IB/hfi1: Add the capability for reserved operations

This fix allows for support of in-kernel reserved operations
without impacting the ULP user.

The low level driver can register a non-zero value which
will be transparently added to the send queue size and hidden
from the ULP in every respect.

ULP post sends will never see a full queue due to a reserved
post send and reserved operations will never exceed that
registered value.

The s_avail will continue to track the ULP swqe availability
and the difference between the reserved value and the reserved
in use will track reserved availabity.

Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

2 files changed, 51 insertions, 0 deletions

diff --git a/include/rdma/rdma_vt.h b/include/rdma/rdma_vt.h
index 7fdba92d4c05..e31502107a58 100644
--- a/include/rdma/rdma_vt.h
+++ b/include/rdma/rdma_vt.h
@@ -158,6 +158,7 @@ struct rvt_driver_params {
 	u32 max_mad_size;
 	u8 qos_shift;
 	u8 max_rdma_atomic;
+	u8 reserved_operations;
 };
 
 /* Protection domain */
diff --git a/include/rdma/rdmavt_qp.h b/include/rdma/rdmavt_qp.h
index b0ab12b30f1e..56adcfcabe0b 100644
--- a/include/rdma/rdmavt_qp.h
+++ b/include/rdma/rdmavt_qp.h
@@ -145,6 +145,11 @@
 	(RVT_PROCESS_SEND_OK | RVT_FLUSH_SEND)
 
 /*
+ * Internal send flags
+ */
+#define RVT_SEND_RESERVE_USED           IB_SEND_RESERVED_START
+
+/*
  * Send work request queue entry.
  * The size of the sg_list is determined when the QP is created and stored
  * in qp->s_max_sge.
@@ -232,6 +237,7 @@ struct rvt_ack_entry {
 #define RVT_OPERATION_ATOMIC      0x00000002
 #define RVT_OPERATION_ATOMIC_SGE  0x00000004
 #define RVT_OPERATION_LOCAL       0x00000008
+#define RVT_OPERATION_USE_RESERVE 0x00000010
 
 #define RVT_OPERATION_MAX (IB_WR_RESERVED10 + 1)
 
@@ -328,6 +334,7 @@ struct rvt_qp {
 	u32 s_next_psn;         /* PSN for next request */
 	u32 s_avail;            /* number of entries avail */
 	u32 s_ssn;              /* SSN of tail entry */
+	atomic_t s_reserved_used; /* reserved entries in use */
 
 	spinlock_t s_lock ____cacheline_aligned_in_smp;
 	u32 s_flags;
@@ -459,6 +466,49 @@ static inline struct rvt_rwqe *rvt_get_rwqe_ptr(struct rvt_rq *rq, unsigned n)
 		  rq->max_sge * sizeof(struct ib_sge)) * n);
 }
 
+/**
+ * rvt_qp_wqe_reserve - reserve operation
+ * @qp - the rvt qp
+ * @wqe - the send wqe
+ *
+ * This routine used in post send to record
+ * a wqe relative reserved operation use.
+ */
+static inline void rvt_qp_wqe_reserve(
+	struct rvt_qp *qp,
+	struct rvt_swqe *wqe)
+{
+	wqe->wr.send_flags |= RVT_SEND_RESERVE_USED;
+	atomic_inc(&qp->s_reserved_used);
+}
+
+/**
+ * rvt_qp_wqe_unreserve - clean reserved operation
+ * @qp - the rvt qp
+ * @wqe - the send wqe
+ *
+ * This decrements the reserve use count.
+ *
+ * This call MUST precede the change to
+ * s_last to insure that post send sees a stable
+ * s_avail.
+ *
+ * An smp_mp__after_atomic() is used to insure
+ * the compiler does not juggle the order of the s_last
+ * ring index and the decrementing of s_reserved_used.
+ */
+static inline void rvt_qp_wqe_unreserve(
+	struct rvt_qp *qp,
+	struct rvt_swqe *wqe)
+{
+	if (unlikely(wqe->wr.send_flags & RVT_SEND_RESERVE_USED)) {
+		wqe->wr.send_flags &= ~RVT_SEND_RESERVE_USED;
+		atomic_dec(&qp->s_reserved_used);
+		/* insure no compiler re-order up to s_last change */
+		smp_mb__after_atomic();
+	}
+}
+
 extern const int  ib_rvt_state_ops[];
 
 struct rvt_dev_info;