +1778,7 @@ int hfi1_setup_eagerbufs(struct hfi1_ctxtdata *rcd)
 			    !HFI1_CAP_KGET_MASK(rcd->flags, MULTI_PKT_EGR)) {
 				dd_dev_err(dd, "ctxt%u: Failed to allocate eager buffers\n",
 					   rcd->ctxt);
+				ret = -ENOMEM;
 				goto bail_rcvegrbuf_phys;
 			}
 

[PATCH 07/15] IB/hfi1: Fix a subcontext memory leak

[Dennis Dalessandro]

From: Michael J. Ruhl <michael.j.ruhl-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

The only context that frees user_exp_rcv data structures is the last
context closed (from a sub-context set).  This leaks the allocations
from the other sub-contexts.  Separate the common frees from the
specific frees and call them at the appropriate time.

Using KEDR to check for memory leaks we get:

Before test:

[leak_check] Possible leaks: 25

After test:

[leak_check] Possible leaks: 31  (6 leaked data structures)

After patch applied (before and after test have the same value)

[leak_check] Possible leaks: 25

Each leak is 192 + 13440 + 6720 = 20352 bytes per sub-context.

Cc: stable-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
Reviewed-by: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/infiniband/hw/hfi1/file_ops.c     |    5 ++++-
 drivers/infiniband/hw/hfi1/user_exp_rcv.c |   32 ++++++++++++++++-------------
 drivers/infiniband/hw/hfi1/user_exp_rcv.h |    1 +
 3 files changed, 23 insertions(+), 15 deletions(-)

diff --git a/drivers/infiniband/hw/hfi1/file_ops.c b/drivers/infiniband/hw/hfi1/file_ops.c
index 3d9bce4..afdf3ef 100644
--- a/drivers/infiniband/hw/hfi1/file_ops.c
+++ b/drivers/infiniband/hw/hfi1/file_ops.c
@@ -756,6 +756,9 @@ static int hfi1_file_close(struct inode *inode, struct file *fp)
 	/* release the cpu */
 	hfi1_put_proc_affinity(fdata->rec_cpu_num);
 
+	/* clean up rcv side */
+	hfi1_user_exp_rcv_free(fdata);
+
 	/*
 	 * Clear any left over, unhandled events so the next process that
 	 * gets this context doesn't get confused.
@@ -795,7 +798,7 @@ static int hfi1_file_close(struct inode *inode, struct file *fp)
 
 	dd->rcd[uctxt->ctxt] = NULL;
 
-	hfi1_user_exp_rcv_free(fdata);
+	hfi1_user_exp_rcv_grp_free(uctxt);
 	hfi1_clear_ctxt_pkey(dd, uctxt->ctxt);
 
 	uctxt->rcvwait_to = 0;
diff --git a/drivers/infiniband/hw/hfi1/user_exp_rcv.c b/drivers/infiniband/hw/hfi1/user_exp_rcv.c
index 35c6e7e..c7f13df 100644
--- a/drivers/infiniband/hw/hfi1/user_exp_rcv.c
+++ b/drivers/infiniband/hw/hfi1/user_exp_rcv.c
@@ -252,36 +252,40 @@ int hfi1_user_exp_rcv_init(struct file *fp)
 	return ret;
 }
 
+void hfi1_user_exp_rcv_grp_free(struct hfi1_ctxtdata *uctxt)
+{
+	struct tid_group *grp, *gptr;
+
+	list_for_each_entry_safe(grp, gptr, &uctxt->tid_group_list.list,
+				 list) {
+		list_del_init(&grp->list);
+		kfree(grp);
+	}
+	hfi1_clear_tids(uctxt);
+}
+
 int hfi1_user_exp_rcv_free(struct hfi1_filedata *fd)
 {
 	struct hfi1_ctxtdata *uctxt = fd->uctxt;
-	struct tid_group *grp, *gptr;
 
-	if (!test_bit(HFI1_CTXT_SETUP_DONE, &uctxt->event_flags))
-		return 0;
 	/*
 	 * The notifier would have been removed when the process'es mm
 	 * was freed.
 	 */
-	if (fd->handler)
+	if (fd->handler) {
 		hfi1_mmu_rb_unregister(fd->handler);
-
-	kfree(fd->invalid_tids);
-
-	if (!uctxt->cnt) {
+	} else {
 		if (!EXP_TID_SET_EMPTY(uctxt->tid_full_list))
 			unlock_exp_tids(uctxt, &uctxt->tid_full_list, fd);
 		if (!EXP_TID_SET_EMPTY(uctxt->tid_used_list))
 			unlock_exp_tids(uctxt, &uctxt->tid_used_list, fd);
-		list_for_each_entry_safe(grp, gptr, &uctxt->tid_group_list.list,
-					 list) {
-			list_del_init(&grp->list);
-			kfree(grp);
-		}
-		hfi1_clear_tids(uctxt);
 	}
 
+	kfree(fd->invalid_tids);
+	fd->invalid_tids = NULL;
+
 	kfree(fd->entry_to_rb);
+	fd->entry_to_rb = NULL;
 	return 0;
 }
 
diff --git a/drivers/infiniband/hw/hfi1/user_exp_rcv.h b/drivers/infiniband/hw/hfi1/user_exp_rcv.h
index 9bc8d9f..d1d7d3d 100644
--- a/drivers/infiniband/hw/hfi1/user_exp_rcv.h
+++ b/drivers/infiniband/hw/hfi1/user_exp_rcv.h
@@ -70,6 +70,7 @@
 		(tid) |= EXP_TID_SET(field, (value));			\
 	} while (0)
 
+void hfi1_user_exp_rcv_grp_free(struct hfi1_ctxtdata *uctxt);
 int hfi1_user_exp_rcv_init(struct file *);
 int hfi1_user_exp_rcv_free(struct hfi1_filedata *);
 int hfi1_user_exp_rcv_setup(struct file *, struct hfi1_tid_info *);

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH 07/15] IB/hfi1: Fix a subcontext memory leak

[Dennis Dalessandro]

From: Michael J. Ruhl <michael.j.ruhl@intel.com>

The only context that frees user_exp_rcv data structures is the last
context closed (from a sub-context set).  This leaks the allocations
from the other sub-contexts.  Separate the common frees from the
specific frees and call them at the appropriate time.

Using KEDR to check for memory leaks we get:

Before test:

[leak_check] Possible leaks: 25

After test:

[leak_check] Possible leaks: 31  (6 leaked data structures)

After patch applied (before and after test have the same value)

[leak_check] Possible leaks: 25

Each leak is 192 + 13440 + 6720 = 20352 bytes per sub-context.

Cc: stable@vger.kernel.org
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
---
 drivers/infiniband/hw/hfi1/file_ops.c     |    5 ++++-
 drivers/infiniband/hw/hfi1/user_exp_rcv.c |   32 ++++++++++++++++-------------
 drivers/infiniband/hw/hfi1/user_exp_rcv.h |    1 +
 3 files changed, 23 insertions(+), 15 deletions(-)

diff --git a/drivers/infiniband/hw/hfi1/file_ops.c b/drivers/infiniband/hw/hfi1/file_ops.c
index 3d9bce4..afdf3ef 100644
--- a/drivers/infiniband/hw/hfi1/file_ops.c
+++ b/drivers/infiniband/hw/hfi1/file_ops.c
@@ -756,6 +756,9 @@ static int hfi1_file_close(struct inode *inode, struct file *fp)
 	/* release the cpu */
 	hfi1_put_proc_affinity(fdata->rec_cpu_num);
 
+	/* clean up rcv side */
+	hfi1_user_exp_rcv_free(fdata);
+
 	/*
 	 * Clear any left over, unhandled events so the next process that
 	 * gets this context doesn't get confused.
@@ -795,7 +798,7 @@ static int hfi1_file_close(struct inode *inode, struct file *fp)
 
 	dd->rcd[uctxt->ctxt] = NULL;
 
-	hfi1_user_exp_rcv_free(fdata);
+	hfi1_user_exp_rcv_grp_free(uctxt);
 	hfi1_clear_ctxt_pkey(dd, uctxt->ctxt);
 
 	uctxt->rcvwait_to = 0;
diff --git a/drivers/infiniband/hw/hfi1/user_exp_rcv.c b/drivers/infiniband/hw/hfi1/user_exp_rcv.c
index 35c6e7e..c7f13df 100644
--- a/drivers/infiniband/hw/hfi1/user_exp_rcv.c
+++ b/drivers/infiniband/hw/hfi1/user_exp_rcv.c
@@ -252,36 +252,40 @@ int hfi1_user_exp_rcv_init(struct file *fp)
 	return ret;
 }
 
+void hfi1_user_exp_rcv_grp_free(struct hfi1_ctxtdata *uctxt)
+{
+	struct tid_group *grp, *gptr;
+
+	list_for_each_entry_safe(grp, gptr, &uctxt->tid_group_list.list,
+				 list) {
+		list_del_init(&grp->list);
+		kfree(grp);
+	}
+	hfi1_clear_tids(uctxt);
+}
+
 int hfi1_user_exp_rcv_free(struct hfi1_filedata *fd)
 {
 	struct hfi1_ctxtdata *uctxt = fd->uctxt;
-	struct tid_group *grp, *gptr;
 
-	if (!test_bit(HFI1_CTXT_SETUP_DONE, &uctxt->event_flags))
-		return 0;
 	/*
 	 * The notifier would have been removed when the process'es mm
 	 * was freed.
 	 */
-	if (fd->handler)
+	if (fd->handler) {
 		hfi1_mmu_rb_unregister(fd->handler);
-
-	kfree(fd->invalid_tids);
-
-	if (!uctxt->cnt) {
+	} else {
 		if (!EXP_TID_SET_EMPTY(uctxt->tid_full_list))
 			unlock_exp_tids(uctxt, &uctxt->tid_full_list, fd);
 		if (!EXP_TID_SET_EMPTY(uctxt->tid_used_list))
 			unlock_exp_tids(uctxt, &uctxt->tid_used_list, fd);
-		list_for_each_entry_safe(grp, gptr, &uctxt->tid_group_list.list,
-					 list) {
-			list_del_init(&grp->list);
-			kfree(grp);
-		}
-		hfi1_clear_tids(uctxt);
 	}
 
+	kfree(fd->invalid_tids);
+	fd->invalid_tids = NULL;
+
 	kfree(fd->entry_to_rb);
+	fd->entry_to_rb = NULL;
 	return 0;
 }
 
diff --git a/drivers/infiniband/hw/hfi1/user_exp_rcv.h b/drivers/infiniband/hw/hfi1/user_exp_rcv.h
index 9bc8d9f..d1d7d3d 100644
--- a/drivers/infiniband/hw/hfi1/user_exp_rcv.h
+++ b/drivers/infiniband/hw/hfi1/user_exp_rcv.h
@@ -70,6 +70,7 @@
 		(tid) |= EXP_TID_SET(field, (value));			\
 	} while (0)
 
+void hfi1_user_exp_rcv_grp_free(struct hfi1_ctxtdata *uctxt);
 int hfi1_user_exp_rcv_init(struct file *);
 int hfi1_user_exp_rcv_free(struct hfi1_filedata *);
 int hfi1_user_exp_rcv_setup(struct file *, struct hfi1_tid_info *);

[PATCH 08/15] IB/hfi1: Name function prototype parameters

[Dennis Dalessandro]

From: Michael J. Ruhl <michael.j.ruhl-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

To improve the readability of function prototypes, give the parameters
names.

Reviewed-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/infiniband/hw/hfi1/chip.c         |    8 ++-
 drivers/infiniband/hw/hfi1/chip.h         |    6 ++
 drivers/infiniband/hw/hfi1/driver.c       |    4 +-
 drivers/infiniband/hw/hfi1/file_ops.c     |   54 ++++++++++++----------
 drivers/infiniband/hw/hfi1/hfi.h          |   72 +++++++++++++++--------------
 drivers/infiniband/hw/hfi1/init.c         |    2 -
 drivers/infiniband/hw/hfi1/user_exp_rcv.c |   30 +++++++-----
 drivers/infiniband/hw/hfi1/user_exp_rcv.h |   12 ++---
 drivers/infiniband/hw/hfi1/user_sdma.c    |   62 +++++++++++++------------
 drivers/infiniband/hw/hfi1/user_sdma.h    |   13 +++--
 10 files changed, 143 insertions(+), 120 deletions(-)

diff --git a/drivers/infiniband/hw/hfi1/chip.c b/drivers/infiniband/hw/hfi1/chip.c
index c96bb3a..5aa95da 100644
--- a/drivers/infiniband/hw/hfi1/chip.c
+++ b/drivers/infiniband/hw/hfi1/chip.c
@@ -1055,7 +1055,7 @@ static void handle_lcb_err(struct hfi1_devdata *dd,
 static void handle_sdma_err(struct hfi1_devdata *dd, u32 unused, u64 reg);
 static void handle_egress_err(struct hfi1_devdata *dd, u32 unused, u64 reg);
 static void handle_txe_err(struct hfi1_devdata *dd, u32 unused, u64 reg);
-static void set_partition_keys(struct hfi1_pportdata *);
+static void set_partition_keys(struct hfi1_pportdata *ppd);
 static const char *link_state_name(u32 state);
 static const char *link_state_reason_name(struct hfi1_pportdata *ppd,
 					  u32 state);
@@ -1068,9 +1068,9 @@ static int wait_logical_linkstate(struct hfi1_pportdata *ppd, u32 state,
 				  int msecs);
 static void read_planned_down_reason_code(struct hfi1_devdata *dd, u8 *pdrrc);
 static void read_link_down_reason(struct hfi1_devdata *dd, u8 *ldr);
-static void handle_temp_err(struct hfi1_devdata *);
-static void dc_shutdown(struct hfi1_devdata *);
-static void dc_start(struct hfi1_devdata *);
+static void handle_temp_err(struct hfi1_devdata *dd);
+static void dc_shutdown(struct hfi1_devdata *dd);
+static void dc_start(struct hfi1_devdata *dd);
 static int qos_rmt_entries(struct hfi1_devdata *dd, unsigned int *mp,
 			   unsigned int *np);
 static void clear_full_mgmt_pkey(struct hfi1_pportdata *ppd);
diff --git a/drivers/infiniband/hw/hfi1/chip.h b/drivers/infiniband/hw/hfi1/chip.h
index b9dbf16..4c3c88d 100644
--- a/drivers/infiniband/hw/hfi1/chip.h
+++ b/drivers/infiniband/hw/hfi1/chip.h
@@ -636,7 +636,8 @@ static inline void write_uctxt_csr(struct hfi1_devdata *dd, int ctxt,
 	write_csr(dd, offset0 + (0x1000 * ctxt), value);
 }
 
-u64 create_pbc(struct hfi1_pportdata *ppd, u64, int, u32, u32);
+u64 create_pbc(struct hfi1_pportdata *ppd, u64 flags, int srate_mbs, u32 vl,
+	       u32 dw_len);
 
 /* firmware.c */
 #define SBUS_MASTER_BROADCAST 0xfd
@@ -728,7 +729,8 @@ void set_link_down_reason(struct hfi1_pportdata *ppd, u8 lcl_reason,
 void set_intr_state(struct hfi1_devdata *dd, u32 enable);
 void apply_link_downgrade_policy(struct hfi1_pportdata *ppd,
 				 int refresh_widths);
-void update_usrhead(struct hfi1_ctxtdata *, u32, u32, u32, u32, u32);
+void update_usrhead(struct hfi1_ctxtdata *rcd, u32 hd, u32 updegr, u32 egrhd,
+		    u32 intr_adjust, u32 npkts);
 int stop_drain_data_vls(struct hfi1_devdata *dd);
 int open_fill_data_vls(struct hfi1_devdata *dd);
 u32 ns_to_cclock(struct hfi1_devdata *dd, u32 ns);
diff --git a/drivers/infiniband/hw/hfi1/driver.c b/drivers/infiniband/hw/hfi1/driver.c
index 5278954..fb319d2 100644
--- a/drivers/infiniband/hw/hfi1/driver.c
+++ b/drivers/infiniband/hw/hfi1/driver.c
@@ -85,8 +85,8 @@
 MODULE_PARM_DESC(cu, "Credit return units");
 
 unsigned long hfi1_cap_mask = HFI1_CAP_MASK_DEFAULT;
-static int hfi1_caps_set(const char *, const struct kernel_param *);
-static int hfi1_caps_get(char *, const struct kernel_param *);
+static int hfi1_caps_set(const char *val, const struct kernel_param *kp);
+static int hfi1_caps_get(char *buffer, const struct kernel_param *kp);
 static const struct kernel_param_ops cap_ops = {
 	.set = hfi1_caps_set,
 	.get = hfi1_caps_get
diff --git a/drivers/infiniband/hw/hfi1/file_ops.c b/drivers/infiniband/hw/hfi1/file_ops.c
index afdf3ef..0b77283 100644
--- a/drivers/infiniband/hw/hfi1/file_ops.c
+++ b/drivers/infiniband/hw/hfi1/file_ops.c
@@ -70,30 +70,36 @@
 /*
  * File operation functions
  */
-static int hfi1_file_open(struct inode *, struct file *);
-static int hfi1_file_close(struct inode *, struct file *);
-static ssize_t hfi1_write_iter(struct kiocb *, struct iov_iter *);
-static unsigned int hfi1_poll(struct file *, struct poll_table_struct *);
-static int hfi1_file_mmap(struct file *, struct vm_area_struct *);
-
-static u64 kvirt_to_phys(void *);
-static int assign_ctxt(struct file *, struct hfi1_user_info *);
-static int init_subctxts(struct hfi1_ctxtdata *, const struct hfi1_user_info *);
-static int user_init(struct file *);
-static int get_ctxt_info(struct file *, void __user *, __u32);
-static int get_base_info(struct file *, void __user *, __u32);
-static int setup_ctxt(struct file *);
-static int setup_subctxt(struct hfi1_ctxtdata *);
-static int get_user_context(struct file *, struct hfi1_user_info *, int);
-static int find_shared_ctxt(struct file *, const struct hfi1_user_info *);
-static int allocate_ctxt(struct file *, struct hfi1_devdata *,
-			 struct hfi1_user_info *);
-static unsigned int poll_urgent(struct file *, struct poll_table_struct *);
-static unsigned int poll_next(struct file *, struct poll_table_struct *);
-static int user_event_ack(struct hfi1_ctxtdata *, int, unsigned long);
-static int set_ctxt_pkey(struct hfi1_ctxtdata *, unsigned, u16);
-static int manage_rcvq(struct hfi1_ctxtdata *, unsigned, int);
-static int vma_fault(struct vm_fault *);
+static int hfi1_file_open(struct inode *inode, struct file *fp);
+static int hfi1_file_close(struct inode *inode, struct file *fp);
+static ssize_t hfi1_write_iter(struct kiocb *kiocb, struct iov_iter *from);
+static unsigned int hfi1_poll(struct file *fp, struct poll_table_struct *pt);
+static int hfi1_file_mmap(struct file *fp, struct vm_area_struct *vma);
+
+static u64 kvirt_to_phys(void *addr);
+static int assign_ctxt(struct file *fp, struct hfi1_user_info *uinfo);
+static int init_subctxts(struct hfi1_ctxtdata *uctxt,
+			 const struct hfi1_user_info *uinfo);
+static int user_init(struct file *fp);
+static int get_ctxt_info(struct file *fp, void __user *ubase, __u32 len);
+static int get_base_info(struct file *fp, void __user *ubase, __u32 len);
+static int setup_ctxt(struct file *fp);
+static int setup_subctxt(struct hfi1_ctxtdata *uctxt);
+static int get_user_context(struct file *fp, struct hfi1_user_info *uinfo,
+			    int devno);
+static int find_shared_ctxt(struct file *fp,
+			    const struct hfi1_user_info *uinfo);
+static int allocate_ctxt(struct file *fp, struct hfi1_devdata *dd,
+			 struct hfi1_user_info *uinfo);
+static unsigned int poll_urgent(struct file *fp, struct poll_table_struct *pt);
+static unsigned int poll_next(struct file *fp, struct poll_table_struct *pt);
+static int user_event_ack(struct hfi1_ctxtdata *uctxt, int subctxt,
+			  unsigned long events);
+static int set_ctxt_pkey(struct hfi1_ctxtdata *uctxt, unsigned subctxt,
+			 u16 pkey);
+static int manage_rcvq(struct hfi1_ctxtdata *uctxt, unsigned subctxt,
+		       int start_stop);
+static int vma_fault(struct vm_fault *vmf);
 static long hfi1_file_ioctl(struct file *fp, unsigned int cmd,
 			    unsigned long arg);
 
diff --git a/drivers/infiniband/hw/hfi1/hfi.h b/drivers/infiniband/hw/hfi1/hfi.h
index f066743..a1e9b8b 100644
--- a/drivers/infiniband/hw/hfi1/hfi.h
+++ b/drivers/infiniband/hw/hfi1/hfi.h
@@ -1263,27 +1263,28 @@ struct hfi1_filedata {
 extern u32 hfi1_cpulist_count;
 extern unsigned long *hfi1_cpulist;
 
-int hfi1_init(struct hfi1_devdata *, int);
+int hfi1_init(struct hfi1_devdata *dd, int reinit);
 int hfi1_count_units(int *npresentp, int *nupp);
 int hfi1_count_active_units(void);
 
-int hfi1_diag_add(struct hfi1_devdata *);
-void hfi1_diag_remove(struct hfi1_devdata *);
+int hfi1_diag_add(struct hfi1_devdata *dd);
+void hfi1_diag_remove(struct hfi1_devdata *dd);
 void handle_linkup_change(struct hfi1_devdata *dd, u32 linkup);
 
 void handle_user_interrupt(struct hfi1_ctxtdata *rcd);
 
-int hfi1_create_rcvhdrq(struct hfi1_devdata *, struct hfi1_ctxtdata *);
-int hfi1_setup_eagerbufs(struct hfi1_ctxtdata *);
+int hfi1_create_rcvhdrq(struct hfi1_devdata *dd, struct hfi1_ctxtdata *rcd);
+int hfi1_setup_eagerbufs(struct hfi1_ctxtdata *rcd);
 int hfi1_create_ctxts(struct hfi1_devdata *dd);
-struct hfi1_ctxtdata *hfi1_create_ctxtdata(struct hfi1_pportdata *, u32, int);
-void hfi1_init_pportdata(struct pci_dev *, struct hfi1_pportdata *,
-			 struct hfi1_devdata *, u8, u8);
-void hfi1_free_ctxtdata(struct hfi1_devdata *, struct hfi1_ctxtdata *);
-
-int handle_receive_interrupt(struct hfi1_ctxtdata *, int);
-int handle_receive_interrupt_nodma_rtail(struct hfi1_ctxtdata *, int);
-int handle_receive_interrupt_dma_rtail(struct hfi1_ctxtdata *, int);
+struct hfi1_ctxtdata *hfi1_create_ctxtdata(struct hfi1_pportdata *ppd, u32 ctxt,
+					   int numa);
+void hfi1_init_pportdata(struct pci_dev *pdev, struct hfi1_pportdata *ppd,
+			 struct hfi1_devdata *dd, u8 hw_pidx, u8 port);
+void hfi1_free_ctxtdata(struct hfi1_devdata *dd, struct hfi1_ctxtdata *rcd);
+
+int handle_receive_interrupt(struct hfi1_ctxtdata *rcd, int thread);
+int handle_receive_interrupt_nodma_rtail(struct hfi1_ctxtdata *rcd, int thread);
+int handle_receive_interrupt_dma_rtail(struct hfi1_ctxtdata *rcd, int thread);
 void set_all_slowpath(struct hfi1_devdata *dd);
 void hfi1_vnic_synchronize_irq(struct hfi1_devdata *dd);
 void hfi1_set_vnic_msix_info(struct hfi1_ctxtdata *rcd);
@@ -1580,7 +1581,7 @@ static inline int rcv_pkey_check(struct hfi1_pportdata *ppd, u16 pkey,
 
 u32 lrh_max_header_bytes(struct hfi1_devdata *dd);
 int mtu_to_enum(u32 mtu, int default_if_bad);
-u16 enum_to_mtu(int);
+u16 enum_to_mtu(int mtu);
 static inline int valid_ib_mtu(unsigned int mtu)
 {
 	return mtu == 256 || mtu == 512 ||
@@ -1594,15 +1595,15 @@ static inline int valid_opa_max_mtu(unsigned int mtu)
 		(valid_ib_mtu(mtu) || mtu == 8192 || mtu == 10240);
 }
 
-int set_mtu(struct hfi1_pportdata *);
+int set_mtu(struct hfi1_pportdata *ppd);
 
-int hfi1_set_lid(struct hfi1_pportdata *, u32, u8);
-void hfi1_disable_after_error(struct hfi1_devdata *);
-int hfi1_set_uevent_bits(struct hfi1_pportdata *, const int);
-int hfi1_rcvbuf_validate(u32, u8, u16 *);
+int hfi1_set_lid(struct hfi1_pportdata *ppd, u32 lid, u8 lmc);
+void hfi1_disable_after_error(struct hfi1_devdata *dd);
+int hfi1_set_uevent_bits(struct hfi1_pportdata *ppd, const int evtbit);
+int hfi1_rcvbuf_validate(u32 size, u8 type, u16 *encode);
 
-int fm_get_table(struct hfi1_pportdata *, int, void *);
-int fm_set_table(struct hfi1_pportdata *, int, void *);
+int fm_get_table(struct hfi1_pportdata *ppd, int which, void *t);
+int fm_set_table(struct hfi1_pportdata *ppd, int which, void *t);
 
 void set_up_vl15(struct hfi1_devdata *dd, u8 vau, u16 vl15buf);
 void reset_link_credits(struct hfi1_devdata *dd);
@@ -1734,9 +1735,9 @@ struct cc_state *get_cc_state_protected(struct hfi1_pportdata *ppd)
 #define HFI1_CTXT_WAITING_URG 5
 
 /* free up any allocated data at closes */
-struct hfi1_devdata *hfi1_init_dd(struct pci_dev *,
-				  const struct pci_device_id *);
-void hfi1_free_devdata(struct hfi1_devdata *);
+struct hfi1_devdata *hfi1_init_dd(struct pci_dev *pdev,
+				  const struct pci_device_id *ent);
+void hfi1_free_devdata(struct hfi1_devdata *dd);
 struct hfi1_devdata *hfi1_alloc_devdata(struct pci_dev *pdev, size_t extra);
 
 /* LED beaconing functions */
@@ -1811,24 +1812,25 @@ static inline u32 get_rcvhdrtail(const struct hfi1_ctxtdata *rcd)
 
 extern const char ib_hfi1_version[];
 
-int hfi1_device_create(struct hfi1_devdata *);
-void hfi1_device_remove(struct hfi1_devdata *);
+int hfi1_device_create(struct hfi1_devdata *dd);
+void hfi1_device_remove(struct hfi1_devdata *dd);
 
 int hfi1_create_port_files(struct ib_device *ibdev, u8 port_num,
 			   struct kobject *kobj);
-int hfi1_verbs_register_sysfs(struct hfi1_devdata *);
-void hfi1_verbs_unregister_sysfs(struct hfi1_devdata *);
+int hfi1_verbs_register_sysfs(struct hfi1_devdata *dd);
+void hfi1_verbs_unregister_sysfs(struct hfi1_devdata *dd);
 /* Hook for sysfs read of QSFP */
 int qsfp_dump(struct hfi1_pportdata *ppd, char *buf, int len);
 
-int hfi1_pcie_init(struct pci_dev *, const struct pci_device_id *);
-void hfi1_pcie_cleanup(struct pci_dev *);
-int hfi1_pcie_ddinit(struct hfi1_devdata *, struct pci_dev *);
+int hfi1_pcie_init(struct pci_dev *pdev, const struct pci_device_id *ent);
+void hfi1_pcie_cleanup(struct pci_dev *pdev);
+int hfi1_pcie_ddinit(struct hfi1_devdata *dd, struct pci_dev *pdev);
 void hfi1_pcie_ddcleanup(struct hfi1_devdata *);
-void hfi1_pcie_flr(struct hfi1_devdata *);
-int pcie_speeds(struct hfi1_devdata *);
-void request_msix(struct hfi1_devdata *, u32 *, struct hfi1_msix_entry *);
-void hfi1_enable_intx(struct pci_dev *);
+void hfi1_pcie_flr(struct hfi1_devdata *dd);
+int pcie_speeds(struct hfi1_devdata *dd);
+void request_msix(struct hfi1_devdata *dd, u32 *nent,
+		  struct hfi1_msix_entry *entry);
+void hfi1_enable_intx(struct pci_dev *pdev);
 void restore_pci_variables(struct hfi1_devdata *dd);
 int do_pcie_gen3_transition(struct hfi1_devdata *dd);
 int parse_platform_config(struct hfi1_devdata *dd);
diff --git a/drivers/infiniband/hw/hfi1/init.c b/drivers/infiniband/hw/hfi1/init.c
index b2db776..e872644 100644
--- a/drivers/infiniband/hw/hfi1/init.c
+++ b/drivers/infiniband/hw/hfi1/init.c
@@ -118,7 +118,7 @@
 module_param(user_credit_return_threshold, uint, S_IRUGO);
 MODULE_PARM_DESC(user_credit_return_threshold, "Credit return threshold for user send contexts, return when unreturned credits passes this many blocks (in percent of allocated blocks, 0 is off)");
 
-static inline u64 encode_rcv_header_entry_size(u16);
+static inline u64 encode_rcv_header_entry_size(u16 size);
 
 static struct idr hfi1_unit_table;
 u32 hfi1_cpulist_count;
diff --git a/drivers/infiniband/hw/hfi1/user_exp_rcv.c b/drivers/infiniband/hw/hfi1/user_exp_rcv.c
index c7f13df..6090933 100644
--- a/drivers/infiniband/hw/hfi1/user_exp_rcv.c
+++ b/drivers/infiniband/hw/hfi1/user_exp_rcv.c
@@ -82,20 +82,26 @@ struct tid_pageset {
 		 (unsigned long)(len) - 1) & PAGE_MASK) -	       \
 	       ((unsigned long)vaddr & PAGE_MASK)) >> PAGE_SHIFT))
 
-static void unlock_exp_tids(struct hfi1_ctxtdata *, struct exp_tid_set *,
-			    struct hfi1_filedata *);
-static u32 find_phys_blocks(struct page **, unsigned, struct tid_pageset *);
-static int set_rcvarray_entry(struct file *, unsigned long, u32,
-			      struct tid_group *, struct page **, unsigned);
-static int tid_rb_insert(void *, struct mmu_rb_node *);
+static void unlock_exp_tids(struct hfi1_ctxtdata *uctxt,
+			    struct exp_tid_set *set,
+			    struct hfi1_filedata *fd);
+static u32 find_phys_blocks(struct page **pages, unsigned npages,
+			    struct tid_pageset *list);
+static int set_rcvarray_entry(struct file *fp, unsigned long vaddr,
+			      u32 rcventry, struct tid_group *grp,
+			      struct page **pages, unsigned npages);
+static int tid_rb_insert(void *arg, struct mmu_rb_node *node);
 static void cacheless_tid_rb_remove(struct hfi1_filedata *fdata,
 				    struct tid_rb_node *tnode);
-static void tid_rb_remove(void *, struct mmu_rb_node *);
-static int tid_rb_invalidate(void *, struct mmu_rb_node *);
-static int program_rcvarray(struct file *, unsigned long, struct tid_group *,
-			    struct tid_pageset *, unsigned, u16, struct page **,
-			    u32 *, unsigned *, unsigned *);
-static int unprogram_rcvarray(struct file *, u32, struct tid_group **);
+static void tid_rb_remove(void *arg, struct mmu_rb_node *node);
+static int tid_rb_invalidate(void *arg, struct mmu_rb_node *mnode);
+static int program_rcvarray(struct file *fp, unsigned long vaddr,
+			    struct tid_group *grp,
+			    struct tid_pageset *sets,
+			    unsigned start, u16 count, struct page **pages,
+			    u32 *tidlist, unsigned *tididx, unsigned *pmapped);
+static int unprogram_rcvarray(struct file *fp, u32 tidinfo,
+			      struct tid_group **grp);
 static void clear_tid_node(struct hfi1_filedata *fd, struct tid_rb_node *node);
 
 static struct mmu_rb_ops tid_rb_ops = {
diff --git a/drivers/infiniband/hw/hfi1/user_exp_rcv.h b/drivers/infiniband/hw/hfi1/user_exp_rcv.h
index d1d7d3d..8a33a09 100644
--- a/drivers/infiniband/hw/hfi1/user_exp_rcv.h
+++ b/drivers/infiniband/hw/hfi1/user_exp_rcv.h
@@ -1,7 +1,7 @@
 #ifndef _HFI1_USER_EXP_RCV_H
 #define _HFI1_USER_EXP_RCV_H
 /*
- * Copyright(c) 2015, 2016 Intel Corporation.
+ * Copyright(c) 2015 - 2017 Intel Corporation.
  *
  * This file is provided under a dual BSD/GPLv2 license.  When using or
  * redistributing this file, you may do so under either license.
@@ -71,10 +71,10 @@
 	} while (0)
 
 void hfi1_user_exp_rcv_grp_free(struct hfi1_ctxtdata *uctxt);
-int hfi1_user_exp_rcv_init(struct file *);
-int hfi1_user_exp_rcv_free(struct hfi1_filedata *);
-int hfi1_user_exp_rcv_setup(struct file *, struct hfi1_tid_info *);
-int hfi1_user_exp_rcv_clear(struct file *, struct hfi1_tid_info *);
-int hfi1_user_exp_rcv_invalid(struct file *, struct hfi1_tid_info *);
+int hfi1_user_exp_rcv_init(struct file *fp);
+int hfi1_user_exp_rcv_free(struct hfi1_filedata *fd);
+int hfi1_user_exp_rcv_setup(struct file *fp, struct hfi1_tid_info *tinfo);
+int hfi1_user_exp_rcv_clear(struct file *fp, struct hfi1_tid_info *tinfo);
+int hfi1_user_exp_rcv_invalid(struct file *fp, struct hfi1_tid_info *tinfo);
 
 #endif /* _HFI1_USER_EXP_RCV_H */
diff --git a/drivers/infiniband/hw/hfi1/user_sdma.c b/drivers/infiniband/hw/hfi1/user_sdma.c
index 8adb6df..17142c5 100644
--- a/drivers/infiniband/hw/hfi1/user_sdma.c
+++ b/drivers/infiniband/hw/hfi1/user_sdma.c
@@ -1,5 +1,5 @@
 /*
- * Copyright(c) 2015, 2016 Intel Corporation.
+ * Copyright(c) 2015 - 2017 Intel Corporation.
  *
  * This file is provided under a dual BSD/GPLv2 license.  When using or
  * redistributing this file, you may do so under either license.
@@ -280,39 +280,43 @@ struct user_sdma_txreq {
 	hfi1_cdbg(SDMA, "[%u:%u:%u] " fmt, (pq)->dd->unit, (pq)->ctxt, \
 		 (pq)->subctxt, ##__VA_ARGS__)
 
-static int user_sdma_send_pkts(struct user_sdma_request *, unsigned);
-static int num_user_pages(const struct iovec *);
-static void user_sdma_txreq_cb(struct sdma_txreq *, int);
-static inline void pq_update(struct hfi1_user_sdma_pkt_q *);
-static void user_sdma_free_request(struct user_sdma_request *, bool);
-static int pin_vector_pages(struct user_sdma_request *,
-			    struct user_sdma_iovec *);
-static void unpin_vector_pages(struct mm_struct *, struct page **, unsigned,
-			       unsigned);
-static int check_header_template(struct user_sdma_request *,
-				 struct hfi1_pkt_header *, u32, u32);
-static int set_txreq_header(struct user_sdma_request *,
-			    struct user_sdma_txreq *, u32);
-static int set_txreq_header_ahg(struct user_sdma_request *,
-				struct user_sdma_txreq *, u32);
-static inline void set_comp_state(struct hfi1_user_sdma_pkt_q *,
-				  struct hfi1_user_sdma_comp_q *,
-				  u16, enum hfi1_sdma_comp_state, int);
-static inline u32 set_pkt_bth_psn(__be32, u8, u32);
+static int user_sdma_send_pkts(struct user_sdma_request *req,
+			       unsigned maxpkts);
+static int num_user_pages(const struct iovec *iov);
+static void user_sdma_txreq_cb(struct sdma_txreq *txreq, int status);
+static inline void pq_update(struct hfi1_user_sdma_pkt_q *pq);
+static void user_sdma_free_request(struct user_sdma_request *req, bool unpin);
+static int pin_vector_pages(struct user_sdma_request *req,
+			    struct user_sdma_iovec *iovec);
+static void unpin_vector_pages(struct mm_struct *mm, struct page **pages,
+			       unsigned start, unsigned npages);
+static int check_header_template(struct user_sdma_request *req,
+				 struct hfi1_pkt_header *hdr, u32 lrhlen,
+				 u32 datalen);
+static int set_txreq_header(struct user_sdma_request *req,
+			    struct user_sdma_txreq *tx, u32 datalen);
+static int set_txreq_header_ahg(struct user_sdma_request *req,
+				struct user_sdma_txreq *tx, u32 len);
+static inline void set_comp_state(struct hfi1_user_sdma_pkt_q *pq,
+				  struct hfi1_user_sdma_comp_q *cq,
+				  u16 idx, enum hfi1_sdma_comp_state state,
+				  int ret);
+static inline u32 set_pkt_bth_psn(__be32 bthpsn, u8 expct, u32 frags);
 static inline u32 get_lrh_len(struct hfi1_pkt_header, u32 len);
 
 static int defer_packet_queue(
-	struct sdma_engine *,
-	struct iowait *,
-	struct sdma_txreq *,
-	unsigned seq);
-static void activate_packet_queue(struct iowait *, int);
-static bool sdma_rb_filter(struct mmu_rb_node *, unsigned long, unsigned long);
-static int sdma_rb_insert(void *, struct mmu_rb_node *);
+	struct sdma_engine *sde,
+	struct iowait *wait,
+	struct sdma_txreq *txreq,
+	unsigned int seq);
+static void activate_packet_queue(struct iowait *wait, int reason);
+static bool sdma_rb_filter(struct mmu_rb_node *node, unsigned long addr,
+			   unsigned long len);
+static int sdma_rb_insert(void *arg, struct mmu_rb_node *mnode);
 static int sdma_rb_evict(void *arg, struct mmu_rb_node *mnode,
 			 void *arg2, bool *stop);
-static void sdma_rb_remove(void *, struct mmu_rb_node *);
-static int sdma_rb_invalidate(void *, struct mmu_rb_node *);
+static void sdma_rb_remove(void *arg, struct mmu_rb_node *mnode);
+static int sdma_rb_invalidate(void *arg, struct mmu_rb_node *mnode);
 
 static struct mmu_rb_ops sdma_rb_ops = {
 	.filter = sdma_rb_filter,
diff --git a/drivers/infiniband/hw/hfi1/user_sdma.h b/drivers/infiniband/hw/hfi1/user_sdma.h
index 3900171..95ec783 100644
--- a/drivers/infiniband/hw/hfi1/user_sdma.h
+++ b/drivers/infiniband/hw/hfi1/user_sdma.h
@@ -1,5 +1,7 @@
+#ifndef _HFI1_USER_SDMA_H
+#define _HFI1_USER_SDMA_H
 /*
- * Copyright(c) 2015, 2016 Intel Corporation.
+ * Copyright(c) 2015 - 2017 Intel Corporation.
  *
  * This file is provided under a dual BSD/GPLv2 license.  When using or
  * redistributing this file, you may do so under either license.
@@ -78,7 +80,8 @@ struct hfi1_user_sdma_comp_q {
 	struct hfi1_sdma_comp_entry *comps;
 };
 
-int hfi1_user_sdma_alloc_queues(struct hfi1_ctxtdata *, struct file *);
-int hfi1_user_sdma_free_queues(struct hfi1_filedata *);
-int hfi1_user_sdma_process_request(struct file *, struct iovec *, unsigned long,
-				   unsigned long *);
+int hfi1_user_sdma_alloc_queues(struct hfi1_ctxtdata *uctxt, struct file *fp);
+int hfi1_user_sdma_free_queues(struct hfi1_filedata *fd);
+int hfi1_user_sdma_process_request(struct file *fp, struct iovec *iovec,
+				   unsigned long dim, unsigned long *count);
+#endif /* _HFI1_USER_SDMA_H */

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH 09/15] IB/hfi1: Use filedata rather than filepointer

[Dennis Dalessandro]

From: Michael J. Ruhl <michael.j.ruhl-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

Since almost all functions that use the hfi1_filedata get the pointer
from the file pointer, simplify by only passing the hfi1_filedata pointer.

Reviewed-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Reviewed-by: Ira Weiny <ira.weiny-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/infiniband/hw/hfi1/file_ops.c     |   68 ++++++++++++++---------------
 drivers/infiniband/hw/hfi1/user_exp_rcv.c |   45 +++++++++----------
 drivers/infiniband/hw/hfi1/user_exp_rcv.h |   11 +++--
 drivers/infiniband/hw/hfi1/user_sdma.c    |   14 +++---
 drivers/infiniband/hw/hfi1/user_sdma.h    |    9 +++-
 5 files changed, 72 insertions(+), 75 deletions(-)

diff --git a/drivers/infiniband/hw/hfi1/file_ops.c b/drivers/infiniband/hw/hfi1/file_ops.c
index 0b77283..ab798a8 100644
--- a/drivers/infiniband/hw/hfi1/file_ops.c
+++ b/drivers/infiniband/hw/hfi1/file_ops.c
@@ -80,16 +80,18 @@
 static int assign_ctxt(struct file *fp, struct hfi1_user_info *uinfo);
 static int init_subctxts(struct hfi1_ctxtdata *uctxt,
 			 const struct hfi1_user_info *uinfo);
-static int user_init(struct file *fp);
-static int get_ctxt_info(struct file *fp, void __user *ubase, __u32 len);
-static int get_base_info(struct file *fp, void __user *ubase, __u32 len);
-static int setup_ctxt(struct file *fp);
+static int user_init(struct hfi1_filedata *fd);
+static int get_ctxt_info(struct hfi1_filedata *fd, void __user *ubase,
+			 __u32 len);
+static int get_base_info(struct hfi1_filedata *fd, void __user *ubase,
+			 __u32 len);
+static int setup_ctxt(struct hfi1_filedata *fd);
 static int setup_subctxt(struct hfi1_ctxtdata *uctxt);
-static int get_user_context(struct file *fp, struct hfi1_user_info *uinfo,
-			    int devno);
-static int find_shared_ctxt(struct file *fp,
+static int get_user_context(struct hfi1_filedata *fd,
+			    struct hfi1_user_info *uinfo, int devno);
+static int find_shared_ctxt(struct hfi1_filedata *fd,
 			    const struct hfi1_user_info *uinfo);
-static int allocate_ctxt(struct file *fp, struct hfi1_devdata *dd,
+static int allocate_ctxt(struct hfi1_filedata *fd, struct hfi1_devdata *dd,
 			 struct hfi1_user_info *uinfo);
 static unsigned int poll_urgent(struct file *fp, struct poll_table_struct *pt);
 static unsigned int poll_next(struct file *fp, struct poll_table_struct *pt);
@@ -238,17 +240,17 @@ static long hfi1_file_ioctl(struct file *fp, unsigned int cmd,
 		ret = assign_ctxt(fp, &uinfo);
 		if (ret < 0)
 			return ret;
-		ret = setup_ctxt(fp);
+		ret = setup_ctxt(fd);
 		if (ret)
 			return ret;
-		ret = user_init(fp);
+		ret = user_init(fd);
 		break;
 	case HFI1_IOCTL_CTXT_INFO:
-		ret = get_ctxt_info(fp, (void __user *)(unsigned long)arg,
+		ret = get_ctxt_info(fd, (void __user *)(unsigned long)arg,
 				    sizeof(struct hfi1_ctxt_info));
 		break;
 	case HFI1_IOCTL_USER_INFO:
-		ret = get_base_info(fp, (void __user *)(unsigned long)arg,
+		ret = get_base_info(fd, (void __user *)(unsigned long)arg,
 				    sizeof(struct hfi1_base_info));
 		break;
 	case HFI1_IOCTL_CREDIT_UPD:
@@ -262,7 +264,7 @@ static long hfi1_file_ioctl(struct file *fp, unsigned int cmd,
 				   sizeof(tinfo)))
 			return -EFAULT;
 
-		ret = hfi1_user_exp_rcv_setup(fp, &tinfo);
+		ret = hfi1_user_exp_rcv_setup(fd, &tinfo);
 		if (!ret) {
 			/*
 			 * Copy the number of tidlist entries we used
@@ -284,7 +286,7 @@ static long hfi1_file_ioctl(struct file *fp, unsigned int cmd,
 				   sizeof(tinfo)))
 			return -EFAULT;
 
-		ret = hfi1_user_exp_rcv_clear(fp, &tinfo);
+		ret = hfi1_user_exp_rcv_clear(fd, &tinfo);
 		if (ret)
 			break;
 		addr = arg + offsetof(struct hfi1_tid_info, tidcnt);
@@ -299,7 +301,7 @@ static long hfi1_file_ioctl(struct file *fp, unsigned int cmd,
 				   sizeof(tinfo)))
 			return -EFAULT;
 
-		ret = hfi1_user_exp_rcv_invalid(fp, &tinfo);
+		ret = hfi1_user_exp_rcv_invalid(fd, &tinfo);
 		if (ret)
 			break;
 		addr = arg + offsetof(struct hfi1_tid_info, tidcnt);
@@ -436,7 +438,7 @@ static ssize_t hfi1_write_iter(struct kiocb *kiocb, struct iov_iter *from)
 		unsigned long count = 0;
 
 		ret = hfi1_user_sdma_process_request(
-			kiocb->ki_filp,	(struct iovec *)(from->iov + done),
+			fd, (struct iovec *)(from->iov + done),
 			dim, &count);
 		if (ret) {
 			reqs = ret;
@@ -863,7 +865,7 @@ static int assign_ctxt(struct file *fp, struct hfi1_user_info *uinfo)
 	if (uinfo->subctxt_cnt) {
 		struct hfi1_filedata *fd = fp->private_data;
 
-		ret = find_shared_ctxt(fp, uinfo);
+		ret = find_shared_ctxt(fd, uinfo);
 		if (ret < 0)
 			goto done_unlock;
 		if (ret) {
@@ -878,7 +880,7 @@ static int assign_ctxt(struct file *fp, struct hfi1_user_info *uinfo)
 	 */
 	if (!ret) {
 		i_minor = iminor(file_inode(fp)) - HFI1_USER_MINOR_BASE;
-		ret = get_user_context(fp, uinfo, i_minor);
+		ret = get_user_context(fp->private_data, uinfo, i_minor);
 	}
 done_unlock:
 	mutex_unlock(&hfi1_mutex);
@@ -886,8 +888,8 @@ static int assign_ctxt(struct file *fp, struct hfi1_user_info *uinfo)
 	return ret;
 }
 
-static int get_user_context(struct file *fp, struct hfi1_user_info *uinfo,
-			    int devno)
+static int get_user_context(struct hfi1_filedata *fd,
+			    struct hfi1_user_info *uinfo, int devno)
 {
 	struct hfi1_devdata *dd = NULL;
 	int devmax, npresent, nup;
@@ -905,15 +907,14 @@ static int get_user_context(struct file *fp, struct hfi1_user_info *uinfo,
 	else if (!dd->freectxts)
 		return -EBUSY;
 
-	return allocate_ctxt(fp, dd, uinfo);
+	return allocate_ctxt(fd, dd, uinfo);
 }
 
-static int find_shared_ctxt(struct file *fp,
+static int find_shared_ctxt(struct hfi1_filedata *fd,
 			    const struct hfi1_user_info *uinfo)
 {
 	int devmax, ndev, i;
 	int ret = 0;
-	struct hfi1_filedata *fd = fp->private_data;
 
 	devmax = hfi1_count_units(NULL, NULL);
 
@@ -960,10 +961,9 @@ static int find_shared_ctxt(struct file *fp,
 	return ret;
 }
 
-static int allocate_ctxt(struct file *fp, struct hfi1_devdata *dd,
+static int allocate_ctxt(struct hfi1_filedata *fd, struct hfi1_devdata *dd,
 			 struct hfi1_user_info *uinfo)
 {
-	struct hfi1_filedata *fd = fp->private_data;
 	struct hfi1_ctxtdata *uctxt;
 	unsigned ctxt;
 	int ret, numa;
@@ -1113,10 +1113,9 @@ static int setup_subctxt(struct hfi1_ctxtdata *uctxt)
 	return ret;
 }
 
-static int user_init(struct file *fp)
+static int user_init(struct hfi1_filedata *fd)
 {
 	unsigned int rcvctrl_ops = 0;
-	struct hfi1_filedata *fd = fp->private_data;
 	struct hfi1_ctxtdata *uctxt = fd->uctxt;
 
 	/* make sure that the context has already been setup */
@@ -1179,10 +1178,10 @@ static int user_init(struct file *fp)
 	return 0;
 }
 
-static int get_ctxt_info(struct file *fp, void __user *ubase, __u32 len)
+static int get_ctxt_info(struct hfi1_filedata *fd, void __user *ubase,
+			 __u32 len)
 {
 	struct hfi1_ctxt_info cinfo;
-	struct hfi1_filedata *fd = fp->private_data;
 	struct hfi1_ctxtdata *uctxt = fd->uctxt;
 	int ret = 0;
 
@@ -1220,9 +1219,8 @@ static int get_ctxt_info(struct file *fp, void __user *ubase, __u32 len)
 	return ret;
 }
 
-static int setup_ctxt(struct file *fp)
+static int setup_ctxt(struct hfi1_filedata *fd)
 {
-	struct hfi1_filedata *fd = fp->private_data;
 	struct hfi1_ctxtdata *uctxt = fd->uctxt;
 	struct hfi1_devdata *dd = uctxt->dd;
 	int ret = 0;
@@ -1257,7 +1255,7 @@ static int setup_ctxt(struct file *fp)
 			goto done;
 	}
 
-	ret = hfi1_user_sdma_alloc_queues(uctxt, fp);
+	ret = hfi1_user_sdma_alloc_queues(uctxt, fd);
 	if (ret)
 		goto done;
 	/*
@@ -1269,7 +1267,7 @@ static int setup_ctxt(struct file *fp)
 	 * (due to the above wait_event_interruptible() until the master
 	 * is setup.
 	 */
-	ret = hfi1_user_exp_rcv_init(fp);
+	ret = hfi1_user_exp_rcv_init(fd);
 	if (ret)
 		goto done;
 
@@ -1278,10 +1276,10 @@ static int setup_ctxt(struct file *fp)
 	return ret;
 }
 
-static int get_base_info(struct file *fp, void __user *ubase, __u32 len)
+static int get_base_info(struct hfi1_filedata *fd, void __user *ubase,
+			 __u32 len)
 {
 	struct hfi1_base_info binfo;
-	struct hfi1_filedata *fd = fp->private_data;
 	struct hfi1_ctxtdata *uctxt = fd->uctxt;
 	struct hfi1_devdata *dd = uctxt->dd;
 	ssize_t sz;
diff --git a/drivers/infiniband/hw/hfi1/user_exp_rcv.c b/drivers/infiniband/hw/hfi1/user_exp_rcv.c
index 6090933..22ffda8 100644
--- a/drivers/infiniband/hw/hfi1/user_exp_rcv.c
+++ b/drivers/infiniband/hw/hfi1/user_exp_rcv.c
@@ -87,7 +87,7 @@ static void unlock_exp_tids(struct hfi1_ctxtdata *uctxt,
 			    struct hfi1_filedata *fd);
 static u32 find_phys_blocks(struct page **pages, unsigned npages,
 			    struct tid_pageset *list);
-static int set_rcvarray_entry(struct file *fp, unsigned long vaddr,
+static int set_rcvarray_entry(struct hfi1_filedata *fd, unsigned long vaddr,
 			      u32 rcventry, struct tid_group *grp,
 			      struct page **pages, unsigned npages);
 static int tid_rb_insert(void *arg, struct mmu_rb_node *node);
@@ -95,12 +95,11 @@ static void cacheless_tid_rb_remove(struct hfi1_filedata *fdata,
 				    struct tid_rb_node *tnode);
 static void tid_rb_remove(void *arg, struct mmu_rb_node *node);
 static int tid_rb_invalidate(void *arg, struct mmu_rb_node *mnode);
-static int program_rcvarray(struct file *fp, unsigned long vaddr,
-			    struct tid_group *grp,
-			    struct tid_pageset *sets,
+static int program_rcvarray(struct hfi1_filedata *fd, unsigned long vaddr,
+			    struct tid_group *grp, struct tid_pageset *sets,
 			    unsigned start, u16 count, struct page **pages,
 			    u32 *tidlist, unsigned *tididx, unsigned *pmapped);
-static int unprogram_rcvarray(struct file *fp, u32 tidinfo,
+static int unprogram_rcvarray(struct hfi1_filedata *fd, u32 tidinfo,
 			      struct tid_group **grp);
 static void clear_tid_node(struct hfi1_filedata *fd, struct tid_rb_node *node);
 
@@ -160,9 +159,8 @@ static inline void tid_group_move(struct tid_group *group,
  * receive caching. This needs to be done after the context has
  * been configured with the eager/expected RcvEntry counts.
  */
-int hfi1_user_exp_rcv_init(struct file *fp)
+int hfi1_user_exp_rcv_init(struct hfi1_filedata *fd)
 {
-	struct hfi1_filedata *fd = fp->private_data;
 	struct hfi1_ctxtdata *uctxt = fd->uctxt;
 	struct hfi1_devdata *dd = uctxt->dd;
 	unsigned tidbase;
@@ -361,10 +359,10 @@ static inline void rcv_array_wc_fill(struct hfi1_devdata *dd, u32 index)
  *          can fit into the group. If the group becomes fully
  *          used, move it to tid_full_list.
  */
-int hfi1_user_exp_rcv_setup(struct file *fp, struct hfi1_tid_info *tinfo)
+int hfi1_user_exp_rcv_setup(struct hfi1_filedata *fd,
+			    struct hfi1_tid_info *tinfo)
 {
 	int ret = 0, need_group = 0, pinned;
-	struct hfi1_filedata *fd = fp->private_data;
 	struct hfi1_ctxtdata *uctxt = fd->uctxt;
 	struct hfi1_devdata *dd = uctxt->dd;
 	unsigned npages, ngroups, pageidx = 0, pageset_count, npagesets,
@@ -461,7 +459,7 @@ int hfi1_user_exp_rcv_setup(struct file *fp, struct hfi1_tid_info *tinfo)
 		struct tid_group *grp =
 			tid_group_pop(&uctxt->tid_group_list);
 
-		ret = program_rcvarray(fp, vaddr, grp, pagesets,
+		ret = program_rcvarray(fd, vaddr, grp, pagesets,
 				       pageidx, dd->rcv_entries.group_size,
 				       pages, tidlist, &tididx, &mapped);
 		/*
@@ -507,7 +505,7 @@ int hfi1_user_exp_rcv_setup(struct file *fp, struct hfi1_tid_info *tinfo)
 			unsigned use = min_t(unsigned, pageset_count - pageidx,
 					     grp->size - grp->used);
 
-			ret = program_rcvarray(fp, vaddr, grp, pagesets,
+			ret = program_rcvarray(fd, vaddr, grp, pagesets,
 					       pageidx, use, pages, tidlist,
 					       &tididx, &mapped);
 			if (ret < 0) {
@@ -557,7 +555,7 @@ int hfi1_user_exp_rcv_setup(struct file *fp, struct hfi1_tid_info *tinfo)
 			 * everything done so far so we don't leak resources.
 			 */
 			tinfo->tidlist = (unsigned long)&tidlist;
-			hfi1_user_exp_rcv_clear(fp, tinfo);
+			hfi1_user_exp_rcv_clear(fd, tinfo);
 			tinfo->tidlist = 0;
 			ret = -EFAULT;
 			goto bail;
@@ -581,10 +579,10 @@ int hfi1_user_exp_rcv_setup(struct file *fp, struct hfi1_tid_info *tinfo)
 	return ret > 0 ? 0 : ret;
 }
 
-int hfi1_user_exp_rcv_clear(struct file *fp, struct hfi1_tid_info *tinfo)
+int hfi1_user_exp_rcv_clear(struct hfi1_filedata *fd,
+			    struct hfi1_tid_info *tinfo)
 {
 	int ret = 0;
-	struct hfi1_filedata *fd = fp->private_data;
 	struct hfi1_ctxtdata *uctxt = fd->uctxt;
 	u32 *tidinfo;
 	unsigned tididx;
@@ -599,7 +597,7 @@ int hfi1_user_exp_rcv_clear(struct file *fp, struct hfi1_tid_info *tinfo)
 
 	mutex_lock(&uctxt->exp_lock);
 	for (tididx = 0; tididx < tinfo->tidcnt; tididx++) {
-		ret = unprogram_rcvarray(fp, tidinfo[tididx], NULL);
+		ret = unprogram_rcvarray(fd, tidinfo[tididx], NULL);
 		if (ret) {
 			hfi1_cdbg(TID, "Failed to unprogram rcv array %d",
 				  ret);
@@ -616,9 +614,9 @@ int hfi1_user_exp_rcv_clear(struct file *fp, struct hfi1_tid_info *tinfo)
 	return ret;
 }
 
-int hfi1_user_exp_rcv_invalid(struct file *fp, struct hfi1_tid_info *tinfo)
+int hfi1_user_exp_rcv_invalid(struct hfi1_filedata *fd,
+			      struct hfi1_tid_info *tinfo)
 {
-	struct hfi1_filedata *fd = fp->private_data;
 	struct hfi1_ctxtdata *uctxt = fd->uctxt;
 	unsigned long *ev = uctxt->dd->events +
 		(((uctxt->ctxt - uctxt->dd->first_dyn_alloc_ctxt) *
@@ -733,7 +731,7 @@ static u32 find_phys_blocks(struct page **pages, unsigned npages,
 
 /**
  * program_rcvarray() - program an RcvArray group with receive buffers
- * @fp: file pointer
+ * @fd: filedata pointer
  * @vaddr: starting user virtual address
  * @grp: RcvArray group
  * @sets: array of struct tid_pageset holding information on physically
@@ -758,13 +756,12 @@ static u32 find_phys_blocks(struct page **pages, unsigned npages,
  * -ENOMEM or -EFAULT on error from set_rcvarray_entry(), or
  * number of RcvArray entries programmed.
  */
-static int program_rcvarray(struct file *fp, unsigned long vaddr,
+static int program_rcvarray(struct hfi1_filedata *fd, unsigned long vaddr,
 			    struct tid_group *grp,
 			    struct tid_pageset *sets,
 			    unsigned start, u16 count, struct page **pages,
 			    u32 *tidlist, unsigned *tididx, unsigned *pmapped)
 {
-	struct hfi1_filedata *fd = fp->private_data;
 	struct hfi1_ctxtdata *uctxt = fd->uctxt;
 	struct hfi1_devdata *dd = uctxt->dd;
 	u16 idx;
@@ -805,7 +802,7 @@ static int program_rcvarray(struct file *fp, unsigned long vaddr,
 		npages = sets[setidx].count;
 		pageidx = sets[setidx].idx;
 
-		ret = set_rcvarray_entry(fp, vaddr + (pageidx * PAGE_SIZE),
+		ret = set_rcvarray_entry(fd, vaddr + (pageidx * PAGE_SIZE),
 					 rcventry, grp, pages + pageidx,
 					 npages);
 		if (ret)
@@ -827,12 +824,11 @@ static int program_rcvarray(struct file *fp, unsigned long vaddr,
 	return idx;
 }
 
-static int set_rcvarray_entry(struct file *fp, unsigned long vaddr,
+static int set_rcvarray_entry(struct hfi1_filedata *fd, unsigned long vaddr,
 			      u32 rcventry, struct tid_group *grp,
 			      struct page **pages, unsigned npages)
 {
 	int ret;
-	struct hfi1_filedata *fd = fp->private_data;
 	struct hfi1_ctxtdata *uctxt = fd->uctxt;
 	struct tid_rb_node *node;
 	struct hfi1_devdata *dd = uctxt->dd;
@@ -886,10 +882,9 @@ static int set_rcvarray_entry(struct file *fp, unsigned long vaddr,
 	return 0;
 }
 
-static int unprogram_rcvarray(struct file *fp, u32 tidinfo,
+static int unprogram_rcvarray(struct hfi1_filedata *fd, u32 tidinfo,
 			      struct tid_group **grp)
 {
-	struct hfi1_filedata *fd = fp->private_data;
 	struct hfi1_ctxtdata *uctxt = fd->uctxt;
 	struct hfi1_devdata *dd = uctxt->dd;
 	struct tid_rb_node *node;
diff --git a/drivers/infiniband/hw/hfi1/user_exp_rcv.h b/drivers/infiniband/hw/hfi1/user_exp_rcv.h
index 8a33a09..9787511 100644
--- a/drivers/infiniband/hw/hfi1/user_exp_rcv.h
+++ b/drivers/infiniband/hw/hfi1/user_exp_rcv.h
@@ -71,10 +71,13 @@
 	} while (0)
 
 void hfi1_user_exp_rcv_grp_free(struct hfi1_ctxtdata *uctxt);
-int hfi1_user_exp_rcv_init(struct file *fp);
+int hfi1_user_exp_rcv_init(struct hfi1_filedata *fd);
 int hfi1_user_exp_rcv_free(struct hfi1_filedata *fd);
-int hfi1_user_exp_rcv_setup(struct file *fp, struct hfi1_tid_info *tinfo);
-int hfi1_user_exp_rcv_clear(struct file *fp, struct hfi1_tid_info *tinfo);
-int hfi1_user_exp_rcv_invalid(struct file *fp, struct hfi1_tid_info *tinfo);
+int hfi1_user_exp_rcv_setup(struct hfi1_filedata *fd,
+			    struct hfi1_tid_info *tinfo);
+int hfi1_user_exp_rcv_clear(struct hfi1_filedata *fd,
+			    struct hfi1_tid_info *tinfo);
+int hfi1_user_exp_rcv_invalid(struct hfi1_filedata *fd,
+			      struct hfi1_tid_info *tinfo);
 
 #endif /* _HFI1_USER_EXP_RCV_H */
diff --git a/drivers/infiniband/hw/hfi1/user_sdma.c b/drivers/infiniband/hw/hfi1/user_sdma.c
index 17142c5..4ac4e9d 100644
--- a/drivers/infiniband/hw/hfi1/user_sdma.c
+++ b/drivers/infiniband/hw/hfi1/user_sdma.c
@@ -372,9 +372,9 @@ static void sdma_kmem_cache_ctor(void *obj)
 	memset(tx, 0, sizeof(*tx));
 }
 
-int hfi1_user_sdma_alloc_queues(struct hfi1_ctxtdata *uctxt, struct file *fp)
+int hfi1_user_sdma_alloc_queues(struct hfi1_ctxtdata *uctxt,
+				struct hfi1_filedata *fd)
 {
-	struct hfi1_filedata *fd;
 	int ret = 0;
 	char buf[64];
 	struct hfi1_devdata *dd;
@@ -382,13 +382,11 @@ int hfi1_user_sdma_alloc_queues(struct hfi1_ctxtdata *uctxt, struct file *fp)
 	struct hfi1_user_sdma_pkt_q *pq;
 	unsigned long flags;
 
-	if (!uctxt || !fp) {
+	if (!uctxt || !fd) {
 		ret = -EBADF;
 		goto done;
 	}
 
-	fd = fp->private_data;
-
 	if (!hfi1_sdma_comp_ring_size) {
 		ret = -EINVAL;
 		goto done;
@@ -536,11 +534,11 @@ static u8 dlid_to_selector(u16 dlid)
 	return mapping[hash];
 }
 
-int hfi1_user_sdma_process_request(struct file *fp, struct iovec *iovec,
-				   unsigned long dim, unsigned long *count)
+int hfi1_user_sdma_process_request(struct hfi1_filedata *fd,
+				   struct iovec *iovec, unsigned long dim,
+				   unsigned long *count)
 {
 	int ret = 0, i;
-	struct hfi1_filedata *fd = fp->private_data;
 	struct hfi1_ctxtdata *uctxt = fd->uctxt;
 	struct hfi1_user_sdma_pkt_q *pq = fd->pq;
 	struct hfi1_user_sdma_comp_q *cq = fd->cq;
diff --git a/drivers/infiniband/hw/hfi1/user_sdma.h b/drivers/infiniband/hw/hfi1/user_sdma.h
index 95ec783..9181d7c 100644
--- a/drivers/infiniband/hw/hfi1/user_sdma.h
+++ b/drivers/infiniband/hw/hfi1/user_sdma.h
@@ -80,8 +80,11 @@ struct hfi1_user_sdma_comp_q {
 	struct hfi1_sdma_comp_entry *comps;
 };
 
-int hfi1_user_sdma_alloc_queues(struct hfi1_ctxtdata *uctxt, struct file *fp);
+int hfi1_user_sdma_alloc_queues(struct hfi1_ctxtdata *uctxt,
+				struct hfi1_filedata *fd);
 int hfi1_user_sdma_free_queues(struct hfi1_filedata *fd);
-int hfi1_user_sdma_process_request(struct file *fp, struct iovec *iovec,
-				   unsigned long dim, unsigned long *count);
+int hfi1_user_sdma_process_request(struct hfi1_filedata *fd,
+				   struct iovec *iovec, unsigned long dim,
+				   unsigned long *count);
+
 #endif /* _HFI1_USER_SDMA_H */

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH 10/15] IB/hfi1: Remove atomic operations for SDMA_REQ_HAVE_AHG bit

[Dennis Dalessandro]

From: Sebastian Sanchez <sebastian.sanchez-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

The AHG index is only accessed in the request call
from user space, so there's no need for atomic semantics.
Replace atomic operations for SDMA_REQ_HAVE_AHG bit
with a test of the AHG index.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Sebastian Sanchez <sebastian.sanchez-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/infiniband/hw/hfi1/user_sdma.c |   22 ++++++++--------------
 1 files changed, 8 insertions(+), 14 deletions(-)

diff --git a/drivers/infiniband/hw/hfi1/user_sdma.c b/drivers/infiniband/hw/hfi1/user_sdma.c
index 4ac4e9d..6b72267 100644
--- a/drivers/infiniband/hw/hfi1/user_sdma.c
+++ b/drivers/infiniband/hw/hfi1/user_sdma.c
@@ -155,9 +155,8 @@
 /* SDMA request flag bits */
 #define SDMA_REQ_FOR_THREAD 1
 #define SDMA_REQ_SEND_DONE  2
-#define SDMA_REQ_HAVE_AHG   3
-#define SDMA_REQ_HAS_ERROR  4
-#define SDMA_REQ_DONE_ERROR 5
+#define SDMA_REQ_HAS_ERROR  3
+#define SDMA_REQ_DONE_ERROR 4
 
 #define SDMA_PKT_Q_INACTIVE BIT(0)
 #define SDMA_PKT_Q_ACTIVE   BIT(1)
@@ -216,7 +215,7 @@ struct user_sdma_request {
 	 * each request will need it's own engine pointer.
 	 */
 	struct sdma_engine *sde;
-	u8 ahg_idx;
+	s8 ahg_idx;
 	u32 ahg[9];
 	/*
 	 * KDETH.Offset (Eager) field
@@ -614,6 +613,7 @@ int hfi1_user_sdma_process_request(struct hfi1_filedata *fd,
 	req->pq = pq;
 	req->cq = cq;
 	req->status = -1;
+	req->ahg_idx = -1;
 	INIT_LIST_HEAD(&req->txps);
 
 	memcpy(&req->info, &info, sizeof(info));
@@ -764,14 +764,8 @@ int hfi1_user_sdma_process_request(struct hfi1_filedata *fd,
 	}
 
 	/* We don't need an AHG entry if the request contains only one packet */
-	if (req->info.npkts > 1 && HFI1_CAP_IS_USET(SDMA_AHG)) {
-		int ahg = sdma_ahg_alloc(req->sde);
-
-		if (likely(ahg >= 0)) {
-			req->ahg_idx = (u8)ahg;
-			set_bit(SDMA_REQ_HAVE_AHG, &req->flags);
-		}
-	}
+	if (req->info.npkts > 1 && HFI1_CAP_IS_USET(SDMA_AHG))
+		req->ahg_idx = sdma_ahg_alloc(req->sde);
 
 	set_comp_state(pq, cq, info.comp_idx, QUEUED, 0);
 	atomic_inc(&pq->n_reqs);
@@ -989,7 +983,7 @@ static int user_sdma_send_pkts(struct user_sdma_request *req, unsigned maxpkts)
 			}
 		}
 
-		if (test_bit(SDMA_REQ_HAVE_AHG, &req->flags)) {
+		if (req->ahg_idx >= 0) {
 			if (!req->seqnum) {
 				u16 pbclen = le16_to_cpu(req->hdr.pbc[0]);
 				u32 lrhlen = get_lrh_len(req->hdr,
@@ -1119,7 +1113,7 @@ static int user_sdma_send_pkts(struct user_sdma_request *req, unsigned maxpkts)
 		 * happen due to the sequential manner in which
 		 * descriptors are processed.
 		 */
-		if (test_bit(SDMA_REQ_HAVE_AHG, &req->flags))
+		if (req->ahg_idx >= 0)
 			sdma_ahg_free(req->sde, req->ahg_idx);
 	}
 	return ret;

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH 11/15] IB/hfi1: Search shared contexts on the opened device, not all devices

[Dennis Dalessandro]

From: Michael J. Ruhl <michael.j.ruhl-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

The search for available shared contexts walks each registered hfi1
device.  This search is too broad because other devices may not
be on the same fabric, and using its contexts could cause unexpected
behavior.

Removed walking the list of devices, limiting the search to the opened
device.  With the device walk removed, the hfi1_devdata (dd) is not
available. Added it to the hfi1_filedata for reference.

With this change, hfi1_count_units() was rendered obsolete and was
removed.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/infiniband/hw/hfi1/driver.c   |   36 -------------
 drivers/infiniband/hw/hfi1/file_ops.c |   94 ++++++++++++++++++---------------
 drivers/infiniband/hw/hfi1/hfi.h      |    2 -
 3 files changed, 52 insertions(+), 80 deletions(-)

diff --git a/drivers/infiniband/hw/hfi1/driver.c b/drivers/infiniband/hw/hfi1/driver.c
index fb319d2..566d152 100644
--- a/drivers/infiniband/hw/hfi1/driver.c
+++ b/drivers/infiniband/hw/hfi1/driver.c
@@ -211,42 +211,6 @@ int hfi1_count_active_units(void)
 }
 
 /*
- * Return count of all units, optionally return in arguments
- * the number of usable (present) units, and the number of
- * ports that are up.
- */
-int hfi1_count_units(int *npresentp, int *nupp)
-{
-	int nunits = 0, npresent = 0, nup = 0;
-	struct hfi1_devdata *dd;
-	unsigned long flags;
-	int pidx;
-	struct hfi1_pportdata *ppd;
-
-	spin_lock_irqsave(&hfi1_devs_lock, flags);
-
-	list_for_each_entry(dd, &hfi1_dev_list, list) {
-		nunits++;
-		if ((dd->flags & HFI1_PRESENT) && dd->kregbase)
-			npresent++;
-		for (pidx = 0; pidx < dd->num_pports; ++pidx) {
-			ppd = dd->pport + pidx;
-			if (ppd->lid && ppd->linkup)
-				nup++;
-		}
-	}
-
-	spin_unlock_irqrestore(&hfi1_devs_lock, flags);
-
-	if (npresentp)
-		*npresentp = npresent;
-	if (nupp)
-		*nupp = nup;
-
-	return nunits;
-}
-
-/*
  * Get address of eager buffer from it's index (allocated in chunks, not
  * contiguous).
  */
diff --git a/drivers/infiniband/hw/hfi1/file_ops.c b/drivers/infiniband/hw/hfi1/file_ops.c
index ab798a8..9fd4b0d 100644
--- a/drivers/infiniband/hw/hfi1/file_ops.c
+++ b/drivers/infiniband/hw/hfi1/file_ops.c
@@ -77,7 +77,7 @@
 static int hfi1_file_mmap(struct file *fp, struct vm_area_struct *vma);
 
 static u64 kvirt_to_phys(void *addr);
-static int assign_ctxt(struct file *fp, struct hfi1_user_info *uinfo);
+static int assign_ctxt(struct hfi1_filedata *fd, struct hfi1_user_info *uinfo);
 static int init_subctxts(struct hfi1_ctxtdata *uctxt,
 			 const struct hfi1_user_info *uinfo);
 static int user_init(struct hfi1_filedata *fd);
@@ -87,8 +87,7 @@ static int get_base_info(struct hfi1_filedata *fd, void __user *ubase,
 			 __u32 len);
 static int setup_ctxt(struct hfi1_filedata *fd);
 static int setup_subctxt(struct hfi1_ctxtdata *uctxt);
-static int get_user_context(struct hfi1_filedata *fd,
-			    struct hfi1_user_info *uinfo, int devno);
+
 static int find_shared_ctxt(struct hfi1_filedata *fd,
 			    const struct hfi1_user_info *uinfo);
 static int allocate_ctxt(struct hfi1_filedata *fd, struct hfi1_devdata *dd,
@@ -181,6 +180,9 @@ static int hfi1_file_open(struct inode *inode, struct file *fp)
 					       struct hfi1_devdata,
 					       user_cdev);
 
+	if (!((dd->flags & HFI1_PRESENT) && dd->kregbase))
+		return -EINVAL;
+
 	if (!atomic_inc_not_zero(&dd->user_refcount))
 		return -ENXIO;
 
@@ -195,6 +197,7 @@ static int hfi1_file_open(struct inode *inode, struct file *fp)
 		fd->rec_cpu_num = -1; /* no cpu affinity by default */
 		fd->mm = current->mm;
 		mmgrab(fd->mm);
+		fd->dd = dd;
 		fp->private_data = fd;
 	} else {
 		fp->private_data = NULL;
@@ -237,7 +240,7 @@ static long hfi1_file_ioctl(struct file *fp, unsigned int cmd,
 				   sizeof(uinfo)))
 			return -EFAULT;
 
-		ret = assign_ctxt(fp, &uinfo);
+		ret = assign_ctxt(fd, &uinfo);
 		if (ret < 0)
 			return ret;
 		ret = setup_ctxt(fd);
@@ -847,9 +850,9 @@ static u64 kvirt_to_phys(void *addr)
 	return paddr;
 }
 
-static int assign_ctxt(struct file *fp, struct hfi1_user_info *uinfo)
+static int assign_ctxt(struct hfi1_filedata *fd, struct hfi1_user_info *uinfo)
 {
-	int i_minor, ret = 0;
+	int ret = 0;
 	unsigned int swmajor, swminor;
 
 	swmajor = uinfo->userversion >> 16;
@@ -863,8 +866,6 @@ static int assign_ctxt(struct file *fp, struct hfi1_user_info *uinfo)
 	mutex_lock(&hfi1_mutex);
 	/* First, lets check if we need to setup a shared context? */
 	if (uinfo->subctxt_cnt) {
-		struct hfi1_filedata *fd = fp->private_data;
-
 		ret = find_shared_ctxt(fd, uinfo);
 		if (ret < 0)
 			goto done_unlock;
@@ -878,51 +879,36 @@ static int assign_ctxt(struct file *fp, struct hfi1_user_info *uinfo)
 	 * We execute the following block if we couldn't find a
 	 * shared context or if context sharing is not required.
 	 */
-	if (!ret) {
-		i_minor = iminor(file_inode(fp)) - HFI1_USER_MINOR_BASE;
-		ret = get_user_context(fp->private_data, uinfo, i_minor);
-	}
+	if (!ret)
+		ret = allocate_ctxt(fd, fd->dd, uinfo);
+
 done_unlock:
 	mutex_unlock(&hfi1_mutex);
 done:
 	return ret;
 }
 
-static int get_user_context(struct hfi1_filedata *fd,
-			    struct hfi1_user_info *uinfo, int devno)
-{
-	struct hfi1_devdata *dd = NULL;
-	int devmax, npresent, nup;
-
-	devmax = hfi1_count_units(&npresent, &nup);
-	if (!npresent)
-		return -ENXIO;
-
-	if (!nup)
-		return -ENETDOWN;
-
-	dd = hfi1_lookup(devno);
-	if (!dd)
-		return -ENODEV;
-	else if (!dd->freectxts)
-		return -EBUSY;
-
-	return allocate_ctxt(fd, dd, uinfo);
-}
-
 static int find_shared_ctxt(struct hfi1_filedata *fd,
 			    const struct hfi1_user_info *uinfo)
 {
-	int devmax, ndev, i;
-	int ret = 0;
+	int i;
+	struct hfi1_devdata *dd = fd->dd;
 
-	devmax = hfi1_count_units(NULL, NULL);
+	for (i = dd->first_user_ctxt; i < dd->num_rcv_contexts; i++) {
+		struct hfi1_ctxtdata *uctxt = dd->rcd[i];
 
-	for (ndev = 0; ndev < devmax; ndev++) {
-		struct hfi1_devdata *dd = hfi1_lookup(ndev);
+		/* Skip ctxts which are not yet open */
+		if (!uctxt || !uctxt->cnt)
+			continue;
 
-		if (!(dd && (dd->flags & HFI1_PRESENT) && dd->kregbase))
+		/* Skip ctxt if it doesn't match the requested one */
+		if (memcmp(uctxt->uuid, uinfo->uuid,
+			   sizeof(uctxt->uuid)) ||
+		    uctxt->jkey != generate_jkey(current_uid()) ||
+		    uctxt->subctxt_id != uinfo->subctxt_id ||
+		    uctxt->subctxt_cnt != uinfo->subctxt_cnt)
 			continue;
+<<<<<<< HEAD
 		for (i = dd->first_dyn_alloc_ctxt;
 		     i < dd->num_rcv_contexts; i++) {
 			struct hfi1_ctxtdata *uctxt = dd->rcd[i];
@@ -954,18 +940,28 @@ static int find_shared_ctxt(struct hfi1_filedata *fd,
 			uctxt->active_slaves |= 1 << fd->subctxt;
 			ret = 1;
 			goto done;
+=======
+
+		/* Verify the sharing process matches the master */
+		if (uctxt->userversion != uinfo->userversion ||
+		    uctxt->cnt >= uctxt->subctxt_cnt) {
+			return -EINVAL;
+>>>>>>> f1f8a68... IB/hfi1: Search shared contexts on the opened device, not all devices
 		}
+		fd->uctxt = uctxt;
+		fd->subctxt  = uctxt->cnt++;
+		uctxt->active_slaves |= 1 << fd->subctxt;
+		return 1;
 	}
 
-done:
-	return ret;
+	return 0;
 }
 
 static int allocate_ctxt(struct hfi1_filedata *fd, struct hfi1_devdata *dd,
 			 struct hfi1_user_info *uinfo)
 {
 	struct hfi1_ctxtdata *uctxt;
-	unsigned ctxt;
+	unsigned int ctxt;
 	int ret, numa;
 
 	if (dd->flags & HFI1_FROZEN) {
@@ -979,8 +975,20 @@ static int allocate_ctxt(struct hfi1_filedata *fd, struct hfi1_devdata *dd,
 		return -EIO;
 	}
 
+<<<<<<< HEAD
 	for (ctxt = dd->first_dyn_alloc_ctxt;
 	     ctxt < dd->num_rcv_contexts; ctxt++)
+=======
+	/*
+	 * This check is sort of redundant to the next EBUSY error. It would
+	 * also indicate an inconsistancy in the driver if this value was
+	 * zero, but there were still contexts available.
+	 */
+	if (!dd->freectxts)
+		return -EBUSY;
+
+	for (ctxt = dd->first_user_ctxt; ctxt < dd->num_rcv_contexts; ctxt++)
+>>>>>>> f1f8a68... IB/hfi1: Search shared contexts on the opened device, not all devices
 		if (!dd->rcd[ctxt])
 			break;
 
diff --git a/drivers/infiniband/hw/hfi1/hfi.h b/drivers/infiniband/hw/hfi1/hfi.h
index a1e9b8b..e0c4c5e 100644
--- a/drivers/infiniband/hw/hfi1/hfi.h
+++ b/drivers/infiniband/hw/hfi1/hfi.h
@@ -1238,6 +1238,7 @@ static inline bool hfi1_vnic_is_rsm_full(struct hfi1_devdata *dd, int spare)
 
 /* Private data for file operations */
 struct hfi1_filedata {
+	struct hfi1_devdata *dd;
 	struct hfi1_ctxtdata *uctxt;
 	unsigned subctxt;
 	struct hfi1_user_sdma_comp_q *cq;
@@ -1264,7 +1265,6 @@ struct hfi1_filedata {
 extern unsigned long *hfi1_cpulist;
 
 int hfi1_init(struct hfi1_devdata *dd, int reinit);
-int hfi1_count_units(int *npresentp, int *nupp);
 int hfi1_count_active_units(void);
 
 int hfi1_diag_add(struct hfi1_devdata *dd);

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH 12/15] IB/hfi1: Correctly clear the pkey

[Dennis Dalessandro]

From: Michael J. Ruhl <michael.j.ruhl-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

In the close path the context is removed from the device array, and then
the clear pkey function is called.  The pkey function trys to get the
context from the device array, but because it was removed the clearing
does not occur.

Rework pkey clear function to work as expected.  Update the function
variable to reflect the correct size and name of the hw_context.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/infiniband/hw/hfi1/chip.c     |   34 ++++++++++++++-------------------
 drivers/infiniband/hw/hfi1/chip.h     |    2 +-
 drivers/infiniband/hw/hfi1/file_ops.c |    2 +-
 3 files changed, 16 insertions(+), 22 deletions(-)

diff --git a/drivers/infiniband/hw/hfi1/chip.c b/drivers/infiniband/hw/hfi1/chip.c
index 5aa95da..9f72d55 100644
--- a/drivers/infiniband/hw/hfi1/chip.c
+++ b/drivers/infiniband/hw/hfi1/chip.c
@@ -14528,30 +14528,24 @@ int hfi1_set_ctxt_pkey(struct hfi1_devdata *dd, unsigned ctxt, u16 pkey)
 	return ret;
 }
 
-int hfi1_clear_ctxt_pkey(struct hfi1_devdata *dd, unsigned ctxt)
+int hfi1_clear_ctxt_pkey(struct hfi1_devdata *dd, struct hfi1_ctxtdata *ctxt)
 {
-	struct hfi1_ctxtdata *rcd;
-	unsigned sctxt;
-	int ret = 0;
+	u8 hw_ctxt;
 	u64 reg;
 
-	if (ctxt < dd->num_rcv_contexts) {
-		rcd = dd->rcd[ctxt];
-	} else {
-		ret = -EINVAL;
-		goto done;
-	}
-	if (!rcd || !rcd->sc) {
-		ret = -EINVAL;
-		goto done;
-	}
-	sctxt = rcd->sc->hw_context;
-	reg = read_kctxt_csr(dd, sctxt, SEND_CTXT_CHECK_ENABLE);
+	if (!ctxt || !ctxt->sc)
+		return -EINVAL;
+
+	if (ctxt->ctxt >= dd->num_rcv_contexts)
+		return -EINVAL;
+
+	hw_ctxt = ctxt->sc->hw_context;
+	reg = read_kctxt_csr(dd, hw_ctxt, SEND_CTXT_CHECK_ENABLE);
 	reg &= ~SEND_CTXT_CHECK_ENABLE_CHECK_PARTITION_KEY_SMASK;
-	write_kctxt_csr(dd, sctxt, SEND_CTXT_CHECK_ENABLE, reg);
-	write_kctxt_csr(dd, sctxt, SEND_CTXT_CHECK_PARTITION_KEY, 0);
-done:
-	return ret;
+	write_kctxt_csr(dd, hw_ctxt, SEND_CTXT_CHECK_ENABLE, reg);
+	write_kctxt_csr(dd, hw_ctxt, SEND_CTXT_CHECK_PARTITION_KEY, 0);
+
+	return 0;
 }
 
 /*
diff --git a/drivers/infiniband/hw/hfi1/chip.h b/drivers/infiniband/hw/hfi1/chip.h
index 4c3c88d..760aa2e 100644
--- a/drivers/infiniband/hw/hfi1/chip.h
+++ b/drivers/infiniband/hw/hfi1/chip.h
@@ -1362,7 +1362,7 @@ void hfi1_put_tid(struct hfi1_devdata *dd, u32 index,
 int hfi1_set_ctxt_jkey(struct hfi1_devdata *dd, unsigned ctxt, u16 jkey);
 int hfi1_clear_ctxt_jkey(struct hfi1_devdata *dd, unsigned ctxt);
 int hfi1_set_ctxt_pkey(struct hfi1_devdata *dd, unsigned ctxt, u16 pkey);
-int hfi1_clear_ctxt_pkey(struct hfi1_devdata *dd, unsigned ctxt);
+int hfi1_clear_ctxt_pkey(struct hfi1_devdata *dd, struct hfi1_ctxtdata *ctxt);
 void hfi1_read_link_quality(struct hfi1_devdata *dd, u8 *link_quality);
 void hfi1_init_vnic_rsm(struct hfi1_devdata *dd);
 void hfi1_deinit_vnic_rsm(struct hfi1_devdata *dd);
diff --git a/drivers/infiniband/hw/hfi1/file_ops.c b/drivers/infiniband/hw/hfi1/file_ops.c
index 9fd4b0d..74c03e8 100644
--- a/drivers/infiniband/hw/hfi1/file_ops.c
+++ b/drivers/infiniband/hw/hfi1/file_ops.c
@@ -810,7 +810,7 @@ static int hfi1_file_close(struct inode *inode, struct file *fp)
 	dd->rcd[uctxt->ctxt] = NULL;
 
 	hfi1_user_exp_rcv_grp_free(uctxt);
-	hfi1_clear_ctxt_pkey(dd, uctxt->ctxt);
+	hfi1_clear_ctxt_pkey(dd, uctxt);
 
 	uctxt->rcvwait_to = 0;
 	uctxt->piowait_to = 0;

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH 13/15] IB/hfi1: Clean up context initialization

[Dennis Dalessandro]

From: Michael J. Ruhl <michael.j.ruhl-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

Context initialization mixes base context init with sub context init.
This is bad because contexts can be reused, and on reuse, reinit things
that should not re-initialized.

Normalize comments and function names to refer to base context and
sub context (not main, shared or slaves).

Separate the base context initialization from sub context initialization.

hfi1_init_ctxt() cannot return an error so changed to a void and remove
error message.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/infiniband/hw/hfi1/chip.c         |    3 
 drivers/infiniband/hw/hfi1/chip.h         |    2 
 drivers/infiniband/hw/hfi1/file_ops.c     |  179 +++++++++++++++--------------
 drivers/infiniband/hw/hfi1/hfi.h          |   13 +-
 drivers/infiniband/hw/hfi1/init.c         |   10 --
 drivers/infiniband/hw/hfi1/trace_ctxts.h  |   17 ++-
 drivers/infiniband/hw/hfi1/user_exp_rcv.c |   83 +++++++------
 drivers/infiniband/hw/hfi1/user_exp_rcv.h |    3 
 8 files changed, 156 insertions(+), 154 deletions(-)

diff --git a/drivers/infiniband/hw/hfi1/chip.c b/drivers/infiniband/hw/hfi1/chip.c
index 9f72d55..972da41 100644
--- a/drivers/infiniband/hw/hfi1/chip.c
+++ b/drivers/infiniband/hw/hfi1/chip.c
@@ -12662,7 +12662,7 @@ u8 hfi1_ibphys_portstate(struct hfi1_pportdata *ppd)
 #define SET_STATIC_RATE_CONTROL_SMASK(r) \
 (r |= SEND_CTXT_CHECK_ENABLE_DISALLOW_PBC_STATIC_RATE_CONTROL_SMASK)
 
-int hfi1_init_ctxt(struct send_context *sc)
+void hfi1_init_ctxt(struct send_context *sc)
 {
 	if (sc) {
 		struct hfi1_devdata *dd = sc->dd;
@@ -12679,7 +12679,6 @@ int hfi1_init_ctxt(struct send_context *sc)
 		write_kctxt_csr(dd, sc->hw_context,
 				SEND_CTXT_CHECK_ENABLE, reg);
 	}
-	return 0;
 }
 
 int hfi1_tempsense_rd(struct hfi1_devdata *dd, struct hfi1_temp *temp)
diff --git a/drivers/infiniband/hw/hfi1/chip.h b/drivers/infiniband/hw/hfi1/chip.h
index 760aa2e..cbe455d 100644
--- a/drivers/infiniband/hw/hfi1/chip.h
+++ b/drivers/infiniband/hw/hfi1/chip.h
@@ -1349,7 +1349,7 @@ enum {
 void hfi1_clear_tids(struct hfi1_ctxtdata *rcd);
 struct ib_header *hfi1_get_msgheader(
 				struct hfi1_devdata *dd, __le32 *rhf_addr);
-int hfi1_init_ctxt(struct send_context *sc);
+void hfi1_init_ctxt(struct send_context *sc);
 void hfi1_put_tid(struct hfi1_devdata *dd, u32 index,
 		  u32 type, unsigned long pa, u16 order);
 void hfi1_quiet_serdes(struct hfi1_pportdata *ppd);
diff --git a/drivers/infiniband/hw/hfi1/file_ops.c b/drivers/infiniband/hw/hfi1/file_ops.c
index 74c03e8..d748e65 100644
--- a/drivers/infiniband/hw/hfi1/file_ops.c
+++ b/drivers/infiniband/hw/hfi1/file_ops.c
@@ -80,16 +80,17 @@
 static int assign_ctxt(struct hfi1_filedata *fd, struct hfi1_user_info *uinfo);
 static int init_subctxts(struct hfi1_ctxtdata *uctxt,
 			 const struct hfi1_user_info *uinfo);
-static int user_init(struct hfi1_filedata *fd);
+static int init_user_ctxt(struct hfi1_filedata *fd);
+static int user_init(struct hfi1_ctxtdata *uctxt);
 static int get_ctxt_info(struct hfi1_filedata *fd, void __user *ubase,
 			 __u32 len);
 static int get_base_info(struct hfi1_filedata *fd, void __user *ubase,
 			 __u32 len);
-static int setup_ctxt(struct hfi1_filedata *fd);
+static int setup_base_ctxt(struct hfi1_filedata *fd);
 static int setup_subctxt(struct hfi1_ctxtdata *uctxt);
 
-static int find_shared_ctxt(struct hfi1_filedata *fd,
-			    const struct hfi1_user_info *uinfo);
+static int find_sub_ctxt(struct hfi1_filedata *fd,
+			 const struct hfi1_user_info *uinfo);
 static int allocate_ctxt(struct hfi1_filedata *fd, struct hfi1_devdata *dd,
 			 struct hfi1_user_info *uinfo);
 static unsigned int poll_urgent(struct file *fp, struct poll_table_struct *pt);
@@ -241,12 +242,6 @@ static long hfi1_file_ioctl(struct file *fp, unsigned int cmd,
 			return -EFAULT;
 
 		ret = assign_ctxt(fd, &uinfo);
-		if (ret < 0)
-			return ret;
-		ret = setup_ctxt(fd);
-		if (ret)
-			return ret;
-		ret = user_init(fd);
 		break;
 	case HFI1_IOCTL_CTXT_INFO:
 		ret = get_ctxt_info(fd, (void __user *)(unsigned long)arg,
@@ -856,40 +851,62 @@ static int assign_ctxt(struct hfi1_filedata *fd, struct hfi1_user_info *uinfo)
 	unsigned int swmajor, swminor;
 
 	swmajor = uinfo->userversion >> 16;
-	if (swmajor != HFI1_USER_SWMAJOR) {
-		ret = -ENODEV;
-		goto done;
-	}
+	if (swmajor != HFI1_USER_SWMAJOR)
+		return -ENODEV;
 
 	swminor = uinfo->userversion & 0xffff;
 
 	mutex_lock(&hfi1_mutex);
-	/* First, lets check if we need to setup a shared context? */
+	/* First, lets check if we need to get a sub context? */
 	if (uinfo->subctxt_cnt) {
-		ret = find_shared_ctxt(fd, uinfo);
-		if (ret < 0)
-			goto done_unlock;
-		if (ret) {
+		/* < 0 error, 0 no context, 1 sub-context found */
+		ret = find_sub_ctxt(fd, uinfo);
+		if (ret > 0) {
 			fd->rec_cpu_num =
 				hfi1_get_proc_affinity(fd->uctxt->numa_id);
 		}
 	}
 
 	/*
-	 * We execute the following block if we couldn't find a
-	 * shared context or if context sharing is not required.
+	 * Allocate a base context f context sharing is not required or we
+	 * couldn't find a sub context.
 	 */
 	if (!ret)
 		ret = allocate_ctxt(fd, fd->dd, uinfo);
 
-done_unlock:
 	mutex_unlock(&hfi1_mutex);
-done:
+
+	/* Depending on the context type, do the appropriate init */
+	if (ret > 0) {
+		/*
+		 * sub-context info can only be set up after the base
+		 * context has been completed.
+		 */
+		ret = wait_event_interruptible(fd->uctxt->wait, !test_bit(
+					       HFI1_CTXT_BASE_UNINIT,
+					       &fd->uctxt->event_flags));
+		/* The only thing a sub context needs is the user_xxx stuff */
+		if (!ret)
+			init_user_ctxt(fd);
+	} else if (!ret) {
+		ret = setup_base_ctxt(fd);
+
+		/*
+		 * Base context is done, notify anybody using a sub-context
+		 * that is waiting for this completion
+		 */
+		if (!ret && fd->uctxt->subctxt_cnt) {
+			clear_bit(HFI1_CTXT_BASE_UNINIT,
+				  &fd->uctxt->event_flags);
+			wake_up(&fd->uctxt->wait);
+		}
+	}
+
 	return ret;
 }
 
-static int find_shared_ctxt(struct hfi1_filedata *fd,
-			    const struct hfi1_user_info *uinfo)
+static int find_sub_ctxt(struct hfi1_filedata *fd,
+			 const struct hfi1_user_info *uinfo)
 {
 	int i;
 	struct hfi1_devdata *dd = fd->dd;
@@ -1030,12 +1047,12 @@ static int allocate_ctxt(struct hfi1_filedata *fd, struct hfi1_devdata *dd,
 		goto ctxdata_free;
 
 	/*
-	 * Setup shared context resources if the user-level has requested
-	 * shared contexts and this is the 'master' process.
+	 * Setup sub context resources if the user-level has requested
+	 * sub contexts.
 	 * This has to be done here so the rest of the sub-contexts find the
 	 * proper master.
 	 */
-	if (uinfo->subctxt_cnt && !fd->subctxt) {
+	if (uinfo->subctxt_cnt) {
 		ret = init_subctxts(uctxt, uinfo);
 		/*
 		 * On error, we don't need to disable and de-allocate the
@@ -1082,7 +1099,7 @@ static int init_subctxts(struct hfi1_ctxtdata *uctxt,
 	uctxt->subctxt_id = uinfo->subctxt_id;
 	uctxt->active_slaves = 1;
 	uctxt->redirect_seq_cnt = 1;
-	set_bit(HFI1_CTXT_MASTER_UNINIT, &uctxt->event_flags);
+	set_bit(HFI1_CTXT_BASE_UNINIT, &uctxt->event_flags);
 
 	return 0;
 }
@@ -1093,10 +1110,9 @@ static int setup_subctxt(struct hfi1_ctxtdata *uctxt)
 	unsigned num_subctxts = uctxt->subctxt_cnt;
 
 	uctxt->subctxt_uregbase = vmalloc_user(PAGE_SIZE);
-	if (!uctxt->subctxt_uregbase) {
-		ret = -ENOMEM;
-		goto bail;
-	}
+	if (!uctxt->subctxt_uregbase)
+		return -ENOMEM;
+
 	/* We can take the size of the RcvHdr Queue from the master */
 	uctxt->subctxt_rcvhdr_base = vmalloc_user(uctxt->rcvhdrq_size *
 						  num_subctxts);
@@ -1111,24 +1127,22 @@ static int setup_subctxt(struct hfi1_ctxtdata *uctxt)
 		ret = -ENOMEM;
 		goto bail_rhdr;
 	}
-	goto bail;
+
+	return 0;
+
 bail_rhdr:
 	vfree(uctxt->subctxt_rcvhdr_base);
+	uctxt->subctxt_rcvhdr_base = NULL;
 bail_ureg:
 	vfree(uctxt->subctxt_uregbase);
 	uctxt->subctxt_uregbase = NULL;
-bail:
+
 	return ret;
 }
 
-static int user_init(struct hfi1_filedata *fd)
+static int user_init(struct hfi1_ctxtdata *uctxt)
 {
 	unsigned int rcvctrl_ops = 0;
-	struct hfi1_ctxtdata *uctxt = fd->uctxt;
-
-	/* make sure that the context has already been setup */
-	if (!test_bit(HFI1_CTXT_SETUP_DONE, &uctxt->event_flags))
-		return -EFAULT;
 
 	/* initialize poll variables... */
 	uctxt->urgent = 0;
@@ -1177,12 +1191,6 @@ static int user_init(struct hfi1_filedata *fd)
 		rcvctrl_ops |= HFI1_RCVCTRL_TAILUPD_DIS;
 	hfi1_rcvctrl(uctxt->dd, rcvctrl_ops, uctxt->ctxt);
 
-	/* Notify any waiting slaves */
-	if (uctxt->subctxt_cnt) {
-		clear_bit(HFI1_CTXT_MASTER_UNINIT, &uctxt->event_flags);
-		wake_up(&uctxt->wait);
-	}
-
 	return 0;
 }
 
@@ -1227,59 +1235,52 @@ static int get_ctxt_info(struct hfi1_filedata *fd, void __user *ubase,
 	return ret;
 }
 
-static int setup_ctxt(struct hfi1_filedata *fd)
+static int init_user_ctxt(struct hfi1_filedata *fd)
+{
+	struct hfi1_ctxtdata *uctxt = fd->uctxt;
+	int ret;
+
+	ret = hfi1_user_sdma_alloc_queues(uctxt, fd);
+	if (ret)
+		return ret;
+
+	ret = hfi1_user_exp_rcv_init(fd);
+
+	return ret;
+}
+
+static int setup_base_ctxt(struct hfi1_filedata *fd)
 {
 	struct hfi1_ctxtdata *uctxt = fd->uctxt;
 	struct hfi1_devdata *dd = uctxt->dd;
 	int ret = 0;
 
-	/*
-	 * Context should be set up only once, including allocation and
-	 * programming of eager buffers. This is done if context sharing
-	 * is not requested or by the master process.
-	 */
-	if (!uctxt->subctxt_cnt || !fd->subctxt) {
-		ret = hfi1_init_ctxt(uctxt->sc);
-		if (ret)
-			goto done;
+	hfi1_init_ctxt(uctxt->sc);
 
-		/* Now allocate the RcvHdr queue and eager buffers. */
-		ret = hfi1_create_rcvhdrq(dd, uctxt);
-		if (ret)
-			goto done;
-		ret = hfi1_setup_eagerbufs(uctxt);
-		if (ret)
-			goto done;
-		if (uctxt->subctxt_cnt && !fd->subctxt) {
-			ret = setup_subctxt(uctxt);
-			if (ret)
-				goto done;
-		}
-	} else {
-		ret = wait_event_interruptible(uctxt->wait, !test_bit(
-					       HFI1_CTXT_MASTER_UNINIT,
-					       &uctxt->event_flags));
-		if (ret)
-			goto done;
-	}
+	/* Now allocate the RcvHdr queue and eager buffers. */
+	ret = hfi1_create_rcvhdrq(dd, uctxt);
+	if (ret)
+		goto done;
 
-	ret = hfi1_user_sdma_alloc_queues(uctxt, fd);
+	ret = hfi1_setup_eagerbufs(uctxt);
 	if (ret)
 		goto done;
-	/*
-	 * Expected receive has to be setup for all processes (including
-	 * shared contexts). However, it has to be done after the master
-	 * context has been fully configured as it depends on the
-	 * eager/expected split of the RcvArray entries.
-	 * Setting it up here ensures that the subcontexts will be waiting
-	 * (due to the above wait_event_interruptible() until the master
-	 * is setup.
-	 */
-	ret = hfi1_user_exp_rcv_init(fd);
+
+	/* If sub-contexts are enabled, do the appropriate setup */
+	if (uctxt->subctxt_cnt)
+		ret = setup_subctxt(uctxt);
+	if (ret)
+		goto done;
+
+	ret = hfi1_user_exp_rcv_grp_init(fd);
+	if (ret)
+		goto done;
+
+	ret = init_user_ctxt(fd);
 	if (ret)
 		goto done;
 
-	set_bit(HFI1_CTXT_SETUP_DONE, &uctxt->event_flags);
+	ret = user_init(uctxt);
 done:
 	return ret;
 }
@@ -1294,7 +1295,7 @@ static int get_base_info(struct hfi1_filedata *fd, void __user *ubase,
 	unsigned offset;
 	int ret = 0;
 
-	trace_hfi1_uctxtdata(uctxt->dd, uctxt);
+	trace_hfi1_uctxtdata(uctxt->dd, uctxt, fd->subctxt);
 
 	memset(&binfo, 0, sizeof(binfo));
 	binfo.hw_version = dd->revision;
diff --git a/drivers/infiniband/hw/hfi1/hfi.h b/drivers/infiniband/hw/hfi1/hfi.h
index e0c4c5e..1b7203a 100644
--- a/drivers/infiniband/hw/hfi1/hfi.h
+++ b/drivers/infiniband/hw/hfi1/hfi.h
@@ -224,13 +224,12 @@ struct hfi1_ctxtdata {
 	 * (ignoring forks, dup, etc. for now)
 	 */
 	int cnt;
+	/* Device context index */
+	unsigned ctxt;
 	/*
-	 * how much space to leave at start of eager TID entries for
-	 * protocol use, on each TID
+	 * non-zero if ctxt can be shared, and defines the maximum number of
+	 * sub contexts allowed.
 	 */
-	/* instead of calculating it */
-	unsigned ctxt;
-	/* non-zero if ctxt is being shared. */
 	u16 subctxt_cnt;
 	/* non-zero if ctxt is being shared. */
 	u16 subctxt_id;
@@ -1725,12 +1724,10 @@ struct cc_state *get_cc_state_protected(struct hfi1_pportdata *ppd)
 #define HFI1_PBC_LENGTH_MASK                     ((1 << 11) - 1)
 
 /* ctxt_flag bit offsets */
-		/* context has been setup */
-#define HFI1_CTXT_SETUP_DONE 1
 		/* waiting for a packet to arrive */
 #define HFI1_CTXT_WAITING_RCV   2
 		/* master has not finished initializing */
-#define HFI1_CTXT_MASTER_UNINIT 4
+#define HFI1_CTXT_BASE_UNINIT 4
 		/* waiting for an urgent packet to arrive */
 #define HFI1_CTXT_WAITING_URG 5
 
diff --git a/drivers/infiniband/hw/hfi1/init.c b/drivers/infiniband/hw/hfi1/init.c
index e872644..52a6364 100644
--- a/drivers/infiniband/hw/hfi1/init.c
+++ b/drivers/infiniband/hw/hfi1/init.c
@@ -176,13 +176,7 @@ int hfi1_create_ctxts(struct hfi1_devdata *dd)
 			goto nomem;
 		}
 
-		ret = hfi1_init_ctxt(rcd->sc);
-		if (ret < 0) {
-			dd_dev_err(dd,
-				   "Failed to setup kernel receive context, failing\n");
-			ret = -EFAULT;
-			goto bail;
-		}
+		hfi1_init_ctxt(rcd->sc);
 	}
 
 	/*
@@ -194,7 +188,7 @@ int hfi1_create_ctxts(struct hfi1_devdata *dd)
 	return 0;
 nomem:
 	ret = -ENOMEM;
-bail:
+
 	if (dd->rcd) {
 		for (i = 0; i < dd->num_rcv_contexts; ++i)
 			hfi1_free_ctxtdata(dd, dd->rcd[i]);
diff --git a/drivers/infiniband/hw/hfi1/trace_ctxts.h b/drivers/infiniband/hw/hfi1/trace_ctxts.h
index 26ae789..4eb4cc7 100644
--- a/drivers/infiniband/hw/hfi1/trace_ctxts.h
+++ b/drivers/infiniband/hw/hfi1/trace_ctxts.h
@@ -57,12 +57,14 @@
 
 #define UCTXT_FMT \
 	"cred:%u, credaddr:0x%llx, piobase:0x%p, rcvhdr_cnt:%u, "	\
-	"rcvbase:0x%llx, rcvegrc:%u, rcvegrb:0x%llx"
+	"rcvbase:0x%llx, rcvegrc:%u, rcvegrb:0x%llx, subctxt_cnt:%u"
 TRACE_EVENT(hfi1_uctxtdata,
-	    TP_PROTO(struct hfi1_devdata *dd, struct hfi1_ctxtdata *uctxt),
-	    TP_ARGS(dd, uctxt),
+	    TP_PROTO(struct hfi1_devdata *dd, struct hfi1_ctxtdata *uctxt,
+		     unsigned int subctxt),
+	    TP_ARGS(dd, uctxt, subctxt),
 	    TP_STRUCT__entry(DD_DEV_ENTRY(dd)
 			     __field(unsigned int, ctxt)
+			     __field(unsigned int, subctxt)
 			     __field(u32, credits)
 			     __field(u64, hw_free)
 			     __field(void __iomem *, piobase)
@@ -70,9 +72,11 @@
 			     __field(u64, rcvhdrq_dma)
 			     __field(u32, eager_cnt)
 			     __field(u64, rcvegr_dma)
+			     __field(unsigned int, subctxt_cnt)
 			     ),
 	    TP_fast_assign(DD_DEV_ASSIGN(dd);
 			   __entry->ctxt = uctxt->ctxt;
+			   __entry->subctxt = subctxt;
 			   __entry->credits = uctxt->sc->credits;
 			   __entry->hw_free = le64_to_cpu(*uctxt->sc->hw_free);
 			   __entry->piobase = uctxt->sc->base_addr;
@@ -80,17 +84,20 @@
 			   __entry->rcvhdrq_dma = uctxt->rcvhdrq_dma;
 			   __entry->eager_cnt = uctxt->egrbufs.alloced;
 			   __entry->rcvegr_dma = uctxt->egrbufs.rcvtids[0].dma;
+			   __entry->subctxt_cnt = uctxt->subctxt_cnt;
 			   ),
-	    TP_printk("[%s] ctxt %u " UCTXT_FMT,
+	    TP_printk("[%s] ctxt %u:%u " UCTXT_FMT,
 		      __get_str(dev),
 		      __entry->ctxt,
+		      __entry->subctxt,
 		      __entry->credits,
 		      __entry->hw_free,
 		      __entry->piobase,
 		      __entry->rcvhdrq_cnt,
 		      __entry->rcvhdrq_dma,
 		      __entry->eager_cnt,
-		      __entry->rcvegr_dma
+		      __entry->rcvegr_dma,
+		      __entry->subctxt_cnt
 		      )
 );
 
diff --git a/drivers/infiniband/hw/hfi1/user_exp_rcv.c b/drivers/infiniband/hw/hfi1/user_exp_rcv.c
index 22ffda8..bfaf159 100644
--- a/drivers/infiniband/hw/hfi1/user_exp_rcv.c
+++ b/drivers/infiniband/hw/hfi1/user_exp_rcv.c
@@ -53,7 +53,7 @@
 
 struct tid_group {
 	struct list_head list;
-	unsigned base;
+	u32 base;
 	u8 size;
 	u8 used;
 	u8 map;
@@ -154,6 +154,40 @@ static inline void tid_group_move(struct tid_group *group,
 	tid_group_add_tail(group, s2);
 }
 
+int hfi1_user_exp_rcv_grp_init(struct hfi1_filedata *fd)
+{
+	struct hfi1_ctxtdata *uctxt = fd->uctxt;
+	struct hfi1_devdata *dd = fd->dd;
+	u32 tidbase;
+	u32 i;
+
+	exp_tid_group_init(&uctxt->tid_group_list);
+	exp_tid_group_init(&uctxt->tid_used_list);
+	exp_tid_group_init(&uctxt->tid_full_list);
+
+	tidbase = uctxt->expected_base;
+	for (i = 0; i < uctxt->expected_count /
+		     dd->rcv_entries.group_size; i++) {
+		struct tid_group *grp;
+
+		grp = kzalloc(sizeof(*grp), GFP_KERNEL);
+		if (!grp) {
+			/*
+			 * If we fail here, the groups already
+			 * allocated will be freed by the close
+			 * call.
+			 */
+			return -ENOMEM;
+		}
+		grp->size = dd->rcv_entries.group_size;
+		grp->base = tidbase;
+		tid_group_add_tail(grp, &uctxt->tid_group_list);
+		tidbase += dd->rcv_entries.group_size;
+	}
+
+	return 0;
+}
+
 /*
  * Initialize context and file private data needed for Expected
  * receive caching. This needs to be done after the context has
@@ -163,54 +197,24 @@ int hfi1_user_exp_rcv_init(struct hfi1_filedata *fd)
 {
 	struct hfi1_ctxtdata *uctxt = fd->uctxt;
 	struct hfi1_devdata *dd = uctxt->dd;
-	unsigned tidbase;
-	int i, ret = 0;
+	int ret = 0;
 
 	spin_lock_init(&fd->tid_lock);
 	spin_lock_init(&fd->invalid_lock);
 
-	if (!uctxt->subctxt_cnt || !fd->subctxt) {
-		exp_tid_group_init(&uctxt->tid_group_list);
-		exp_tid_group_init(&uctxt->tid_used_list);
-		exp_tid_group_init(&uctxt->tid_full_list);
-
-		tidbase = uctxt->expected_base;
-		for (i = 0; i < uctxt->expected_count /
-			     dd->rcv_entries.group_size; i++) {
-			struct tid_group *grp;
-
-			grp = kzalloc(sizeof(*grp), GFP_KERNEL);
-			if (!grp) {
-				/*
-				 * If we fail here, the groups already
-				 * allocated will be freed by the close
-				 * call.
-				 */
-				ret = -ENOMEM;
-				goto done;
-			}
-			grp->size = dd->rcv_entries.group_size;
-			grp->base = tidbase;
-			tid_group_add_tail(grp, &uctxt->tid_group_list);
-			tidbase += dd->rcv_entries.group_size;
-		}
-	}
-
 	fd->entry_to_rb = kcalloc(uctxt->expected_count,
-				     sizeof(struct rb_node *),
-				     GFP_KERNEL);
+				  sizeof(struct rb_node *),
+				  GFP_KERNEL);
 	if (!fd->entry_to_rb)
 		return -ENOMEM;
 
 	if (!HFI1_CAP_UGET_MASK(uctxt->flags, TID_UNMAP)) {
 		fd->invalid_tid_idx = 0;
 		fd->invalid_tids = kcalloc(uctxt->expected_count,
-					   sizeof(*fd->invalid_tids),
+					   sizeof(u32),
 					   GFP_KERNEL);
-		if (!fd->invalid_tids) {
-			ret = -ENOMEM;
-			goto done;
-		}
+		if (!fd->invalid_tids)
+			return -ENOMEM;
 
 		/*
 		 * Register MMU notifier callbacks. If the registration
@@ -252,7 +256,7 @@ int hfi1_user_exp_rcv_init(struct hfi1_filedata *fd)
 		fd->tid_limit = uctxt->expected_count;
 	}
 	spin_unlock(&fd->tid_lock);
-done:
+
 	return ret;
 }
 
@@ -268,7 +272,7 @@ void hfi1_user_exp_rcv_grp_free(struct hfi1_ctxtdata *uctxt)
 	hfi1_clear_tids(uctxt);
 }
 
-int hfi1_user_exp_rcv_free(struct hfi1_filedata *fd)
+void hfi1_user_exp_rcv_free(struct hfi1_filedata *fd)
 {
 	struct hfi1_ctxtdata *uctxt = fd->uctxt;
 
@@ -290,7 +294,6 @@ int hfi1_user_exp_rcv_free(struct hfi1_filedata *fd)
 
 	kfree(fd->entry_to_rb);
 	fd->entry_to_rb = NULL;
-	return 0;
 }
 
 /*
diff --git a/drivers/infiniband/hw/hfi1/user_exp_rcv.h b/drivers/infiniband/hw/hfi1/user_exp_rcv.h
index 9787511..5250c89 100644
--- a/drivers/infiniband/hw/hfi1/user_exp_rcv.h
+++ b/drivers/infiniband/hw/hfi1/user_exp_rcv.h
@@ -71,8 +71,9 @@
 	} while (0)
 
 void hfi1_user_exp_rcv_grp_free(struct hfi1_ctxtdata *uctxt);
+int hfi1_user_exp_rcv_grp_init(struct hfi1_filedata *fd);
 int hfi1_user_exp_rcv_init(struct hfi1_filedata *fd);
-int hfi1_user_exp_rcv_free(struct hfi1_filedata *fd);
+void hfi1_user_exp_rcv_free(struct hfi1_filedata *fd);
 int hfi1_user_exp_rcv_setup(struct hfi1_filedata *fd,
 			    struct hfi1_tid_info *tinfo);
 int hfi1_user_exp_rcv_clear(struct hfi1_filedata *fd,

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH 14/15] IB/hfi1: Fix an assign/ordering issue with shared context IDs

[Dennis Dalessandro]

From: Michael J. Ruhl <michael.j.ruhl-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

The current algorithm for generating sub-context IDs is FILO.  If the
contexts are not closed in that order, the uniqueness of the ID will be
compromised. I.e. logging the creation/deletion of context IDs with an
application that assigns and closes in a FIFO order reveals:

cache_id: assign: uctxt: 3    sub_ctxt: 0
cache_id: assign: uctxt: 3    sub_ctxt: 1
cache_id: assign: uctxt: 3    sub_ctxt: 2
cache_id: close:  uctxt: 3    sub_ctxt: 0
cache_id: assign: uctxt: 3    sub_ctxt: 2 <<<

The sub_ctxt ID 2 is reused incorrectly.

Update the sub-context ID assign algorithm to use a bitmask of in_use
contexts.  The new algorithm will allow the contexts to be closed in any
order, and will only re-use unused contexts.

Size subctxt and subctxt_cnt to match the user API size.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/infiniband/hw/hfi1/driver.c    |    2 +
 drivers/infiniband/hw/hfi1/file_ops.c  |   53 +++++++++++++++++++++-----------
 drivers/infiniband/hw/hfi1/hfi.h       |    8 ++---
 drivers/infiniband/hw/hfi1/init.c      |    3 +-
 drivers/infiniband/hw/hfi1/intr.c      |    3 +-
 drivers/infiniband/hw/hfi1/user_sdma.h |    2 +
 6 files changed, 44 insertions(+), 27 deletions(-)

diff --git a/drivers/infiniband/hw/hfi1/driver.c b/drivers/infiniband/hw/hfi1/driver.c
index 566d152..a50870e 100644
--- a/drivers/infiniband/hw/hfi1/driver.c
+++ b/drivers/infiniband/hw/hfi1/driver.c
@@ -1289,7 +1289,7 @@ int hfi1_reset_device(int unit)
 	if (dd->rcd)
 		for (i = dd->first_dyn_alloc_ctxt;
 		     i < dd->num_rcv_contexts; i++) {
-			if (!dd->rcd[i] || !dd->rcd[i]->cnt)
+			if (!dd->rcd[i])
 				continue;
 			spin_unlock_irqrestore(&dd->uctxt_lock, flags);
 			ret = -EBUSY;
diff --git a/drivers/infiniband/hw/hfi1/file_ops.c b/drivers/infiniband/hw/hfi1/file_ops.c
index d748e65..84e6342 100644
--- a/drivers/infiniband/hw/hfi1/file_ops.c
+++ b/drivers/infiniband/hw/hfi1/file_ops.c
@@ -49,6 +49,7 @@
 #include <linux/vmalloc.h>
 #include <linux/io.h>
 #include <linux/sched/mm.h>
+#include <linux/bitmap.h>
 
 #include <rdma/ib.h>
 
@@ -95,11 +96,10 @@ static int allocate_ctxt(struct hfi1_filedata *fd, struct hfi1_devdata *dd,
 			 struct hfi1_user_info *uinfo);
 static unsigned int poll_urgent(struct file *fp, struct poll_table_struct *pt);
 static unsigned int poll_next(struct file *fp, struct poll_table_struct *pt);
-static int user_event_ack(struct hfi1_ctxtdata *uctxt, int subctxt,
+static int user_event_ack(struct hfi1_ctxtdata *uctxt, u16 subctxt,
 			  unsigned long events);
-static int set_ctxt_pkey(struct hfi1_ctxtdata *uctxt, unsigned subctxt,
-			 u16 pkey);
-static int manage_rcvq(struct hfi1_ctxtdata *uctxt, unsigned subctxt,
+static int set_ctxt_pkey(struct hfi1_ctxtdata *uctxt, u16 subctxt, u16 pkey);
+static int manage_rcvq(struct hfi1_ctxtdata *uctxt, u16 subctxt,
 		       int start_stop);
 static int vma_fault(struct vm_fault *vmf);
 static long hfi1_file_ioctl(struct file *fp, unsigned int cmd,
@@ -773,8 +773,8 @@ static int hfi1_file_close(struct inode *inode, struct file *fp)
 			   HFI1_MAX_SHARED_CTXTS) + fdata->subctxt;
 	*ev = 0;
 
-	if (--uctxt->cnt) {
-		uctxt->active_slaves &= ~(1 << fdata->subctxt);
+	__clear_bit(fdata->subctxt, uctxt->in_use_ctxts);
+	if (!bitmap_empty(uctxt->in_use_ctxts, HFI1_MAX_SHARED_CTXTS)) {
 		mutex_unlock(&hfi1_mutex);
 		goto done;
 	}
@@ -868,7 +868,7 @@ static int assign_ctxt(struct hfi1_filedata *fd, struct hfi1_user_info *uinfo)
 	}
 
 	/*
-	 * Allocate a base context f context sharing is not required or we
+	 * Allocate a base context if context sharing is not required or we
 	 * couldn't find a sub context.
 	 */
 	if (!ret)
@@ -905,17 +905,24 @@ static int assign_ctxt(struct hfi1_filedata *fd, struct hfi1_user_info *uinfo)
 	return ret;
 }
 
+/*
+ * The hfi1_mutex must be held when this function is called.  It is
+ * necessary to ensure serialized access to the bitmask in_use_ctxts.
+ */
 static int find_sub_ctxt(struct hfi1_filedata *fd,
 			 const struct hfi1_user_info *uinfo)
 {
 	int i;
 	struct hfi1_devdata *dd = fd->dd;
+	u16 subctxt;
 
 	for (i = dd->first_user_ctxt; i < dd->num_rcv_contexts; i++) {
 		struct hfi1_ctxtdata *uctxt = dd->rcd[i];
 
 		/* Skip ctxts which are not yet open */
-		if (!uctxt || !uctxt->cnt)
+		if (!uctxt ||
+		    bitmap_empty(uctxt->in_use_ctxts,
+				 HFI1_MAX_SHARED_CTXTS))
 			continue;
 
 		/* Skip ctxt if it doesn't match the requested one */
@@ -960,14 +967,24 @@ static int find_sub_ctxt(struct hfi1_filedata *fd,
 =======
 
 		/* Verify the sharing process matches the master */
-		if (uctxt->userversion != uinfo->userversion ||
-		    uctxt->cnt >= uctxt->subctxt_cnt) {
+		if (uctxt->userversion != uinfo->userversion)
 			return -EINVAL;
+<<<<<<< HEAD
 >>>>>>> f1f8a68... IB/hfi1: Search shared contexts on the opened device, not all devices
 		}
+=======
+
+		/* Find an unused context */
+		subctxt = find_first_zero_bit(uctxt->in_use_ctxts,
+					      HFI1_MAX_SHARED_CTXTS);
+		if (subctxt >= uctxt->subctxt_cnt)
+			return -EINVAL;
+
+>>>>>>> d409d3d... IB/hfi1: Fix an assign/ordering issue with shared context IDs
 		fd->uctxt = uctxt;
-		fd->subctxt  = uctxt->cnt++;
-		uctxt->active_slaves |= 1 << fd->subctxt;
+		fd->subctxt = subctxt;
+		__set_bit(fd->subctxt, uctxt->in_use_ctxts);
+
 		return 1;
 	}
 
@@ -1089,7 +1106,7 @@ static int allocate_ctxt(struct hfi1_filedata *fd, struct hfi1_devdata *dd,
 static int init_subctxts(struct hfi1_ctxtdata *uctxt,
 			 const struct hfi1_user_info *uinfo)
 {
-	unsigned num_subctxts;
+	u16 num_subctxts;
 
 	num_subctxts = uinfo->subctxt_cnt;
 	if (num_subctxts > HFI1_MAX_SHARED_CTXTS)
@@ -1097,7 +1114,6 @@ static int init_subctxts(struct hfi1_ctxtdata *uctxt,
 
 	uctxt->subctxt_cnt = uinfo->subctxt_cnt;
 	uctxt->subctxt_id = uinfo->subctxt_id;
-	uctxt->active_slaves = 1;
 	uctxt->redirect_seq_cnt = 1;
 	set_bit(HFI1_CTXT_BASE_UNINIT, &uctxt->event_flags);
 
@@ -1107,7 +1123,7 @@ static int init_subctxts(struct hfi1_ctxtdata *uctxt,
 static int setup_subctxt(struct hfi1_ctxtdata *uctxt)
 {
 	int ret = 0;
-	unsigned num_subctxts = uctxt->subctxt_cnt;
+	u16 num_subctxts = uctxt->subctxt_cnt;
 
 	uctxt->subctxt_uregbase = vmalloc_user(PAGE_SIZE);
 	if (!uctxt->subctxt_uregbase)
@@ -1459,7 +1475,7 @@ int hfi1_set_uevent_bits(struct hfi1_pportdata *ppd, const int evtbit)
  * overflow conditions.  start_stop==1 re-enables, to be used to
  * re-init the software copy of the head register
  */
-static int manage_rcvq(struct hfi1_ctxtdata *uctxt, unsigned subctxt,
+static int manage_rcvq(struct hfi1_ctxtdata *uctxt, u16 subctxt,
 		       int start_stop)
 {
 	struct hfi1_devdata *dd = uctxt->dd;
@@ -1494,7 +1510,7 @@ static int manage_rcvq(struct hfi1_ctxtdata *uctxt, unsigned subctxt,
  * User process then performs actions appropriate to bit having been
  * set, if desired, and checks again in future.
  */
-static int user_event_ack(struct hfi1_ctxtdata *uctxt, int subctxt,
+static int user_event_ack(struct hfi1_ctxtdata *uctxt, u16 subctxt,
 			  unsigned long events)
 {
 	int i;
@@ -1515,8 +1531,7 @@ static int user_event_ack(struct hfi1_ctxtdata *uctxt, int subctxt,
 	return 0;
 }
 
-static int set_ctxt_pkey(struct hfi1_ctxtdata *uctxt, unsigned subctxt,
-			 u16 pkey)
+static int set_ctxt_pkey(struct hfi1_ctxtdata *uctxt, u16 subctxt, u16 pkey)
 {
 	int ret = -ENOENT, i, intable = 0;
 	struct hfi1_pportdata *ppd = uctxt->ppd;
diff --git a/drivers/infiniband/hw/hfi1/hfi.h b/drivers/infiniband/hw/hfi1/hfi.h
index 1b7203a..f3d75fc 100644
--- a/drivers/infiniband/hw/hfi1/hfi.h
+++ b/drivers/infiniband/hw/hfi1/hfi.h
@@ -228,7 +228,7 @@ struct hfi1_ctxtdata {
 	unsigned ctxt;
 	/*
 	 * non-zero if ctxt can be shared, and defines the maximum number of
-	 * sub contexts allowed.
+	 * sub-contexts for this device context.
 	 */
 	u16 subctxt_cnt;
 	/* non-zero if ctxt is being shared. */
@@ -287,10 +287,10 @@ struct hfi1_ctxtdata {
 	void *subctxt_rcvegrbuf;
 	/* An array of pages for the eager header queue entries * N */
 	void *subctxt_rcvhdr_base;
+	/* Bitmask of in use context(s) */
+	DECLARE_BITMAP(in_use_ctxts, HFI1_MAX_SHARED_CTXTS);
 	/* The version of the library which opened this ctxt */
 	u32 userversion;
-	/* Bitmask of active slaves */
-	u32 active_slaves;
 	/* Type of packets or conditions we want to poll for */
 	u16 poll_type;
 	/* receive packet sequence counter */
@@ -1239,9 +1239,9 @@ static inline bool hfi1_vnic_is_rsm_full(struct hfi1_devdata *dd, int spare)
 struct hfi1_filedata {
 	struct hfi1_devdata *dd;
 	struct hfi1_ctxtdata *uctxt;
-	unsigned subctxt;
 	struct hfi1_user_sdma_comp_q *cq;
 	struct hfi1_user_sdma_pkt_q *pq;
+	u16 subctxt;
 	/* for cpu affinity; -1 if none */
 	int rec_cpu_num;
 	u32 tid_n_pinned;
diff --git a/drivers/infiniband/hw/hfi1/init.c b/drivers/infiniband/hw/hfi1/init.c
index 52a6364..694a8ec 100644
--- a/drivers/infiniband/hw/hfi1/init.c
+++ b/drivers/infiniband/hw/hfi1/init.c
@@ -53,6 +53,7 @@
 #include <linux/module.h>
 #include <linux/printk.h>
 #include <linux/hrtimer.h>
+#include <linux/bitmap.h>
 #include <rdma/rdma_vt.h>
 
 #include "hfi.h"
@@ -222,7 +223,7 @@ struct hfi1_ctxtdata *hfi1_create_ctxtdata(struct hfi1_pportdata *ppd, u32 ctxt,
 		INIT_LIST_HEAD(&rcd->qp_wait_list);
 		rcd->ppd = ppd;
 		rcd->dd = dd;
-		rcd->cnt = 1;
+		__set_bit(0, rcd->in_use_ctxts);
 		rcd->ctxt = ctxt;
 		dd->rcd[ctxt] = rcd;
 		rcd->numa_id = numa;
diff --git a/drivers/infiniband/hw/hfi1/intr.c b/drivers/infiniband/hw/hfi1/intr.c
index 232014d..ba265d0 100644
--- a/drivers/infiniband/hw/hfi1/intr.c
+++ b/drivers/infiniband/hw/hfi1/intr.c
@@ -47,6 +47,7 @@
 
 #include <linux/pci.h>
 #include <linux/delay.h>
+#include <linux/bitmap.h>
 
 #include "hfi.h"
 #include "common.h"
@@ -189,7 +190,7 @@ void handle_user_interrupt(struct hfi1_ctxtdata *rcd)
 	unsigned long flags;
 
 	spin_lock_irqsave(&dd->uctxt_lock, flags);
-	if (!rcd->cnt)
+	if (bitmap_empty(rcd->in_use_ctxts, HFI1_MAX_SHARED_CTXTS))
 		goto done;
 
 	if (test_and_clear_bit(HFI1_CTXT_WAITING_RCV, &rcd->event_flags)) {
diff --git a/drivers/infiniband/hw/hfi1/user_sdma.h b/drivers/infiniband/hw/hfi1/user_sdma.h
index 9181d7c..e5b10ae 100644
--- a/drivers/infiniband/hw/hfi1/user_sdma.h
+++ b/drivers/infiniband/hw/hfi1/user_sdma.h
@@ -58,7 +58,7 @@
 struct hfi1_user_sdma_pkt_q {
 	struct list_head list;
 	unsigned ctxt;
-	unsigned subctxt;
+	u16 subctxt;
 	u16 n_max_reqs;
 	atomic_t n_reqs;
 	u16 reqidx;

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH 15/15] IB/hfi1: Clean up on context initialization failure

[Dennis Dalessandro]

From: Michael J. Ruhl <michael.j.ruhl-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

The error path for context initialization is not consistent. Cleanup all
resources on failure.

Removed unused variable user_event_mask.

Add the _BASE_FAILED bit to the event flags so that a base context can
notify waiting sub contexts that they cannot continue.

Running out of sub contexts is an EBUSY result, not EINVAL.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/infiniband/hw/hfi1/file_ops.c     |  114 +++++++++++------------------
 drivers/infiniband/hw/hfi1/hfi.h          |   16 ++--
 drivers/infiniband/hw/hfi1/init.c         |   10 +--
 drivers/infiniband/hw/hfi1/user_exp_rcv.c |   28 ++++---
 drivers/infiniband/hw/hfi1/user_sdma.c    |   60 +++++++--------
 drivers/infiniband/hw/hfi1/vnic_main.c    |    8 +-
 6 files changed, 103 insertions(+), 133 deletions(-)

diff --git a/drivers/infiniband/hw/hfi1/file_ops.c b/drivers/infiniband/hw/hfi1/file_ops.c
index 84e6342..0ea842d 100644
--- a/drivers/infiniband/hw/hfi1/file_ops.c
+++ b/drivers/infiniband/hw/hfi1/file_ops.c
@@ -82,7 +82,7 @@
 static int init_subctxts(struct hfi1_ctxtdata *uctxt,
 			 const struct hfi1_user_info *uinfo);
 static int init_user_ctxt(struct hfi1_filedata *fd);
-static int user_init(struct hfi1_ctxtdata *uctxt);
+static void user_init(struct hfi1_ctxtdata *uctxt);
 static int get_ctxt_info(struct hfi1_filedata *fd, void __user *ubase,
 			 __u32 len);
 static int get_base_info(struct hfi1_filedata *fd, void __user *ubase,
@@ -847,7 +847,7 @@ static u64 kvirt_to_phys(void *addr)
 
 static int assign_ctxt(struct hfi1_filedata *fd, struct hfi1_user_info *uinfo)
 {
-	int ret = 0;
+	int ret;
 	unsigned int swmajor, swminor;
 
 	swmajor = uinfo->userversion >> 16;
@@ -857,14 +857,16 @@ static int assign_ctxt(struct hfi1_filedata *fd, struct hfi1_user_info *uinfo)
 	swminor = uinfo->userversion & 0xffff;
 
 	mutex_lock(&hfi1_mutex);
-	/* First, lets check if we need to get a sub context? */
+	/*
+	 * Get a sub context if necessary.
+	 * ret < 0 error, 0 no context, 1 sub-context found
+	 */
+	ret = 0;
 	if (uinfo->subctxt_cnt) {
-		/* < 0 error, 0 no context, 1 sub-context found */
 		ret = find_sub_ctxt(fd, uinfo);
-		if (ret > 0) {
+		if (ret > 0)
 			fd->rec_cpu_num =
 				hfi1_get_proc_affinity(fd->uctxt->numa_id);
-		}
 	}
 
 	/*
@@ -885,17 +887,27 @@ static int assign_ctxt(struct hfi1_filedata *fd, struct hfi1_user_info *uinfo)
 		ret = wait_event_interruptible(fd->uctxt->wait, !test_bit(
 					       HFI1_CTXT_BASE_UNINIT,
 					       &fd->uctxt->event_flags));
+		if (test_bit(HFI1_CTXT_BASE_FAILED, &fd->uctxt->event_flags)) {
+			clear_bit(fd->subctxt, fd->uctxt->in_use_ctxts);
+			return -ENOMEM;
+		}
 		/* The only thing a sub context needs is the user_xxx stuff */
 		if (!ret)
-			init_user_ctxt(fd);
+			ret = init_user_ctxt(fd);
+
+		if (ret)
+			clear_bit(fd->subctxt, fd->uctxt->in_use_ctxts);
 	} else if (!ret) {
 		ret = setup_base_ctxt(fd);
-
-		/*
-		 * Base context is done, notify anybody using a sub-context
-		 * that is waiting for this completion
-		 */
-		if (!ret && fd->uctxt->subctxt_cnt) {
+		if (fd->uctxt->subctxt_cnt) {
+			/* If there is an error, set the failed bit. */
+			if (ret)
+				set_bit(HFI1_CTXT_BASE_FAILED,
+					&fd->uctxt->event_flags);
+			/*
+			 * Base context is done, notify anybody using a
+			 * sub-context that is waiting for this completion
+			 */
 			clear_bit(HFI1_CTXT_BASE_UNINIT,
 				  &fd->uctxt->event_flags);
 			wake_up(&fd->uctxt->wait);
@@ -916,7 +928,7 @@ static int find_sub_ctxt(struct hfi1_filedata *fd,
 	struct hfi1_devdata *dd = fd->dd;
 	u16 subctxt;
 
-	for (i = dd->first_user_ctxt; i < dd->num_rcv_contexts; i++) {
+	for (i = dd->first_dyn_alloc_ctxt; i < dd->num_rcv_contexts; i++) {
 		struct hfi1_ctxtdata *uctxt = dd->rcd[i];
 
 		/* Skip ctxts which are not yet open */
@@ -925,6 +937,10 @@ static int find_sub_ctxt(struct hfi1_filedata *fd,
 				 HFI1_MAX_SHARED_CTXTS))
 			continue;
 
+		/* Skip dynamically allocated kernel contexts */
+		if (uctxt->sc && (uctxt->sc->type == SC_KERNEL))
+			continue;
+
 		/* Skip ctxt if it doesn't match the requested one */
 		if (memcmp(uctxt->uuid, uinfo->uuid,
 			   sizeof(uctxt->uuid)) ||
@@ -932,55 +948,17 @@ static int find_sub_ctxt(struct hfi1_filedata *fd,
 		    uctxt->subctxt_id != uinfo->subctxt_id ||
 		    uctxt->subctxt_cnt != uinfo->subctxt_cnt)
 			continue;
-<<<<<<< HEAD
-		for (i = dd->first_dyn_alloc_ctxt;
-		     i < dd->num_rcv_contexts; i++) {
-			struct hfi1_ctxtdata *uctxt = dd->rcd[i];
-
-			/* Skip ctxts which are not yet open */
-			if (!uctxt || !uctxt->cnt)
-				continue;
-
-			/* Skip dynamically allocted kernel contexts */
-			if (uctxt->sc && (uctxt->sc->type == SC_KERNEL))
-				continue;
-
-			/* Skip ctxt if it doesn't match the requested one */
-			if (memcmp(uctxt->uuid, uinfo->uuid,
-				   sizeof(uctxt->uuid)) ||
-			    uctxt->jkey != generate_jkey(current_uid()) ||
-			    uctxt->subctxt_id != uinfo->subctxt_id ||
-			    uctxt->subctxt_cnt != uinfo->subctxt_cnt)
-				continue;
-
-			/* Verify the sharing process matches the master */
-			if (uctxt->userversion != uinfo->userversion ||
-			    uctxt->cnt >= uctxt->subctxt_cnt) {
-				ret = -EINVAL;
-				goto done;
-			}
-			fd->uctxt = uctxt;
-			fd->subctxt  = uctxt->cnt++;
-			uctxt->active_slaves |= 1 << fd->subctxt;
-			ret = 1;
-			goto done;
-=======
 
 		/* Verify the sharing process matches the master */
 		if (uctxt->userversion != uinfo->userversion)
 			return -EINVAL;
-<<<<<<< HEAD
->>>>>>> f1f8a68... IB/hfi1: Search shared contexts on the opened device, not all devices
-		}
-=======
 
 		/* Find an unused context */
 		subctxt = find_first_zero_bit(uctxt->in_use_ctxts,
 					      HFI1_MAX_SHARED_CTXTS);
 		if (subctxt >= uctxt->subctxt_cnt)
-			return -EINVAL;
+			return -EBUSY;
 
->>>>>>> d409d3d... IB/hfi1: Fix an assign/ordering issue with shared context IDs
 		fd->uctxt = uctxt;
 		fd->subctxt = subctxt;
 		__set_bit(fd->subctxt, uctxt->in_use_ctxts);
@@ -1009,10 +987,6 @@ static int allocate_ctxt(struct hfi1_filedata *fd, struct hfi1_devdata *dd,
 		return -EIO;
 	}
 
-<<<<<<< HEAD
-	for (ctxt = dd->first_dyn_alloc_ctxt;
-	     ctxt < dd->num_rcv_contexts; ctxt++)
-=======
 	/*
 	 * This check is sort of redundant to the next EBUSY error. It would
 	 * also indicate an inconsistancy in the driver if this value was
@@ -1021,8 +995,8 @@ static int allocate_ctxt(struct hfi1_filedata *fd, struct hfi1_devdata *dd,
 	if (!dd->freectxts)
 		return -EBUSY;
 
-	for (ctxt = dd->first_user_ctxt; ctxt < dd->num_rcv_contexts; ctxt++)
->>>>>>> f1f8a68... IB/hfi1: Search shared contexts on the opened device, not all devices
+	for (ctxt = dd->first_dyn_alloc_ctxt;
+	     ctxt < dd->num_rcv_contexts; ctxt++)
 		if (!dd->rcd[ctxt])
 			break;
 
@@ -1156,7 +1130,7 @@ static int setup_subctxt(struct hfi1_ctxtdata *uctxt)
 	return ret;
 }
 
-static int user_init(struct hfi1_ctxtdata *uctxt)
+static void user_init(struct hfi1_ctxtdata *uctxt)
 {
 	unsigned int rcvctrl_ops = 0;
 
@@ -1206,8 +1180,6 @@ static int user_init(struct hfi1_ctxtdata *uctxt)
 	else
 		rcvctrl_ops |= HFI1_RCVCTRL_TAILUPD_DIS;
 	hfi1_rcvctrl(uctxt->dd, rcvctrl_ops, uctxt->ctxt);
-
-	return 0;
 }
 
 static int get_ctxt_info(struct hfi1_filedata *fd, void __user *ubase,
@@ -1276,28 +1248,32 @@ static int setup_base_ctxt(struct hfi1_filedata *fd)
 	/* Now allocate the RcvHdr queue and eager buffers. */
 	ret = hfi1_create_rcvhdrq(dd, uctxt);
 	if (ret)
-		goto done;
+		return ret;
 
 	ret = hfi1_setup_eagerbufs(uctxt);
 	if (ret)
-		goto done;
+		goto setup_failed;
 
 	/* If sub-contexts are enabled, do the appropriate setup */
 	if (uctxt->subctxt_cnt)
 		ret = setup_subctxt(uctxt);
 	if (ret)
-		goto done;
+		goto setup_failed;
 
 	ret = hfi1_user_exp_rcv_grp_init(fd);
 	if (ret)
-		goto done;
+		goto setup_failed;
 
 	ret = init_user_ctxt(fd);
 	if (ret)
-		goto done;
+		goto setup_failed;
 
-	ret = user_init(uctxt);
-done:
+	user_init(uctxt);
+
+	return 0;
+
+setup_failed:
+	hfi1_free_ctxtdata(dd, uctxt);
 	return ret;
 }
 
diff --git a/drivers/infiniband/hw/hfi1/hfi.h b/drivers/infiniband/hw/hfi1/hfi.h
index f3d75fc..509df98 100644
--- a/drivers/infiniband/hw/hfi1/hfi.h
+++ b/drivers/infiniband/hw/hfi1/hfi.h
@@ -196,12 +196,6 @@ struct hfi1_ctxtdata {
 	void *rcvhdrq;
 	/* kernel virtual address where hdrqtail is updated */
 	volatile __le64 *rcvhdrtail_kvaddr;
-	/*
-	 * Shared page for kernel to signal user processes that send buffers
-	 * need disarming.  The process should call HFI1_CMD_DISARM_BUFS
-	 * or HFI1_CMD_ACK_EVENT with IPATH_EVENT_DISARM_BUFS set.
-	 */
-	unsigned long *user_event_mask;
 	/* when waiting for rcv or pioavail */
 	wait_queue_head_t wait;
 	/* rcvhdrq size (for freeing) */
@@ -1724,12 +1718,14 @@ struct cc_state *get_cc_state_protected(struct hfi1_pportdata *ppd)
 #define HFI1_PBC_LENGTH_MASK                     ((1 << 11) - 1)
 
 /* ctxt_flag bit offsets */
+		/* base context has not finished initializing */
+#define HFI1_CTXT_BASE_UNINIT 1
+		/* base context initaliation failed */
+#define HFI1_CTXT_BASE_FAILED 2
 		/* waiting for a packet to arrive */
-#define HFI1_CTXT_WAITING_RCV   2
-		/* master has not finished initializing */
-#define HFI1_CTXT_BASE_UNINIT 4
+#define HFI1_CTXT_WAITING_RCV 3
 		/* waiting for an urgent packet to arrive */
-#define HFI1_CTXT_WAITING_URG 5
+#define HFI1_CTXT_WAITING_URG 4
 
 /* free up any allocated data at closes */
 struct hfi1_devdata *hfi1_init_dd(struct pci_dev *pdev,
diff --git a/drivers/infiniband/hw/hfi1/init.c b/drivers/infiniband/hw/hfi1/init.c
index 694a8ec..4a11d4d 100644
--- a/drivers/infiniband/hw/hfi1/init.c
+++ b/drivers/infiniband/hw/hfi1/init.c
@@ -964,7 +964,6 @@ void hfi1_free_ctxtdata(struct hfi1_devdata *dd, struct hfi1_ctxtdata *rcd)
 	kfree(rcd->egrbufs.buffers);
 
 	sc_free(rcd->sc);
-	vfree(rcd->user_event_mask);
 	vfree(rcd->subctxt_uregbase);
 	vfree(rcd->subctxt_rcvegrbuf);
 	vfree(rcd->subctxt_rcvhdr_base);
@@ -1683,8 +1682,6 @@ int hfi1_create_rcvhdrq(struct hfi1_devdata *dd, struct hfi1_ctxtdata *rcd)
 	dd_dev_err(dd,
 		   "attempt to allocate 1 page for ctxt %u rcvhdrqtailaddr failed\n",
 		   rcd->ctxt);
-	vfree(rcd->user_event_mask);
-	rcd->user_event_mask = NULL;
 	dma_free_coherent(&dd->pcidev->dev, amt, rcd->rcvhdrq,
 			  rcd->rcvhdrq_dma);
 	rcd->rcvhdrq = NULL;
@@ -1851,7 +1848,7 @@ int hfi1_setup_eagerbufs(struct hfi1_ctxtdata *rcd)
 			  "ctxt%u: current Eager buffer size is invalid %u\n",
 			  rcd->ctxt, rcd->egrbufs.rcvtid_size);
 		ret = -EINVAL;
-		goto bail;
+		goto bail_rcvegrbuf_phys;
 	}
 
 	for (idx = 0; idx < rcd->egrbufs.alloced; idx++) {
@@ -1859,7 +1856,8 @@ int hfi1_setup_eagerbufs(struct hfi1_ctxtdata *rcd)
 			     rcd->egrbufs.rcvtids[idx].dma, order);
 		cond_resched();
 	}
-	goto bail;
+
+	return 0;
 
 bail_rcvegrbuf_phys:
 	for (idx = 0; idx < rcd->egrbufs.alloced &&
@@ -1873,6 +1871,6 @@ int hfi1_setup_eagerbufs(struct hfi1_ctxtdata *rcd)
 		rcd->egrbufs.buffers[idx].dma = 0;
 		rcd->egrbufs.buffers[idx].len = 0;
 	}
-bail:
+
 	return ret;
 }
diff --git a/drivers/infiniband/hw/hfi1/user_exp_rcv.c b/drivers/infiniband/hw/hfi1/user_exp_rcv.c
index bfaf159..0462d9d 100644
--- a/drivers/infiniband/hw/hfi1/user_exp_rcv.c
+++ b/drivers/infiniband/hw/hfi1/user_exp_rcv.c
@@ -160,6 +160,7 @@ int hfi1_user_exp_rcv_grp_init(struct hfi1_filedata *fd)
 	struct hfi1_devdata *dd = fd->dd;
 	u32 tidbase;
 	u32 i;
+	struct tid_group *grp, *gptr;
 
 	exp_tid_group_init(&uctxt->tid_group_list);
 	exp_tid_group_init(&uctxt->tid_used_list);
@@ -168,17 +169,10 @@ int hfi1_user_exp_rcv_grp_init(struct hfi1_filedata *fd)
 	tidbase = uctxt->expected_base;
 	for (i = 0; i < uctxt->expected_count /
 		     dd->rcv_entries.group_size; i++) {
-		struct tid_group *grp;
-
 		grp = kzalloc(sizeof(*grp), GFP_KERNEL);
-		if (!grp) {
-			/*
-			 * If we fail here, the groups already
-			 * allocated will be freed by the close
-			 * call.
-			 */
-			return -ENOMEM;
-		}
+		if (!grp)
+			goto grp_failed;
+
 		grp->size = dd->rcv_entries.group_size;
 		grp->base = tidbase;
 		tid_group_add_tail(grp, &uctxt->tid_group_list);
@@ -186,6 +180,15 @@ int hfi1_user_exp_rcv_grp_init(struct hfi1_filedata *fd)
 	}
 
 	return 0;
+
+grp_failed:
+	list_for_each_entry_safe(grp, gptr, &uctxt->tid_group_list.list,
+				 list) {
+		list_del_init(&grp->list);
+		kfree(grp);
+	}
+
+	return -ENOMEM;
 }
 
 /*
@@ -213,8 +216,11 @@ int hfi1_user_exp_rcv_init(struct hfi1_filedata *fd)
 		fd->invalid_tids = kcalloc(uctxt->expected_count,
 					   sizeof(u32),
 					   GFP_KERNEL);
-		if (!fd->invalid_tids)
+		if (!fd->invalid_tids) {
+			kfree(fd->entry_to_rb);
+			fd->entry_to_rb = NULL;
 			return -ENOMEM;
+		}
 
 		/*
 		 * Register MMU notifier callbacks. If the registration
diff --git a/drivers/infiniband/hw/hfi1/user_sdma.c b/drivers/infiniband/hw/hfi1/user_sdma.c
index 6b72267..e38fa88 100644
--- a/drivers/infiniband/hw/hfi1/user_sdma.c
+++ b/drivers/infiniband/hw/hfi1/user_sdma.c
@@ -374,40 +374,25 @@ static void sdma_kmem_cache_ctor(void *obj)
 int hfi1_user_sdma_alloc_queues(struct hfi1_ctxtdata *uctxt,
 				struct hfi1_filedata *fd)
 {
-	int ret = 0;
+	int ret = -ENOMEM;
+	unsigned memsize;
 	char buf[64];
 	struct hfi1_devdata *dd;
 	struct hfi1_user_sdma_comp_q *cq;
 	struct hfi1_user_sdma_pkt_q *pq;
 	unsigned long flags;
 
-	if (!uctxt || !fd) {
-		ret = -EBADF;
-		goto done;
-	}
+	if (!uctxt || !fd)
+		return -EBADF;
 
-	if (!hfi1_sdma_comp_ring_size) {
-		ret = -EINVAL;
-		goto done;
-	}
+	if (!hfi1_sdma_comp_ring_size)
+		return -EINVAL;
 
 	dd = uctxt->dd;
 
 	pq = kzalloc(sizeof(*pq), GFP_KERNEL);
 	if (!pq)
-		goto pq_nomem;
-
-	pq->reqs = kcalloc(hfi1_sdma_comp_ring_size,
-			   sizeof(*pq->reqs),
-			   GFP_KERNEL);
-	if (!pq->reqs)
-		goto pq_reqs_nomem;
-
-	pq->req_in_use = kcalloc(BITS_TO_LONGS(hfi1_sdma_comp_ring_size),
-				 sizeof(*pq->req_in_use),
-				 GFP_KERNEL);
-	if (!pq->req_in_use)
-		goto pq_reqs_no_in_use;
+		return -ENOMEM;
 
 	INIT_LIST_HEAD(&pq->list);
 	pq->dd = dd;
@@ -423,10 +408,21 @@ int hfi1_user_sdma_alloc_queues(struct hfi1_ctxtdata *uctxt,
 	iowait_init(&pq->busy, 0, NULL, defer_packet_queue,
 		    activate_packet_queue, NULL);
 	pq->reqidx = 0;
+
+	memsize = sizeof(*pq->reqs) * hfi1_sdma_comp_ring_size;
+	pq->reqs = kzalloc(memsize, GFP_KERNEL);
+	if (!pq->reqs)
+		goto pq_reqs_nomem;
+
+	memsize = BITS_TO_LONGS(hfi1_sdma_comp_ring_size) * sizeof(long);
+	pq->req_in_use = kzalloc(memsize, GFP_KERNEL);
+	if (!pq->req_in_use)
+		goto pq_reqs_no_in_use;
+
 	snprintf(buf, 64, "txreq-kmem-cache-%u-%u-%u", dd->unit, uctxt->ctxt,
 		 fd->subctxt);
 	pq->txreq_cache = kmem_cache_create(buf,
-			       sizeof(struct user_sdma_txreq),
+					    sizeof(struct user_sdma_txreq),
 					    L1_CACHE_BYTES,
 					    SLAB_HWCACHE_ALIGN,
 					    sdma_kmem_cache_ctor);
@@ -435,7 +431,7 @@ int hfi1_user_sdma_alloc_queues(struct hfi1_ctxtdata *uctxt,
 			   uctxt->ctxt);
 		goto pq_txreq_nomem;
 	}
-	fd->pq = pq;
+
 	cq = kzalloc(sizeof(*cq), GFP_KERNEL);
 	if (!cq)
 		goto cq_nomem;
@@ -446,20 +442,25 @@ int hfi1_user_sdma_alloc_queues(struct hfi1_ctxtdata *uctxt,
 		goto cq_comps_nomem;
 
 	cq->nentries = hfi1_sdma_comp_ring_size;
-	fd->cq = cq;
 
 	ret = hfi1_mmu_rb_register(pq, pq->mm, &sdma_rb_ops, dd->pport->hfi1_wq,
 				   &pq->handler);
 	if (ret) {
 		dd_dev_err(dd, "Failed to register with MMU %d", ret);
-		goto done;
+		goto pq_mmu_fail;
 	}
 
+	fd->pq = pq;
+	fd->cq = cq;
+
 	spin_lock_irqsave(&uctxt->sdma_qlock, flags);
 	list_add(&pq->list, &uctxt->sdma_queues);
 	spin_unlock_irqrestore(&uctxt->sdma_qlock, flags);
-	goto done;
 
+	return 0;
+
+pq_mmu_fail:
+	vfree(cq->comps);
 cq_comps_nomem:
 	kfree(cq);
 cq_nomem:
@@ -470,10 +471,7 @@ int hfi1_user_sdma_alloc_queues(struct hfi1_ctxtdata *uctxt,
 	kfree(pq->reqs);
 pq_reqs_nomem:
 	kfree(pq);
-	fd->pq = NULL;
-pq_nomem:
-	ret = -ENOMEM;
-done:
+
 	return ret;
 }
 
diff --git a/drivers/infiniband/hw/hfi1/vnic_main.c b/drivers/infiniband/hw/hfi1/vnic_main.c
index 392f4d5..b601c29 100644
--- a/drivers/infiniband/hw/hfi1/vnic_main.c
+++ b/drivers/infiniband/hw/hfi1/vnic_main.c
@@ -67,9 +67,7 @@ static int setup_vnic_ctxt(struct hfi1_devdata *dd, struct hfi1_ctxtdata *uctxt)
 	unsigned int rcvctrl_ops = 0;
 	int ret;
 
-	ret = hfi1_init_ctxt(uctxt->sc);
-	if (ret)
-		goto done;
+	hfi1_init_ctxt(uctxt->sc);
 
 	uctxt->do_interrupt = &handle_receive_interrupt;
 
@@ -82,8 +80,6 @@ static int setup_vnic_ctxt(struct hfi1_devdata *dd, struct hfi1_ctxtdata *uctxt)
 	if (ret)
 		goto done;
 
-	set_bit(HFI1_CTXT_SETUP_DONE, &uctxt->event_flags);
-
 	if (uctxt->rcvhdrtail_kvaddr)
 		clear_rcvhdrtail(uctxt);
 
@@ -209,7 +205,7 @@ static void deallocate_vnic_ctxt(struct hfi1_devdata *dd,
 	uctxt->event_flags = 0;
 
 	hfi1_clear_tids(uctxt);
-	hfi1_clear_ctxt_pkey(dd, uctxt->ctxt);
+	hfi1_clear_ctxt_pkey(dd, uctxt);
 
 	hfi1_stats.sps_ctxts--;
 	hfi1_free_ctxtdata(dd, uctxt);

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH 15/15] IB/hfi1: Clean up on context initialization failure

[Leon Romanovsky]

[-- Attachment #1: Type: text/plain, Size: 18588 bytes --]

On Tue, May 02, 2017 at 05:42:30PM -0700, Dennis Dalessandro wrote:
> From: Michael J. Ruhl <michael.j.ruhl-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
>
> The error path for context initialization is not consistent. Cleanup all
> resources on failure.
>
> Removed unused variable user_event_mask.
>
> Add the _BASE_FAILED bit to the event flags so that a base context can
> notify waiting sub contexts that they cannot continue.
>
> Running out of sub contexts is an EBUSY result, not EINVAL.
>
> Reviewed-by: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
> Signed-off-by: Michael J. Ruhl <michael.j.ruhl-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
> Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
> ---
>  drivers/infiniband/hw/hfi1/file_ops.c     |  114 +++++++++++------------------
>  drivers/infiniband/hw/hfi1/hfi.h          |   16 ++--
>  drivers/infiniband/hw/hfi1/init.c         |   10 +--
>  drivers/infiniband/hw/hfi1/user_exp_rcv.c |   28 ++++---
>  drivers/infiniband/hw/hfi1/user_sdma.c    |   60 +++++++--------
>  drivers/infiniband/hw/hfi1/vnic_main.c    |    8 +-
>  6 files changed, 103 insertions(+), 133 deletions(-)
>
> diff --git a/drivers/infiniband/hw/hfi1/file_ops.c b/drivers/infiniband/hw/hfi1/file_ops.c
> index 84e6342..0ea842d 100644
> --- a/drivers/infiniband/hw/hfi1/file_ops.c
> +++ b/drivers/infiniband/hw/hfi1/file_ops.c
> @@ -82,7 +82,7 @@
>  static int init_subctxts(struct hfi1_ctxtdata *uctxt,
>  			 const struct hfi1_user_info *uinfo);
>  static int init_user_ctxt(struct hfi1_filedata *fd);
> -static int user_init(struct hfi1_ctxtdata *uctxt);
> +static void user_init(struct hfi1_ctxtdata *uctxt);
>  static int get_ctxt_info(struct hfi1_filedata *fd, void __user *ubase,
>  			 __u32 len);
>  static int get_base_info(struct hfi1_filedata *fd, void __user *ubase,
> @@ -847,7 +847,7 @@ static u64 kvirt_to_phys(void *addr)
>
>  static int assign_ctxt(struct hfi1_filedata *fd, struct hfi1_user_info *uinfo)
>  {
> -	int ret = 0;
> +	int ret;
>  	unsigned int swmajor, swminor;
>
>  	swmajor = uinfo->userversion >> 16;
> @@ -857,14 +857,16 @@ static int assign_ctxt(struct hfi1_filedata *fd, struct hfi1_user_info *uinfo)
>  	swminor = uinfo->userversion & 0xffff;
>
>  	mutex_lock(&hfi1_mutex);
> -	/* First, lets check if we need to get a sub context? */
> +	/*
> +	 * Get a sub context if necessary.
> +	 * ret < 0 error, 0 no context, 1 sub-context found
> +	 */
> +	ret = 0;
>  	if (uinfo->subctxt_cnt) {
> -		/* < 0 error, 0 no context, 1 sub-context found */
>  		ret = find_sub_ctxt(fd, uinfo);
> -		if (ret > 0) {
> +		if (ret > 0)
>  			fd->rec_cpu_num =
>  				hfi1_get_proc_affinity(fd->uctxt->numa_id);
> -		}
>  	}
>
>  	/*
> @@ -885,17 +887,27 @@ static int assign_ctxt(struct hfi1_filedata *fd, struct hfi1_user_info *uinfo)
>  		ret = wait_event_interruptible(fd->uctxt->wait, !test_bit(
>  					       HFI1_CTXT_BASE_UNINIT,
>  					       &fd->uctxt->event_flags));
> +		if (test_bit(HFI1_CTXT_BASE_FAILED, &fd->uctxt->event_flags)) {
> +			clear_bit(fd->subctxt, fd->uctxt->in_use_ctxts);
> +			return -ENOMEM;
> +		}
>  		/* The only thing a sub context needs is the user_xxx stuff */
>  		if (!ret)
> -			init_user_ctxt(fd);
> +			ret = init_user_ctxt(fd);
> +
> +		if (ret)
> +			clear_bit(fd->subctxt, fd->uctxt->in_use_ctxts);
>  	} else if (!ret) {
>  		ret = setup_base_ctxt(fd);
> -
> -		/*
> -		 * Base context is done, notify anybody using a sub-context
> -		 * that is waiting for this completion
> -		 */
> -		if (!ret && fd->uctxt->subctxt_cnt) {
> +		if (fd->uctxt->subctxt_cnt) {
> +			/* If there is an error, set the failed bit. */
> +			if (ret)
> +				set_bit(HFI1_CTXT_BASE_FAILED,
> +					&fd->uctxt->event_flags);
> +			/*
> +			 * Base context is done, notify anybody using a
> +			 * sub-context that is waiting for this completion
> +			 */
>  			clear_bit(HFI1_CTXT_BASE_UNINIT,
>  				  &fd->uctxt->event_flags);
>  			wake_up(&fd->uctxt->wait);
> @@ -916,7 +928,7 @@ static int find_sub_ctxt(struct hfi1_filedata *fd,
>  	struct hfi1_devdata *dd = fd->dd;
>  	u16 subctxt;
>
> -	for (i = dd->first_user_ctxt; i < dd->num_rcv_contexts; i++) {
> +	for (i = dd->first_dyn_alloc_ctxt; i < dd->num_rcv_contexts; i++) {
>  		struct hfi1_ctxtdata *uctxt = dd->rcd[i];
>
>  		/* Skip ctxts which are not yet open */
> @@ -925,6 +937,10 @@ static int find_sub_ctxt(struct hfi1_filedata *fd,
>  				 HFI1_MAX_SHARED_CTXTS))
>  			continue;
>
> +		/* Skip dynamically allocated kernel contexts */
> +		if (uctxt->sc && (uctxt->sc->type == SC_KERNEL))
> +			continue;
> +
>  		/* Skip ctxt if it doesn't match the requested one */
>  		if (memcmp(uctxt->uuid, uinfo->uuid,
>  			   sizeof(uctxt->uuid)) ||
> @@ -932,55 +948,17 @@ static int find_sub_ctxt(struct hfi1_filedata *fd,
>  		    uctxt->subctxt_id != uinfo->subctxt_id ||
>  		    uctxt->subctxt_cnt != uinfo->subctxt_cnt)
>  			continue;
> -<<<<<<< HEAD
> -		for (i = dd->first_dyn_alloc_ctxt;
> -		     i < dd->num_rcv_contexts; i++) {
> -			struct hfi1_ctxtdata *uctxt = dd->rcd[i];
> -
> -			/* Skip ctxts which are not yet open */
> -			if (!uctxt || !uctxt->cnt)
> -				continue;
> -
> -			/* Skip dynamically allocted kernel contexts */
> -			if (uctxt->sc && (uctxt->sc->type == SC_KERNEL))
> -				continue;
> -
> -			/* Skip ctxt if it doesn't match the requested one */
> -			if (memcmp(uctxt->uuid, uinfo->uuid,
> -				   sizeof(uctxt->uuid)) ||
> -			    uctxt->jkey != generate_jkey(current_uid()) ||
> -			    uctxt->subctxt_id != uinfo->subctxt_id ||
> -			    uctxt->subctxt_cnt != uinfo->subctxt_cnt)
> -				continue;
> -
> -			/* Verify the sharing process matches the master */
> -			if (uctxt->userversion != uinfo->userversion ||
> -			    uctxt->cnt >= uctxt->subctxt_cnt) {
> -				ret = -EINVAL;
> -				goto done;
> -			}
> -			fd->uctxt = uctxt;
> -			fd->subctxt  = uctxt->cnt++;
> -			uctxt->active_slaves |= 1 << fd->subctxt;
> -			ret = 1;
> -			goto done;
> -=======
>
>  		/* Verify the sharing process matches the master */
>  		if (uctxt->userversion != uinfo->userversion)
>  			return -EINVAL;
> -<<<<<<< HEAD
> ->>>>>>> f1f8a68... IB/hfi1: Search shared contexts on the opened device, not all devices
> -		}
> -=======
>
>  		/* Find an unused context */
>  		subctxt = find_first_zero_bit(uctxt->in_use_ctxts,
>  					      HFI1_MAX_SHARED_CTXTS);
>  		if (subctxt >= uctxt->subctxt_cnt)
> -			return -EINVAL;
> +			return -EBUSY;
>
> ->>>>>>> d409d3d... IB/hfi1: Fix an assign/ordering issue with shared context IDs

Denny,

You forgot to fix the rebase conflict :)


>  		fd->uctxt = uctxt;
>  		fd->subctxt = subctxt;
>  		__set_bit(fd->subctxt, uctxt->in_use_ctxts);
> @@ -1009,10 +987,6 @@ static int allocate_ctxt(struct hfi1_filedata *fd, struct hfi1_devdata *dd,
>  		return -EIO;
>  	}
>
> -<<<<<<< HEAD
> -	for (ctxt = dd->first_dyn_alloc_ctxt;
> -	     ctxt < dd->num_rcv_contexts; ctxt++)
> -=======
>  	/*
>  	 * This check is sort of redundant to the next EBUSY error. It would
>  	 * also indicate an inconsistancy in the driver if this value was
> @@ -1021,8 +995,8 @@ static int allocate_ctxt(struct hfi1_filedata *fd, struct hfi1_devdata *dd,
>  	if (!dd->freectxts)
>  		return -EBUSY;
>
> -	for (ctxt = dd->first_user_ctxt; ctxt < dd->num_rcv_contexts; ctxt++)
> ->>>>>>> f1f8a68... IB/hfi1: Search shared contexts on the opened device, not all devices
> +	for (ctxt = dd->first_dyn_alloc_ctxt;
> +	     ctxt < dd->num_rcv_contexts; ctxt++)
>  		if (!dd->rcd[ctxt])
>  			break;
>
> @@ -1156,7 +1130,7 @@ static int setup_subctxt(struct hfi1_ctxtdata *uctxt)
>  	return ret;
>  }
>
> -static int user_init(struct hfi1_ctxtdata *uctxt)
> +static void user_init(struct hfi1_ctxtdata *uctxt)
>  {
>  	unsigned int rcvctrl_ops = 0;
>
> @@ -1206,8 +1180,6 @@ static int user_init(struct hfi1_ctxtdata *uctxt)
>  	else
>  		rcvctrl_ops |= HFI1_RCVCTRL_TAILUPD_DIS;
>  	hfi1_rcvctrl(uctxt->dd, rcvctrl_ops, uctxt->ctxt);
> -
> -	return 0;
>  }
>
>  static int get_ctxt_info(struct hfi1_filedata *fd, void __user *ubase,
> @@ -1276,28 +1248,32 @@ static int setup_base_ctxt(struct hfi1_filedata *fd)
>  	/* Now allocate the RcvHdr queue and eager buffers. */
>  	ret = hfi1_create_rcvhdrq(dd, uctxt);
>  	if (ret)
> -		goto done;
> +		return ret;
>
>  	ret = hfi1_setup_eagerbufs(uctxt);
>  	if (ret)
> -		goto done;
> +		goto setup_failed;
>
>  	/* If sub-contexts are enabled, do the appropriate setup */
>  	if (uctxt->subctxt_cnt)
>  		ret = setup_subctxt(uctxt);
>  	if (ret)
> -		goto done;
> +		goto setup_failed;
>
>  	ret = hfi1_user_exp_rcv_grp_init(fd);
>  	if (ret)
> -		goto done;
> +		goto setup_failed;
>
>  	ret = init_user_ctxt(fd);
>  	if (ret)
> -		goto done;
> +		goto setup_failed;
>
> -	ret = user_init(uctxt);
> -done:
> +	user_init(uctxt);
> +
> +	return 0;
> +
> +setup_failed:
> +	hfi1_free_ctxtdata(dd, uctxt);
>  	return ret;
>  }
>
> diff --git a/drivers/infiniband/hw/hfi1/hfi.h b/drivers/infiniband/hw/hfi1/hfi.h
> index f3d75fc..509df98 100644
> --- a/drivers/infiniband/hw/hfi1/hfi.h
> +++ b/drivers/infiniband/hw/hfi1/hfi.h
> @@ -196,12 +196,6 @@ struct hfi1_ctxtdata {
>  	void *rcvhdrq;
>  	/* kernel virtual address where hdrqtail is updated */
>  	volatile __le64 *rcvhdrtail_kvaddr;
> -	/*
> -	 * Shared page for kernel to signal user processes that send buffers
> -	 * need disarming.  The process should call HFI1_CMD_DISARM_BUFS
> -	 * or HFI1_CMD_ACK_EVENT with IPATH_EVENT_DISARM_BUFS set.
> -	 */
> -	unsigned long *user_event_mask;
>  	/* when waiting for rcv or pioavail */
>  	wait_queue_head_t wait;
>  	/* rcvhdrq size (for freeing) */
> @@ -1724,12 +1718,14 @@ struct cc_state *get_cc_state_protected(struct hfi1_pportdata *ppd)
>  #define HFI1_PBC_LENGTH_MASK                     ((1 << 11) - 1)
>
>  /* ctxt_flag bit offsets */
> +		/* base context has not finished initializing */
> +#define HFI1_CTXT_BASE_UNINIT 1
> +		/* base context initaliation failed */
> +#define HFI1_CTXT_BASE_FAILED 2
>  		/* waiting for a packet to arrive */
> -#define HFI1_CTXT_WAITING_RCV   2
> -		/* master has not finished initializing */
> -#define HFI1_CTXT_BASE_UNINIT 4
> +#define HFI1_CTXT_WAITING_RCV 3
>  		/* waiting for an urgent packet to arrive */
> -#define HFI1_CTXT_WAITING_URG 5
> +#define HFI1_CTXT_WAITING_URG 4
>
>  /* free up any allocated data at closes */
>  struct hfi1_devdata *hfi1_init_dd(struct pci_dev *pdev,
> diff --git a/drivers/infiniband/hw/hfi1/init.c b/drivers/infiniband/hw/hfi1/init.c
> index 694a8ec..4a11d4d 100644
> --- a/drivers/infiniband/hw/hfi1/init.c
> +++ b/drivers/infiniband/hw/hfi1/init.c
> @@ -964,7 +964,6 @@ void hfi1_free_ctxtdata(struct hfi1_devdata *dd, struct hfi1_ctxtdata *rcd)
>  	kfree(rcd->egrbufs.buffers);
>
>  	sc_free(rcd->sc);
> -	vfree(rcd->user_event_mask);
>  	vfree(rcd->subctxt_uregbase);
>  	vfree(rcd->subctxt_rcvegrbuf);
>  	vfree(rcd->subctxt_rcvhdr_base);
> @@ -1683,8 +1682,6 @@ int hfi1_create_rcvhdrq(struct hfi1_devdata *dd, struct hfi1_ctxtdata *rcd)
>  	dd_dev_err(dd,
>  		   "attempt to allocate 1 page for ctxt %u rcvhdrqtailaddr failed\n",
>  		   rcd->ctxt);
> -	vfree(rcd->user_event_mask);
> -	rcd->user_event_mask = NULL;
>  	dma_free_coherent(&dd->pcidev->dev, amt, rcd->rcvhdrq,
>  			  rcd->rcvhdrq_dma);
>  	rcd->rcvhdrq = NULL;
> @@ -1851,7 +1848,7 @@ int hfi1_setup_eagerbufs(struct hfi1_ctxtdata *rcd)
>  			  "ctxt%u: current Eager buffer size is invalid %u\n",
>  			  rcd->ctxt, rcd->egrbufs.rcvtid_size);
>  		ret = -EINVAL;
> -		goto bail;
> +		goto bail_rcvegrbuf_phys;
>  	}
>
>  	for (idx = 0; idx < rcd->egrbufs.alloced; idx++) {
> @@ -1859,7 +1856,8 @@ int hfi1_setup_eagerbufs(struct hfi1_ctxtdata *rcd)
>  			     rcd->egrbufs.rcvtids[idx].dma, order);
>  		cond_resched();
>  	}
> -	goto bail;
> +
> +	return 0;
>
>  bail_rcvegrbuf_phys:
>  	for (idx = 0; idx < rcd->egrbufs.alloced &&
> @@ -1873,6 +1871,6 @@ int hfi1_setup_eagerbufs(struct hfi1_ctxtdata *rcd)
>  		rcd->egrbufs.buffers[idx].dma = 0;
>  		rcd->egrbufs.buffers[idx].len = 0;
>  	}
> -bail:
> +
>  	return ret;
>  }
> diff --git a/drivers/infiniband/hw/hfi1/user_exp_rcv.c b/drivers/infiniband/hw/hfi1/user_exp_rcv.c
> index bfaf159..0462d9d 100644
> --- a/drivers/infiniband/hw/hfi1/user_exp_rcv.c
> +++ b/drivers/infiniband/hw/hfi1/user_exp_rcv.c
> @@ -160,6 +160,7 @@ int hfi1_user_exp_rcv_grp_init(struct hfi1_filedata *fd)
>  	struct hfi1_devdata *dd = fd->dd;
>  	u32 tidbase;
>  	u32 i;
> +	struct tid_group *grp, *gptr;
>
>  	exp_tid_group_init(&uctxt->tid_group_list);
>  	exp_tid_group_init(&uctxt->tid_used_list);
> @@ -168,17 +169,10 @@ int hfi1_user_exp_rcv_grp_init(struct hfi1_filedata *fd)
>  	tidbase = uctxt->expected_base;
>  	for (i = 0; i < uctxt->expected_count /
>  		     dd->rcv_entries.group_size; i++) {
> -		struct tid_group *grp;
> -
>  		grp = kzalloc(sizeof(*grp), GFP_KERNEL);
> -		if (!grp) {
> -			/*
> -			 * If we fail here, the groups already
> -			 * allocated will be freed by the close
> -			 * call.
> -			 */
> -			return -ENOMEM;
> -		}
> +		if (!grp)
> +			goto grp_failed;
> +
>  		grp->size = dd->rcv_entries.group_size;
>  		grp->base = tidbase;
>  		tid_group_add_tail(grp, &uctxt->tid_group_list);
> @@ -186,6 +180,15 @@ int hfi1_user_exp_rcv_grp_init(struct hfi1_filedata *fd)
>  	}
>
>  	return 0;
> +
> +grp_failed:
> +	list_for_each_entry_safe(grp, gptr, &uctxt->tid_group_list.list,
> +				 list) {
> +		list_del_init(&grp->list);
> +		kfree(grp);
> +	}
> +
> +	return -ENOMEM;
>  }
>
>  /*
> @@ -213,8 +216,11 @@ int hfi1_user_exp_rcv_init(struct hfi1_filedata *fd)
>  		fd->invalid_tids = kcalloc(uctxt->expected_count,
>  					   sizeof(u32),
>  					   GFP_KERNEL);
> -		if (!fd->invalid_tids)
> +		if (!fd->invalid_tids) {
> +			kfree(fd->entry_to_rb);
> +			fd->entry_to_rb = NULL;
>  			return -ENOMEM;
> +		}
>
>  		/*
>  		 * Register MMU notifier callbacks. If the registration
> diff --git a/drivers/infiniband/hw/hfi1/user_sdma.c b/drivers/infiniband/hw/hfi1/user_sdma.c
> index 6b72267..e38fa88 100644
> --- a/drivers/infiniband/hw/hfi1/user_sdma.c
> +++ b/drivers/infiniband/hw/hfi1/user_sdma.c
> @@ -374,40 +374,25 @@ static void sdma_kmem_cache_ctor(void *obj)
>  int hfi1_user_sdma_alloc_queues(struct hfi1_ctxtdata *uctxt,
>  				struct hfi1_filedata *fd)
>  {
> -	int ret = 0;
> +	int ret = -ENOMEM;
> +	unsigned memsize;
>  	char buf[64];
>  	struct hfi1_devdata *dd;
>  	struct hfi1_user_sdma_comp_q *cq;
>  	struct hfi1_user_sdma_pkt_q *pq;
>  	unsigned long flags;
>
> -	if (!uctxt || !fd) {
> -		ret = -EBADF;
> -		goto done;
> -	}
> +	if (!uctxt || !fd)
> +		return -EBADF;
>
> -	if (!hfi1_sdma_comp_ring_size) {
> -		ret = -EINVAL;
> -		goto done;
> -	}
> +	if (!hfi1_sdma_comp_ring_size)
> +		return -EINVAL;
>
>  	dd = uctxt->dd;
>
>  	pq = kzalloc(sizeof(*pq), GFP_KERNEL);
>  	if (!pq)
> -		goto pq_nomem;
> -
> -	pq->reqs = kcalloc(hfi1_sdma_comp_ring_size,
> -			   sizeof(*pq->reqs),
> -			   GFP_KERNEL);
> -	if (!pq->reqs)
> -		goto pq_reqs_nomem;
> -
> -	pq->req_in_use = kcalloc(BITS_TO_LONGS(hfi1_sdma_comp_ring_size),
> -				 sizeof(*pq->req_in_use),
> -				 GFP_KERNEL);
> -	if (!pq->req_in_use)
> -		goto pq_reqs_no_in_use;
> +		return -ENOMEM;
>
>  	INIT_LIST_HEAD(&pq->list);
>  	pq->dd = dd;
> @@ -423,10 +408,21 @@ int hfi1_user_sdma_alloc_queues(struct hfi1_ctxtdata *uctxt,
>  	iowait_init(&pq->busy, 0, NULL, defer_packet_queue,
>  		    activate_packet_queue, NULL);
>  	pq->reqidx = 0;
> +
> +	memsize = sizeof(*pq->reqs) * hfi1_sdma_comp_ring_size;
> +	pq->reqs = kzalloc(memsize, GFP_KERNEL);
> +	if (!pq->reqs)
> +		goto pq_reqs_nomem;
> +
> +	memsize = BITS_TO_LONGS(hfi1_sdma_comp_ring_size) * sizeof(long);
> +	pq->req_in_use = kzalloc(memsize, GFP_KERNEL);
> +	if (!pq->req_in_use)
> +		goto pq_reqs_no_in_use;
> +
>  	snprintf(buf, 64, "txreq-kmem-cache-%u-%u-%u", dd->unit, uctxt->ctxt,
>  		 fd->subctxt);
>  	pq->txreq_cache = kmem_cache_create(buf,
> -			       sizeof(struct user_sdma_txreq),
> +					    sizeof(struct user_sdma_txreq),
>  					    L1_CACHE_BYTES,
>  					    SLAB_HWCACHE_ALIGN,
>  					    sdma_kmem_cache_ctor);
> @@ -435,7 +431,7 @@ int hfi1_user_sdma_alloc_queues(struct hfi1_ctxtdata *uctxt,
>  			   uctxt->ctxt);
>  		goto pq_txreq_nomem;
>  	}
> -	fd->pq = pq;
> +
>  	cq = kzalloc(sizeof(*cq), GFP_KERNEL);
>  	if (!cq)
>  		goto cq_nomem;
> @@ -446,20 +442,25 @@ int hfi1_user_sdma_alloc_queues(struct hfi1_ctxtdata *uctxt,
>  		goto cq_comps_nomem;
>
>  	cq->nentries = hfi1_sdma_comp_ring_size;
> -	fd->cq = cq;
>
>  	ret = hfi1_mmu_rb_register(pq, pq->mm, &sdma_rb_ops, dd->pport->hfi1_wq,
>  				   &pq->handler);
>  	if (ret) {
>  		dd_dev_err(dd, "Failed to register with MMU %d", ret);
> -		goto done;
> +		goto pq_mmu_fail;
>  	}
>
> +	fd->pq = pq;
> +	fd->cq = cq;
> +
>  	spin_lock_irqsave(&uctxt->sdma_qlock, flags);
>  	list_add(&pq->list, &uctxt->sdma_queues);
>  	spin_unlock_irqrestore(&uctxt->sdma_qlock, flags);
> -	goto done;
>
> +	return 0;
> +
> +pq_mmu_fail:
> +	vfree(cq->comps);
>  cq_comps_nomem:
>  	kfree(cq);
>  cq_nomem:
> @@ -470,10 +471,7 @@ int hfi1_user_sdma_alloc_queues(struct hfi1_ctxtdata *uctxt,
>  	kfree(pq->reqs);
>  pq_reqs_nomem:
>  	kfree(pq);
> -	fd->pq = NULL;
> -pq_nomem:
> -	ret = -ENOMEM;
> -done:
> +
>  	return ret;
>  }
>
> diff --git a/drivers/infiniband/hw/hfi1/vnic_main.c b/drivers/infiniband/hw/hfi1/vnic_main.c
> index 392f4d5..b601c29 100644
> --- a/drivers/infiniband/hw/hfi1/vnic_main.c
> +++ b/drivers/infiniband/hw/hfi1/vnic_main.c
> @@ -67,9 +67,7 @@ static int setup_vnic_ctxt(struct hfi1_devdata *dd, struct hfi1_ctxtdata *uctxt)
>  	unsigned int rcvctrl_ops = 0;
>  	int ret;
>
> -	ret = hfi1_init_ctxt(uctxt->sc);
> -	if (ret)
> -		goto done;
> +	hfi1_init_ctxt(uctxt->sc);
>
>  	uctxt->do_interrupt = &handle_receive_interrupt;
>
> @@ -82,8 +80,6 @@ static int setup_vnic_ctxt(struct hfi1_devdata *dd, struct hfi1_ctxtdata *uctxt)
>  	if (ret)
>  		goto done;
>
> -	set_bit(HFI1_CTXT_SETUP_DONE, &uctxt->event_flags);
> -
>  	if (uctxt->rcvhdrtail_kvaddr)
>  		clear_rcvhdrtail(uctxt);
>
> @@ -209,7 +205,7 @@ static void deallocate_vnic_ctxt(struct hfi1_devdata *dd,
>  	uctxt->event_flags = 0;
>
>  	hfi1_clear_tids(uctxt);
> -	hfi1_clear_ctxt_pkey(dd, uctxt->ctxt);
> +	hfi1_clear_ctxt_pkey(dd, uctxt);
>
>  	hfi1_stats.sps_ctxts--;
>  	hfi1_free_ctxtdata(dd, uctxt);
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCH 14/15] IB/hfi1: Fix an assign/ordering issue with shared context IDs

[Leon Romanovsky]

[-- Attachment #1: Type: text/plain, Size: 11598 bytes --]

On Tue, May 02, 2017 at 05:42:24PM -0700, Dennis Dalessandro wrote:
> From: Michael J. Ruhl <michael.j.ruhl-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
>
> The current algorithm for generating sub-context IDs is FILO.  If the
> contexts are not closed in that order, the uniqueness of the ID will be
> compromised. I.e. logging the creation/deletion of context IDs with an
> application that assigns and closes in a FIFO order reveals:
>
> cache_id: assign: uctxt: 3    sub_ctxt: 0
> cache_id: assign: uctxt: 3    sub_ctxt: 1
> cache_id: assign: uctxt: 3    sub_ctxt: 2
> cache_id: close:  uctxt: 3    sub_ctxt: 0
> cache_id: assign: uctxt: 3    sub_ctxt: 2 <<<
>
> The sub_ctxt ID 2 is reused incorrectly.
>
> Update the sub-context ID assign algorithm to use a bitmask of in_use
> contexts.  The new algorithm will allow the contexts to be closed in any
> order, and will only re-use unused contexts.
>
> Size subctxt and subctxt_cnt to match the user API size.
>
> Reviewed-by: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
> Signed-off-by: Michael J. Ruhl <michael.j.ruhl-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
> Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
> ---
>  drivers/infiniband/hw/hfi1/driver.c    |    2 +
>  drivers/infiniband/hw/hfi1/file_ops.c  |   53 +++++++++++++++++++++-----------
>  drivers/infiniband/hw/hfi1/hfi.h       |    8 ++---
>  drivers/infiniband/hw/hfi1/init.c      |    3 +-
>  drivers/infiniband/hw/hfi1/intr.c      |    3 +-
>  drivers/infiniband/hw/hfi1/user_sdma.h |    2 +
>  6 files changed, 44 insertions(+), 27 deletions(-)
>
> diff --git a/drivers/infiniband/hw/hfi1/driver.c b/drivers/infiniband/hw/hfi1/driver.c
> index 566d152..a50870e 100644
> --- a/drivers/infiniband/hw/hfi1/driver.c
> +++ b/drivers/infiniband/hw/hfi1/driver.c
> @@ -1289,7 +1289,7 @@ int hfi1_reset_device(int unit)
>  	if (dd->rcd)
>  		for (i = dd->first_dyn_alloc_ctxt;
>  		     i < dd->num_rcv_contexts; i++) {
> -			if (!dd->rcd[i] || !dd->rcd[i]->cnt)
> +			if (!dd->rcd[i])
>  				continue;
>  			spin_unlock_irqrestore(&dd->uctxt_lock, flags);
>  			ret = -EBUSY;
> diff --git a/drivers/infiniband/hw/hfi1/file_ops.c b/drivers/infiniband/hw/hfi1/file_ops.c
> index d748e65..84e6342 100644
> --- a/drivers/infiniband/hw/hfi1/file_ops.c
> +++ b/drivers/infiniband/hw/hfi1/file_ops.c
> @@ -49,6 +49,7 @@
>  #include <linux/vmalloc.h>
>  #include <linux/io.h>
>  #include <linux/sched/mm.h>
> +#include <linux/bitmap.h>
>
>  #include <rdma/ib.h>
>
> @@ -95,11 +96,10 @@ static int allocate_ctxt(struct hfi1_filedata *fd, struct hfi1_devdata *dd,
>  			 struct hfi1_user_info *uinfo);
>  static unsigned int poll_urgent(struct file *fp, struct poll_table_struct *pt);
>  static unsigned int poll_next(struct file *fp, struct poll_table_struct *pt);
> -static int user_event_ack(struct hfi1_ctxtdata *uctxt, int subctxt,
> +static int user_event_ack(struct hfi1_ctxtdata *uctxt, u16 subctxt,
>  			  unsigned long events);
> -static int set_ctxt_pkey(struct hfi1_ctxtdata *uctxt, unsigned subctxt,
> -			 u16 pkey);
> -static int manage_rcvq(struct hfi1_ctxtdata *uctxt, unsigned subctxt,
> +static int set_ctxt_pkey(struct hfi1_ctxtdata *uctxt, u16 subctxt, u16 pkey);
> +static int manage_rcvq(struct hfi1_ctxtdata *uctxt, u16 subctxt,
>  		       int start_stop);
>  static int vma_fault(struct vm_fault *vmf);
>  static long hfi1_file_ioctl(struct file *fp, unsigned int cmd,
> @@ -773,8 +773,8 @@ static int hfi1_file_close(struct inode *inode, struct file *fp)
>  			   HFI1_MAX_SHARED_CTXTS) + fdata->subctxt;
>  	*ev = 0;
>
> -	if (--uctxt->cnt) {
> -		uctxt->active_slaves &= ~(1 << fdata->subctxt);
> +	__clear_bit(fdata->subctxt, uctxt->in_use_ctxts);
> +	if (!bitmap_empty(uctxt->in_use_ctxts, HFI1_MAX_SHARED_CTXTS)) {
>  		mutex_unlock(&hfi1_mutex);
>  		goto done;
>  	}
> @@ -868,7 +868,7 @@ static int assign_ctxt(struct hfi1_filedata *fd, struct hfi1_user_info *uinfo)
>  	}
>
>  	/*
> -	 * Allocate a base context f context sharing is not required or we
> +	 * Allocate a base context if context sharing is not required or we
>  	 * couldn't find a sub context.
>  	 */
>  	if (!ret)
> @@ -905,17 +905,24 @@ static int assign_ctxt(struct hfi1_filedata *fd, struct hfi1_user_info *uinfo)
>  	return ret;
>  }
>
> +/*
> + * The hfi1_mutex must be held when this function is called.  It is
> + * necessary to ensure serialized access to the bitmask in_use_ctxts.
> + */
>  static int find_sub_ctxt(struct hfi1_filedata *fd,
>  			 const struct hfi1_user_info *uinfo)
>  {
>  	int i;
>  	struct hfi1_devdata *dd = fd->dd;
> +	u16 subctxt;
>
>  	for (i = dd->first_user_ctxt; i < dd->num_rcv_contexts; i++) {
>  		struct hfi1_ctxtdata *uctxt = dd->rcd[i];
>
>  		/* Skip ctxts which are not yet open */
> -		if (!uctxt || !uctxt->cnt)
> +		if (!uctxt ||
> +		    bitmap_empty(uctxt->in_use_ctxts,
> +				 HFI1_MAX_SHARED_CTXTS))
>  			continue;
>
>  		/* Skip ctxt if it doesn't match the requested one */
> @@ -960,14 +967,24 @@ static int find_sub_ctxt(struct hfi1_filedata *fd,
>  =======
>
>  		/* Verify the sharing process matches the master */
> -		if (uctxt->userversion != uinfo->userversion ||
> -		    uctxt->cnt >= uctxt->subctxt_cnt) {
> +		if (uctxt->userversion != uinfo->userversion)
>  			return -EINVAL;
> +<<<<<<< HEAD
>  >>>>>>> f1f8a68... IB/hfi1: Search shared contexts on the opened device, not all devices
>  		}
> +=======
> +
> +		/* Find an unused context */
> +		subctxt = find_first_zero_bit(uctxt->in_use_ctxts,
> +					      HFI1_MAX_SHARED_CTXTS);
> +		if (subctxt >= uctxt->subctxt_cnt)
> +			return -EINVAL;
> +
> +>>>>>>> d409d3d... IB/hfi1: Fix an assign/ordering issue with shared context IDs
>  		fd->uctxt = uctxt;

Again, rebase conflict.

> -		fd->subctxt  = uctxt->cnt++;
> -		uctxt->active_slaves |= 1 << fd->subctxt;
> +		fd->subctxt = subctxt;
> +		__set_bit(fd->subctxt, uctxt->in_use_ctxts);
> +
>  		return 1;
>  	}
>
> @@ -1089,7 +1106,7 @@ static int allocate_ctxt(struct hfi1_filedata *fd, struct hfi1_devdata *dd,
>  static int init_subctxts(struct hfi1_ctxtdata *uctxt,
>  			 const struct hfi1_user_info *uinfo)
>  {
> -	unsigned num_subctxts;
> +	u16 num_subctxts;
>
>  	num_subctxts = uinfo->subctxt_cnt;
>  	if (num_subctxts > HFI1_MAX_SHARED_CTXTS)
> @@ -1097,7 +1114,6 @@ static int init_subctxts(struct hfi1_ctxtdata *uctxt,
>
>  	uctxt->subctxt_cnt = uinfo->subctxt_cnt;
>  	uctxt->subctxt_id = uinfo->subctxt_id;
> -	uctxt->active_slaves = 1;
>  	uctxt->redirect_seq_cnt = 1;
>  	set_bit(HFI1_CTXT_BASE_UNINIT, &uctxt->event_flags);
>
> @@ -1107,7 +1123,7 @@ static int init_subctxts(struct hfi1_ctxtdata *uctxt,
>  static int setup_subctxt(struct hfi1_ctxtdata *uctxt)
>  {
>  	int ret = 0;
> -	unsigned num_subctxts = uctxt->subctxt_cnt;
> +	u16 num_subctxts = uctxt->subctxt_cnt;
>
>  	uctxt->subctxt_uregbase = vmalloc_user(PAGE_SIZE);
>  	if (!uctxt->subctxt_uregbase)
> @@ -1459,7 +1475,7 @@ int hfi1_set_uevent_bits(struct hfi1_pportdata *ppd, const int evtbit)
>   * overflow conditions.  start_stop==1 re-enables, to be used to
>   * re-init the software copy of the head register
>   */
> -static int manage_rcvq(struct hfi1_ctxtdata *uctxt, unsigned subctxt,
> +static int manage_rcvq(struct hfi1_ctxtdata *uctxt, u16 subctxt,
>  		       int start_stop)
>  {
>  	struct hfi1_devdata *dd = uctxt->dd;
> @@ -1494,7 +1510,7 @@ static int manage_rcvq(struct hfi1_ctxtdata *uctxt, unsigned subctxt,
>   * User process then performs actions appropriate to bit having been
>   * set, if desired, and checks again in future.
>   */
> -static int user_event_ack(struct hfi1_ctxtdata *uctxt, int subctxt,
> +static int user_event_ack(struct hfi1_ctxtdata *uctxt, u16 subctxt,
>  			  unsigned long events)
>  {
>  	int i;
> @@ -1515,8 +1531,7 @@ static int user_event_ack(struct hfi1_ctxtdata *uctxt, int subctxt,
>  	return 0;
>  }
>
> -static int set_ctxt_pkey(struct hfi1_ctxtdata *uctxt, unsigned subctxt,
> -			 u16 pkey)
> +static int set_ctxt_pkey(struct hfi1_ctxtdata *uctxt, u16 subctxt, u16 pkey)
>  {
>  	int ret = -ENOENT, i, intable = 0;
>  	struct hfi1_pportdata *ppd = uctxt->ppd;
> diff --git a/drivers/infiniband/hw/hfi1/hfi.h b/drivers/infiniband/hw/hfi1/hfi.h
> index 1b7203a..f3d75fc 100644
> --- a/drivers/infiniband/hw/hfi1/hfi.h
> +++ b/drivers/infiniband/hw/hfi1/hfi.h
> @@ -228,7 +228,7 @@ struct hfi1_ctxtdata {
>  	unsigned ctxt;
>  	/*
>  	 * non-zero if ctxt can be shared, and defines the maximum number of
> -	 * sub contexts allowed.
> +	 * sub-contexts for this device context.
>  	 */
>  	u16 subctxt_cnt;
>  	/* non-zero if ctxt is being shared. */
> @@ -287,10 +287,10 @@ struct hfi1_ctxtdata {
>  	void *subctxt_rcvegrbuf;
>  	/* An array of pages for the eager header queue entries * N */
>  	void *subctxt_rcvhdr_base;
> +	/* Bitmask of in use context(s) */
> +	DECLARE_BITMAP(in_use_ctxts, HFI1_MAX_SHARED_CTXTS);
>  	/* The version of the library which opened this ctxt */
>  	u32 userversion;
> -	/* Bitmask of active slaves */
> -	u32 active_slaves;
>  	/* Type of packets or conditions we want to poll for */
>  	u16 poll_type;
>  	/* receive packet sequence counter */
> @@ -1239,9 +1239,9 @@ static inline bool hfi1_vnic_is_rsm_full(struct hfi1_devdata *dd, int spare)
>  struct hfi1_filedata {
>  	struct hfi1_devdata *dd;
>  	struct hfi1_ctxtdata *uctxt;
> -	unsigned subctxt;
>  	struct hfi1_user_sdma_comp_q *cq;
>  	struct hfi1_user_sdma_pkt_q *pq;
> +	u16 subctxt;
>  	/* for cpu affinity; -1 if none */
>  	int rec_cpu_num;
>  	u32 tid_n_pinned;
> diff --git a/drivers/infiniband/hw/hfi1/init.c b/drivers/infiniband/hw/hfi1/init.c
> index 52a6364..694a8ec 100644
> --- a/drivers/infiniband/hw/hfi1/init.c
> +++ b/drivers/infiniband/hw/hfi1/init.c
> @@ -53,6 +53,7 @@
>  #include <linux/module.h>
>  #include <linux/printk.h>
>  #include <linux/hrtimer.h>
> +#include <linux/bitmap.h>
>  #include <rdma/rdma_vt.h>
>
>  #include "hfi.h"
> @@ -222,7 +223,7 @@ struct hfi1_ctxtdata *hfi1_create_ctxtdata(struct hfi1_pportdata *ppd, u32 ctxt,
>  		INIT_LIST_HEAD(&rcd->qp_wait_list);
>  		rcd->ppd = ppd;
>  		rcd->dd = dd;
> -		rcd->cnt = 1;
> +		__set_bit(0, rcd->in_use_ctxts);
>  		rcd->ctxt = ctxt;
>  		dd->rcd[ctxt] = rcd;
>  		rcd->numa_id = numa;
> diff --git a/drivers/infiniband/hw/hfi1/intr.c b/drivers/infiniband/hw/hfi1/intr.c
> index 232014d..ba265d0 100644
> --- a/drivers/infiniband/hw/hfi1/intr.c
> +++ b/drivers/infiniband/hw/hfi1/intr.c
> @@ -47,6 +47,7 @@
>
>  #include <linux/pci.h>
>  #include <linux/delay.h>
> +#include <linux/bitmap.h>
>
>  #include "hfi.h"
>  #include "common.h"
> @@ -189,7 +190,7 @@ void handle_user_interrupt(struct hfi1_ctxtdata *rcd)
>  	unsigned long flags;
>
>  	spin_lock_irqsave(&dd->uctxt_lock, flags);
> -	if (!rcd->cnt)
> +	if (bitmap_empty(rcd->in_use_ctxts, HFI1_MAX_SHARED_CTXTS))
>  		goto done;
>
>  	if (test_and_clear_bit(HFI1_CTXT_WAITING_RCV, &rcd->event_flags)) {
> diff --git a/drivers/infiniband/hw/hfi1/user_sdma.h b/drivers/infiniband/hw/hfi1/user_sdma.h
> index 9181d7c..e5b10ae 100644
> --- a/drivers/infiniband/hw/hfi1/user_sdma.h
> +++ b/drivers/infiniband/hw/hfi1/user_sdma.h
> @@ -58,7 +58,7 @@
>  struct hfi1_user_sdma_pkt_q {
>  	struct list_head list;
>  	unsigned ctxt;
> -	unsigned subctxt;
> +	u16 subctxt;
>  	u16 n_max_reqs;
>  	atomic_t n_reqs;
>  	u16 reqidx;
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCH 14/15] IB/hfi1: Fix an assign/ordering issue with shared context IDs

[Dennis Dalessandro]

On 05/03/2017 03:36 AM, Leon Romanovsky wrote:
>> +=======
>> +
>> +		/* Find an unused context */
>> +		subctxt = find_first_zero_bit(uctxt->in_use_ctxts,
>> +					      HFI1_MAX_SHARED_CTXTS);
>> +		if (subctxt >= uctxt->subctxt_cnt)
>> +			return -EINVAL;
>> +
>> +>>>>>>> d409d3d... IB/hfi1: Fix an assign/ordering issue with shared context IDs
>>  		fd->uctxt = uctxt;
>
> Again, rebase conflict.
>

I see what happened. Will get it fixed up and resend just this and the 
following patch.

Thanks for catching this Leon.

-Denny

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html