IB/rdmavt: Compress adjacent SGEs in rvt_lkey_ok()

SGEs that are contiguous needlessly consume driver dependent TX resources.

The lkey validation logic is enhanced to compress the SGE that ends
up in the send wqe when consecutive addresses are detected.

The lkey validation API used to return 1 (success) or 0 (fail).

The return value is now an -errno, 0 (compressed), or 1 (uncompressed).  A
additional argument is added to pass the last SQE for the compression.

Loopback callers always pass a NULL to last_sge since the optimization is
of little benefit in that situation.

Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Brian Welty <brian.welty@intel.com>
Signed-off-by: Venkata Sandeep Dhanalakota <venkata.s.dhanalakota@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>

1 files changed, 44 insertions, 7 deletions

diff --git a/drivers/infiniband/sw/rdmavt/mr.c b/drivers/infiniband/sw/rdmavt/mr.c
index aa5f9ea318e4..ea95672d9675 100644
--- a/drivers/infiniband/sw/rdmavt/mr.c
+++ b/drivers/infiniband/sw/rdmavt/mr.c
@@ -778,23 +778,54 @@ out:
 }
 
 /**
+ * rvt_sge_adjacent - is isge compressible
+ * @isge: outgoing internal SGE
+ * @last_sge: last outgoing SGE written
+ * @sge: SGE to check
+ *
+ * If adjacent will update last_sge to add length.
+ *
+ * Return: true if isge is adjacent to last sge
+ */
+static inline bool rvt_sge_adjacent(struct rvt_sge *isge,
+				    struct rvt_sge *last_sge,
+				    struct ib_sge *sge)
+{
+	if (last_sge && sge->lkey == last_sge->mr->lkey &&
+	    ((uint64_t)(last_sge->vaddr + last_sge->length) == sge->addr)) {
+		if (sge->lkey) {
+			if (unlikely((sge->addr - last_sge->mr->user_base +
+			      sge->length > last_sge->mr->length)))
+				return false; /* overrun, caller will catch */
+		} else {
+			last_sge->length += sge->length;
+		}
+		last_sge->sge_length += sge->length;
+		trace_rvt_sge_adjacent(last_sge, sge);
+		return true;
+	}
+	return false;
+}
+
+/**
  * rvt_lkey_ok - check IB SGE for validity and initialize
  * @rkt: table containing lkey to check SGE against
  * @pd: protection domain
  * @isge: outgoing internal SGE
+ * @last_sge: last outgoing SGE written
  * @sge: SGE to check
  * @acc: access flags
  *
  * Check the IB SGE for validity and initialize our internal version
  * of it.
  *
- * Return: 1 if valid and successful, otherwise returns 0.
- *
- * increments the reference count upon success
+ * Increments the reference count when a new sge is stored.
  *
+ * Return: 0 if compressed, 1 if added , otherwise returns -errno.
  */
 int rvt_lkey_ok(struct rvt_lkey_table *rkt, struct rvt_pd *pd,
-		struct rvt_sge *isge, struct ib_sge *sge, int acc)
+		struct rvt_sge *isge, struct rvt_sge *last_sge,
+		struct ib_sge *sge, int acc)
 {
 	struct rvt_mregion *mr;
 	unsigned n, m;
@@ -804,12 +835,14 @@ int rvt_lkey_ok(struct rvt_lkey_table *rkt, struct rvt_pd *pd,
 	 * We use LKEY == zero for kernel virtual addresses
 	 * (see rvt_get_dma_mr() and dma_virt_ops).
 	 */
-	rcu_read_lock();
 	if (sge->lkey == 0) {
 		struct rvt_dev_info *dev = ib_to_rvt(pd->ibpd.device);
 
 		if (pd->user)
-			goto bail;
+			return -EINVAL;
+		if (rvt_sge_adjacent(isge, last_sge, sge))
+			return 0;
+		rcu_read_lock();
 		mr = rcu_dereference(dev->dma_mr);
 		if (!mr)
 			goto bail;
@@ -824,6 +857,9 @@ int rvt_lkey_ok(struct rvt_lkey_table *rkt, struct rvt_pd *pd,
 		isge->n = 0;
 		goto ok;
 	}
+	if (rvt_sge_adjacent(isge, last_sge, sge))
+		return 0;
+	rcu_read_lock();
 	mr = rcu_dereference(rkt->table[sge->lkey >> rkt->shift]);
 	if (!mr)
 		goto bail;
@@ -874,12 +910,13 @@ int rvt_lkey_ok(struct rvt_lkey_table *rkt, struct rvt_pd *pd,
 	isge->m = m;
 	isge->n = n;
 ok:
+	trace_rvt_sge_new(isge, sge);
 	return 1;
 bail_unref:
 	rvt_put_mr(mr);
 bail:
 	rcu_read_unlock();
-	return 0;
+	return -EINVAL;
 }
 EXPORT_SYMBOL(rvt_lkey_ok);