[PATCH 0/6] IB/hfi1,rdmavt: Few more for 4.9 incl stable fix

[PATCH 0/6] IB/hfi1,rdmavt: Few more for 4.9 incl stable fix

[Dennis Dalessandro]

Doug,

Here are a couple more patches we have collected for 4.9. There is a stable fix
here that fixes a breakage in IPoIB introduced by a change in 4.8.

Patches apply on your k.o/for-4.9-hfi1 branch.

Can also be found in GitHub repo at:
https://github.com/ddalessa/kernel/tree/for-4.9

---

Dennis Dalessandro (1):
      IB/rdmavt: rdmavt can handle non aligned page maps

Jakub Pawlak (1):
      IB/hfi1: Fix integrity check flags default values

Mike Marciniszyn (3):
      IB/hfi1: Add unique txwait_lock for txreq events
      IB/hfi1: Inline sdma_txclean() for verbs pio
      IB/hfi1: Optimize lkey validation structures

Tadeusz Struk (1):
      IB/hfi1: Remove redundant sysfs irq affinity entry


 drivers/infiniband/hw/hfi1/affinity.c    |   72 ------------------------------
 drivers/infiniband/hw/hfi1/affinity.h    |    4 --
 drivers/infiniband/hw/hfi1/hfi.h         |   40 +++++++++++------
 drivers/infiniband/hw/hfi1/iowait.h      |    8 +++
 drivers/infiniband/hw/hfi1/pio.c         |   14 ++----
 drivers/infiniband/hw/hfi1/qp.c          |   11 +++--
 drivers/infiniband/hw/hfi1/sdma.c        |   37 +++++----------
 drivers/infiniband/hw/hfi1/sdma.h        |   12 ++++-
 drivers/infiniband/hw/hfi1/sysfs.c       |   25 ----------
 drivers/infiniband/hw/hfi1/verbs.c       |    4 ++
 drivers/infiniband/hw/hfi1/verbs.h       |   13 +++--
 drivers/infiniband/hw/hfi1/verbs_txreq.c |   13 +++--
 drivers/infiniband/sw/rdmavt/dma.c       |    3 -
 drivers/infiniband/sw/rdmavt/mr.c        |   10 ++--
 include/rdma/rdmavt_mr.h                 |   10 +++-
 15 files changed, 97 insertions(+), 179 deletions(-)

--
-Denny

[PATCH 1/6] IB/hfi1: Add unique txwait_lock for txreq events

[Dennis Dalessandro]

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

Profiling suggests that the read_seqbegin() in
the txreq put logic is colliding with other uses
of the iowait lock.

The packet at a time use of this lock dictates a unique
lock to avoid reader/writer collisions when the number
of vTxWait events is low.

In order to support a unique lock the iowait struct embedded
in the QP is extended to remember the lock that protects the queue
head.

The QP destroy removes that QP from any wait list.  It doesn't
need to know the head because of the linked list API, but it does
need to know the lock required to protect the head.

This also opens up the wait logic to have unique per resources locks
which needs to be in future refinement.

Reviewed-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/iowait.h      |    8 ++++++++
 drivers/infiniband/hw/hfi1/pio.c         |    1 +
 drivers/infiniband/hw/hfi1/qp.c          |   11 ++++++++---
 drivers/infiniband/hw/hfi1/verbs.c       |    4 ++++
 drivers/infiniband/hw/hfi1/verbs.h       |   13 +++++++------
 drivers/infiniband/hw/hfi1/verbs_txreq.c |   13 +++++++------
 6 files changed, 35 insertions(+), 15 deletions(-)

diff --git a/drivers/infiniband/hw/hfi1/iowait.h b/drivers/infiniband/hw/hfi1/iowait.h
index 2ec6ef3..d9740dd 100644
--- a/drivers/infiniband/hw/hfi1/iowait.h
+++ b/drivers/infiniband/hw/hfi1/iowait.h
@@ -64,6 +64,7 @@ struct sdma_engine;
 /**
  * struct iowait - linkage for delayed progress/waiting
  * @list: used to add/insert into QP/PQ wait lists
+ * @lock: uses to record the list head lock
  * @tx_head: overflow list of sdma_txreq's
  * @sleep: no space callback
  * @wakeup: space callback wakeup
@@ -91,6 +92,11 @@ struct sdma_engine;
  * so sleeping is not allowed.
  *
  * The wait_dma member along with the iow
+ *
+ * The lock field is used by waiters to record
+ * the seqlock_t that guards the list head.
+ * Waiters explicity know that, but the destroy
+ * code that unwaits QPs does not.
  */
 
 struct iowait {
@@ -103,6 +109,7 @@ struct iowait {
 		unsigned seq);
 	void (*wakeup)(struct iowait *wait, int reason);
 	void (*sdma_drained)(struct iowait *wait);
+	seqlock_t *lock;
 	struct work_struct iowork;
 	wait_queue_head_t wait_dma;
 	wait_queue_head_t wait_pio;
@@ -141,6 +148,7 @@ static inline void iowait_init(
 	void (*sdma_drained)(struct iowait *wait))
 {
 	wait->count = 0;
+	wait->lock = NULL;
 	INIT_LIST_HEAD(&wait->list);
 	INIT_LIST_HEAD(&wait->tx_head);
 	INIT_WORK(&wait->iowork, func);
diff --git a/drivers/infiniband/hw/hfi1/pio.c b/drivers/infiniband/hw/hfi1/pio.c
index 50a3a36..385e4dc 100644
--- a/drivers/infiniband/hw/hfi1/pio.c
+++ b/drivers/infiniband/hw/hfi1/pio.c
@@ -1580,6 +1580,7 @@ static void sc_piobufavail(struct send_context *sc)
 		qp = iowait_to_qp(wait);
 		priv = qp->priv;
 		list_del_init(&priv->s_iowait.list);
+		priv->s_iowait.lock = NULL;
 		/* refcount held until actual wake up */
 		qps[n++] = qp;
 	}
diff --git a/drivers/infiniband/hw/hfi1/qp.c b/drivers/infiniband/hw/hfi1/qp.c
index 9fc75e7..d752d67 100644
--- a/drivers/infiniband/hw/hfi1/qp.c
+++ b/drivers/infiniband/hw/hfi1/qp.c
@@ -196,15 +196,18 @@ static void flush_tx_list(struct rvt_qp *qp)
 static void flush_iowait(struct rvt_qp *qp)
 {
 	struct hfi1_qp_priv *priv = qp->priv;
-	struct hfi1_ibdev *dev = to_idev(qp->ibqp.device);
 	unsigned long flags;
+	seqlock_t *lock = priv->s_iowait.lock;
 
-	write_seqlock_irqsave(&dev->iowait_lock, flags);
+	if (!lock)
+		return;
+	write_seqlock_irqsave(lock, flags);
 	if (!list_empty(&priv->s_iowait.list)) {
 		list_del_init(&priv->s_iowait.list);
+		priv->s_iowait.lock = NULL;
 		rvt_put_qp(qp);
 	}
-	write_sequnlock_irqrestore(&dev->iowait_lock, flags);
+	write_sequnlock_irqrestore(lock, flags);
 }
 
 static inline int opa_mtu_enum_to_int(int mtu)
@@ -543,6 +546,7 @@ static int iowait_sleep(
 			ibp->rvp.n_dmawait++;
 			qp->s_flags |= RVT_S_WAIT_DMA_DESC;
 			list_add_tail(&priv->s_iowait.list, &sde->dmawait);
+			priv->s_iowait.lock = &dev->iowait_lock;
 			trace_hfi1_qpsleep(qp, RVT_S_WAIT_DMA_DESC);
 			rvt_get_qp(qp);
 		}
@@ -964,6 +968,7 @@ void notify_error_qp(struct rvt_qp *qp)
 	if (!list_empty(&priv->s_iowait.list) && !(qp->s_flags & RVT_S_BUSY)) {
 		qp->s_flags &= ~RVT_S_ANY_WAIT_IO;
 		list_del_init(&priv->s_iowait.list);
+		priv->s_iowait.lock = NULL;
 		rvt_put_qp(qp);
 	}
 	write_sequnlock(&dev->iowait_lock);
diff --git a/drivers/infiniband/hw/hfi1/verbs.c b/drivers/infiniband/hw/hfi1/verbs.c
index f2f6b5a..9c3c237 100644
--- a/drivers/infiniband/hw/hfi1/verbs.c
+++ b/drivers/infiniband/hw/hfi1/verbs.c
@@ -694,6 +694,7 @@ static void mem_timer(unsigned long data)
 		qp = iowait_to_qp(wait);
 		priv = qp->priv;
 		list_del_init(&priv->s_iowait.list);
+		priv->s_iowait.lock = NULL;
 		/* refcount held until actual wake up */
 		if (!list_empty(list))
 			mod_timer(&dev->mem_timer, jiffies + 1);
@@ -769,6 +770,7 @@ static int wait_kmem(struct hfi1_ibdev *dev,
 				mod_timer(&dev->mem_timer, jiffies + 1);
 			qp->s_flags |= RVT_S_WAIT_KMEM;
 			list_add_tail(&priv->s_iowait.list, &dev->memwait);
+			priv->s_iowait.lock = &dev->iowait_lock;
 			trace_hfi1_qpsleep(qp, RVT_S_WAIT_KMEM);
 			rvt_get_qp(qp);
 		}
@@ -980,6 +982,7 @@ static int pio_wait(struct rvt_qp *qp,
 			qp->s_flags |= flag;
 			was_empty = list_empty(&sc->piowait);
 			list_add_tail(&priv->s_iowait.list, &sc->piowait);
+			priv->s_iowait.lock = &dev->iowait_lock;
 			trace_hfi1_qpsleep(qp, RVT_S_WAIT_PIO);
 			rvt_get_qp(qp);
 			/* counting: only call wantpiobuf_intr if first user */
@@ -1631,6 +1634,7 @@ int hfi1_register_ib_device(struct hfi1_devdata *dd)
 	setup_timer(&dev->mem_timer, mem_timer, (unsigned long)dev);
 
 	seqlock_init(&dev->iowait_lock);
+	seqlock_init(&dev->txwait_lock);
 	INIT_LIST_HEAD(&dev->txwait);
 	INIT_LIST_HEAD(&dev->memwait);
 
diff --git a/drivers/infiniband/hw/hfi1/verbs.h b/drivers/infiniband/hw/hfi1/verbs.h
index 1c3815d..7a8af39 100644
--- a/drivers/infiniband/hw/hfi1/verbs.h
+++ b/drivers/infiniband/hw/hfi1/verbs.h
@@ -180,18 +180,19 @@ struct hfi1_ibdev {
 	struct rvt_dev_info rdi; /* Must be first */
 
 	/* QP numbers are shared by all IB ports */
-	/* protect wait lists */
-	seqlock_t iowait_lock;
+	/* protect txwait list */
+	seqlock_t txwait_lock ____cacheline_aligned_in_smp;
 	struct list_head txwait;        /* list for wait verbs_txreq */
 	struct list_head memwait;       /* list for wait kernel memory */
-	struct list_head txreq_free;
 	struct kmem_cache *verbs_txreq_cache;
-	struct timer_list mem_timer;
+	u64 n_txwait;
+	u64 n_kmem_wait;
 
+	/* protect iowait lists */
+	seqlock_t iowait_lock ____cacheline_aligned_in_smp;
 	u64 n_piowait;
 	u64 n_piodrain;
-	u64 n_txwait;
-	u64 n_kmem_wait;
+	struct timer_list mem_timer;
 
 #ifdef CONFIG_DEBUG_FS
 	/* per HFI debugfs */
diff --git a/drivers/infiniband/hw/hfi1/verbs_txreq.c b/drivers/infiniband/hw/hfi1/verbs_txreq.c
index 094ab82..5d23172 100644
--- a/drivers/infiniband/hw/hfi1/verbs_txreq.c
+++ b/drivers/infiniband/hw/hfi1/verbs_txreq.c
@@ -72,22 +72,22 @@ void hfi1_put_txreq(struct verbs_txreq *tx)
 	kmem_cache_free(dev->verbs_txreq_cache, tx);
 
 	do {
-		seq = read_seqbegin(&dev->iowait_lock);
+		seq = read_seqbegin(&dev->txwait_lock);
 		if (!list_empty(&dev->txwait)) {
 			struct iowait *wait;
 
-			write_seqlock_irqsave(&dev->iowait_lock, flags);
+			write_seqlock_irqsave(&dev->txwait_lock, flags);
 			wait = list_first_entry(&dev->txwait, struct iowait,
 						list);
 			qp = iowait_to_qp(wait);
 			priv = qp->priv;
 			list_del_init(&priv->s_iowait.list);
 			/* refcount held until actual wake up */
-			write_sequnlock_irqrestore(&dev->iowait_lock, flags);
+			write_sequnlock_irqrestore(&dev->txwait_lock, flags);
 			hfi1_qp_wakeup(qp, RVT_S_WAIT_TX);
 			break;
 		}
-	} while (read_seqretry(&dev->iowait_lock, seq));
+	} while (read_seqretry(&dev->txwait_lock, seq));
 }
 
 struct verbs_txreq *__get_txreq(struct hfi1_ibdev *dev,
@@ -96,7 +96,7 @@ struct verbs_txreq *__get_txreq(struct hfi1_ibdev *dev,
 {
 	struct verbs_txreq *tx = ERR_PTR(-EBUSY);
 
-	write_seqlock(&dev->iowait_lock);
+	write_seqlock(&dev->txwait_lock);
 	if (ib_rvt_state_ops[qp->state] & RVT_PROCESS_RECV_OK) {
 		struct hfi1_qp_priv *priv;
 
@@ -108,13 +108,14 @@ struct verbs_txreq *__get_txreq(struct hfi1_ibdev *dev,
 			dev->n_txwait++;
 			qp->s_flags |= RVT_S_WAIT_TX;
 			list_add_tail(&priv->s_iowait.list, &dev->txwait);
+			priv->s_iowait.lock = &dev->txwait_lock;
 			trace_hfi1_qpsleep(qp, RVT_S_WAIT_TX);
 			rvt_get_qp(qp);
 		}
 		qp->s_flags &= ~RVT_S_BUSY;
 	}
 out:
-	write_sequnlock(&dev->iowait_lock);
+	write_sequnlock(&dev->txwait_lock);
 	return tx;
 }
 

--
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 2/6] IB/hfi1: Inline sdma_txclean() for verbs pio

[Dennis Dalessandro]

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

Short circuit sdma_txclean() by adding an __sdma_txclean()
that is only called when the tx has sdma mappings.

Convert internal calls to __sdma_txclean().

This removes a call from the critical path.

Reviewed-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/sdma.c |   18 +++++++++---------
 drivers/infiniband/hw/hfi1/sdma.h |   12 +++++++++---
 2 files changed, 18 insertions(+), 12 deletions(-)

diff --git a/drivers/infiniband/hw/hfi1/sdma.c b/drivers/infiniband/hw/hfi1/sdma.c
index fd39bca..7102a07 100644
--- a/drivers/infiniband/hw/hfi1/sdma.c
+++ b/drivers/infiniband/hw/hfi1/sdma.c
@@ -375,7 +375,7 @@ static inline void complete_tx(struct sdma_engine *sde,
 			   sde->head_sn, tx->sn);
 	sde->head_sn++;
 #endif
-	sdma_txclean(sde->dd, tx);
+	__sdma_txclean(sde->dd, tx);
 	if (complete)
 		(*complete)(tx, res);
 	if (wait && iowait_sdma_dec(wait))
@@ -1643,7 +1643,7 @@ static inline u8 ahg_mode(struct sdma_txreq *tx)
 }
 
 /**
- * sdma_txclean() - clean tx of mappings, descp *kmalloc's
+ * __sdma_txclean() - clean tx of mappings, descp *kmalloc's
  * @dd: hfi1_devdata for unmapping
  * @tx: tx request to clean
  *
@@ -1653,7 +1653,7 @@ static inline u8 ahg_mode(struct sdma_txreq *tx)
  * The code can be called multiple times without issue.
  *
  */
-void sdma_txclean(
+void __sdma_txclean(
 	struct hfi1_devdata *dd,
 	struct sdma_txreq *tx)
 {
@@ -3080,7 +3080,7 @@ static int _extend_sdma_tx_descs(struct hfi1_devdata *dd, struct sdma_txreq *tx)
 		tx->descp[i] = tx->descs[i];
 	return 0;
 enomem:
-	sdma_txclean(dd, tx);
+	__sdma_txclean(dd, tx);
 	return -ENOMEM;
 }
 
@@ -3109,14 +3109,14 @@ int ext_coal_sdma_tx_descs(struct hfi1_devdata *dd, struct sdma_txreq *tx,
 
 	rval = _extend_sdma_tx_descs(dd, tx);
 	if (rval) {
-		sdma_txclean(dd, tx);
+		__sdma_txclean(dd, tx);
 		return rval;
 	}
 
 	/* If coalesce buffer is allocated, copy data into it */
 	if (tx->coalesce_buf) {
 		if (type == SDMA_MAP_NONE) {
-			sdma_txclean(dd, tx);
+			__sdma_txclean(dd, tx);
 			return -EINVAL;
 		}
 
@@ -3124,7 +3124,7 @@ int ext_coal_sdma_tx_descs(struct hfi1_devdata *dd, struct sdma_txreq *tx,
 			kvaddr = kmap(page);
 			kvaddr += offset;
 		} else if (WARN_ON(!kvaddr)) {
-			sdma_txclean(dd, tx);
+			__sdma_txclean(dd, tx);
 			return -EINVAL;
 		}
 
@@ -3154,7 +3154,7 @@ int ext_coal_sdma_tx_descs(struct hfi1_devdata *dd, struct sdma_txreq *tx,
 				      DMA_TO_DEVICE);
 
 		if (unlikely(dma_mapping_error(&dd->pcidev->dev, addr))) {
-			sdma_txclean(dd, tx);
+			__sdma_txclean(dd, tx);
 			return -ENOSPC;
 		}
 
@@ -3196,7 +3196,7 @@ int _pad_sdma_tx_descs(struct hfi1_devdata *dd, struct sdma_txreq *tx)
 	if ((unlikely(tx->num_desc == tx->desc_limit))) {
 		rval = _extend_sdma_tx_descs(dd, tx);
 		if (rval) {
-			sdma_txclean(dd, tx);
+			__sdma_txclean(dd, tx);
 			return rval;
 		}
 	}
diff --git a/drivers/infiniband/hw/hfi1/sdma.h b/drivers/infiniband/hw/hfi1/sdma.h
index 56257ea..21f1e28 100644
--- a/drivers/infiniband/hw/hfi1/sdma.h
+++ b/drivers/infiniband/hw/hfi1/sdma.h
@@ -667,7 +667,13 @@ int ext_coal_sdma_tx_descs(struct hfi1_devdata *dd, struct sdma_txreq *tx,
 			   int type, void *kvaddr, struct page *page,
 			   unsigned long offset, u16 len);
 int _pad_sdma_tx_descs(struct hfi1_devdata *, struct sdma_txreq *);
-void sdma_txclean(struct hfi1_devdata *, struct sdma_txreq *);
+void __sdma_txclean(struct hfi1_devdata *, struct sdma_txreq *);
+
+static inline void sdma_txclean(struct hfi1_devdata *dd, struct sdma_txreq *tx)
+{
+	if (tx->num_desc)
+		__sdma_txclean(dd, tx);
+}
 
 /* helpers used by public routines */
 static inline void _sdma_close_tx(struct hfi1_devdata *dd,
@@ -753,7 +759,7 @@ static inline int sdma_txadd_page(
 		       DMA_TO_DEVICE);
 
 	if (unlikely(dma_mapping_error(&dd->pcidev->dev, addr))) {
-		sdma_txclean(dd, tx);
+		__sdma_txclean(dd, tx);
 		return -ENOSPC;
 	}
 
@@ -834,7 +840,7 @@ static inline int sdma_txadd_kvaddr(
 		       DMA_TO_DEVICE);
 
 	if (unlikely(dma_mapping_error(&dd->pcidev->dev, addr))) {
-		sdma_txclean(dd, tx);
+		__sdma_txclean(dd, tx);
 		return -ENOSPC;
 	}
 

--
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 3/6] IB/hfi1: Optimize lkey validation structures

[Dennis Dalessandro]

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

Profiling shows that the key validation is susceptible
to cache line trading when accessing the lkey table.

Fix by separating out the read mostly fields from the write
fields.   In addition the shift amount, which is function
of the lkey table size, is precomputed and stored with the
table pointer.   Since both the shift and table pointer
are in the same read mostly cacheline, this saves a cache
line in this hot path.

Reviewed-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/sw/rdmavt/mr.c |   10 +++++-----
 include/rdma/rdmavt_mr.h          |   10 +++++++---
 2 files changed, 12 insertions(+), 8 deletions(-)

diff --git a/drivers/infiniband/sw/rdmavt/mr.c b/drivers/infiniband/sw/rdmavt/mr.c
index 46b6497..4acf179 100644
--- a/drivers/infiniband/sw/rdmavt/mr.c
+++ b/drivers/infiniband/sw/rdmavt/mr.c
@@ -84,6 +84,7 @@ int rvt_driver_mr_init(struct rvt_dev_info *rdi)
 		lkey_table_size = rdi->dparms.lkey_table_size;
 	}
 	rdi->lkey_table.max = 1 << lkey_table_size;
+	rdi->lkey_table.shift = 32 - lkey_table_size;
 	lk_tab_size = rdi->lkey_table.max * sizeof(*rdi->lkey_table.table);
 	rdi->lkey_table.table = (struct rvt_mregion __rcu **)
 			       vmalloc_node(lk_tab_size, rdi->dparms.node);
@@ -774,7 +775,6 @@ int rvt_lkey_ok(struct rvt_lkey_table *rkt, struct rvt_pd *pd,
 	struct rvt_mregion *mr;
 	unsigned n, m;
 	size_t off;
-	struct rvt_dev_info *dev = ib_to_rvt(pd->ibpd.device);
 
 	/*
 	 * We use LKEY == zero for kernel virtual addresses
@@ -782,6 +782,8 @@ int rvt_lkey_ok(struct rvt_lkey_table *rkt, struct rvt_pd *pd,
 	 */
 	rcu_read_lock();
 	if (sge->lkey == 0) {
+		struct rvt_dev_info *dev = ib_to_rvt(pd->ibpd.device);
+
 		if (pd->user)
 			goto bail;
 		mr = rcu_dereference(dev->dma_mr);
@@ -798,8 +800,7 @@ int rvt_lkey_ok(struct rvt_lkey_table *rkt, struct rvt_pd *pd,
 		isge->n = 0;
 		goto ok;
 	}
-	mr = rcu_dereference(
-		rkt->table[(sge->lkey >> (32 - dev->dparms.lkey_table_size))]);
+	mr = rcu_dereference(rkt->table[sge->lkey >> rkt->shift]);
 	if (unlikely(!mr || atomic_read(&mr->lkey_invalid) ||
 		     mr->lkey != sge->lkey || mr->pd != &pd->ibpd))
 		goto bail;
@@ -899,8 +900,7 @@ int rvt_rkey_ok(struct rvt_qp *qp, struct rvt_sge *sge,
 		goto ok;
 	}
 
-	mr = rcu_dereference(
-		rkt->table[(rkey >> (32 - dev->dparms.lkey_table_size))]);
+	mr = rcu_dereference(rkt->table[rkey >> rkt->shift]);
 	if (unlikely(!mr || atomic_read(&mr->lkey_invalid) ||
 		     mr->lkey != rkey || qp->ibqp.pd != mr->pd))
 		goto bail;
diff --git a/include/rdma/rdmavt_mr.h b/include/rdma/rdmavt_mr.h
index 6b3c6c8..de59de2 100644
--- a/include/rdma/rdmavt_mr.h
+++ b/include/rdma/rdmavt_mr.h
@@ -90,11 +90,15 @@ struct rvt_mregion {
 #define RVT_MAX_LKEY_TABLE_BITS 23
 
 struct rvt_lkey_table {
-	spinlock_t lock; /* protect changes in this struct */
-	u32 next;               /* next unused index (speeds search) */
-	u32 gen;                /* generation count */
+	/* read mostly fields */
 	u32 max;                /* size of the table */
+	u32 shift;              /* lkey/rkey shift */
 	struct rvt_mregion __rcu **table;
+	/* writeable fields */
+	/* protect changes in this struct */
+	spinlock_t lock ____cacheline_aligned_in_smp;
+	u32 next;               /* next unused index (speeds search) */
+	u32 gen;                /* generation count */
 };
 
 /*

--
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 4/6] IB/rdmavt: rdmavt can handle non aligned page maps

[Dennis Dalessandro]

The initial code for rdmavt carried with it a restriction that was a
vestige from the qib driver, that to dma map a page it had to be less
than a page size. This is not the case on modern hardware, both qib and
hfi1 will be just fine with unaligned map requests.

This fixes a 4.8 regression where by an IPoIB transfer of > PAGE_SIZE
will hang because the dma map page call always fails. This was
introduced after commit 5faba5469522 ("IB/ipoib: Report SG feature
regardless of HW UD CSUM capability") added the capability to use SG by
default. Rather than override this, the HW supports it, so allow SG.

Cc: Stable <stable-u79uwXL29TY76Z2rM5mHXA@public.gmane.org> # 4.8
Reviewed-by: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/infiniband/sw/rdmavt/dma.c |    3 ---
 1 files changed, 0 insertions(+), 3 deletions(-)

diff --git a/drivers/infiniband/sw/rdmavt/dma.c b/drivers/infiniband/sw/rdmavt/dma.c
index 01f71ca..f2cefb0 100644
--- a/drivers/infiniband/sw/rdmavt/dma.c
+++ b/drivers/infiniband/sw/rdmavt/dma.c
@@ -90,9 +90,6 @@ static u64 rvt_dma_map_page(struct ib_device *dev, struct page *page,
 	if (WARN_ON(!valid_dma_direction(direction)))
 		return BAD_DMA_ADDRESS;
 
-	if (offset + size > PAGE_SIZE)
-		return BAD_DMA_ADDRESS;
-
 	addr = (u64)page_address(page);
 	if (addr)
 		addr += offset;

--
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 4/6] IB/rdmavt: rdmavt can handle non aligned page maps

[Dennis Dalessandro]

The initial code for rdmavt carried with it a restriction that was a
vestige from the qib driver, that to dma map a page it had to be less
than a page size. This is not the case on modern hardware, both qib and
hfi1 will be just fine with unaligned map requests.

This fixes a 4.8 regression where by an IPoIB transfer of > PAGE_SIZE
will hang because the dma map page call always fails. This was
introduced after commit 5faba5469522 ("IB/ipoib: Report SG feature
regardless of HW UD CSUM capability") added the capability to use SG by
default. Rather than override this, the HW supports it, so allow SG.

Cc: Stable <stable@vger.kernel.org> # 4.8
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
---
 drivers/infiniband/sw/rdmavt/dma.c |    3 ---
 1 files changed, 0 insertions(+), 3 deletions(-)

diff --git a/drivers/infiniband/sw/rdmavt/dma.c b/drivers/infiniband/sw/rdmavt/dma.c
index 01f71ca..f2cefb0 100644
--- a/drivers/infiniband/sw/rdmavt/dma.c
+++ b/drivers/infiniband/sw/rdmavt/dma.c
@@ -90,9 +90,6 @@ static u64 rvt_dma_map_page(struct ib_device *dev, struct page *page,
 	if (WARN_ON(!valid_dma_direction(direction)))
 		return BAD_DMA_ADDRESS;
 
-	if (offset + size > PAGE_SIZE)
-		return BAD_DMA_ADDRESS;
-
 	addr = (u64)page_address(page);
 	if (addr)
 		addr += offset;

[PATCH 5/6] IB/hfi1: Remove redundant sysfs irq affinity entry

[Dennis Dalessandro]

From: Tadeusz Struk <tadeusz.struk-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

The IRQ affinity entry is not needed after the irq notifier patch has been
added to the hfi1 driver.
The irq affinity settings for SDMA engine should be set using the standard
/proc/irq/<N>/ interface.

Reviewed-by: Jianxin Xiong <jianxin.xiong-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Tadeusz Struk <tadeusz.struk-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/infiniband/hw/hfi1/affinity.c |   72 ---------------------------------
 drivers/infiniband/hw/hfi1/affinity.h |    4 --
 drivers/infiniband/hw/hfi1/sysfs.c    |   25 -----------
 3 files changed, 0 insertions(+), 101 deletions(-)

diff --git a/drivers/infiniband/hw/hfi1/affinity.c b/drivers/infiniband/hw/hfi1/affinity.c
index a26a9a0..67ea85a 100644
--- a/drivers/infiniband/hw/hfi1/affinity.c
+++ b/drivers/infiniband/hw/hfi1/affinity.c
@@ -775,75 +775,3 @@ void hfi1_put_proc_affinity(int cpu)
 	}
 	mutex_unlock(&affinity->lock);
 }
-
-int hfi1_set_sdma_affinity(struct hfi1_devdata *dd, const char *buf,
-			   size_t count)
-{
-	struct hfi1_affinity_node *entry;
-	cpumask_var_t mask;
-	int ret, i;
-
-	mutex_lock(&node_affinity.lock);
-	entry = node_affinity_lookup(dd->node);
-
-	if (!entry) {
-		ret = -EINVAL;
-		goto unlock;
-	}
-
-	ret = zalloc_cpumask_var(&mask, GFP_KERNEL);
-	if (!ret) {
-		ret = -ENOMEM;
-		goto unlock;
-	}
-
-	ret = cpulist_parse(buf, mask);
-	if (ret)
-		goto out;
-
-	if (!cpumask_subset(mask, cpu_online_mask) || cpumask_empty(mask)) {
-		dd_dev_warn(dd, "Invalid CPU mask\n");
-		ret = -EINVAL;
-		goto out;
-	}
-
-	/* reset the SDMA interrupt affinity details */
-	init_cpu_mask_set(&entry->def_intr);
-	cpumask_copy(&entry->def_intr.mask, mask);
-
-	/* Reassign the affinity for each SDMA interrupt. */
-	for (i = 0; i < dd->num_msix_entries; i++) {
-		struct hfi1_msix_entry *msix;
-
-		msix = &dd->msix_entries[i];
-		if (msix->type != IRQ_SDMA)
-			continue;
-
-		ret = get_irq_affinity(dd, msix);
-
-		if (ret)
-			break;
-	}
-out:
-	free_cpumask_var(mask);
-unlock:
-	mutex_unlock(&node_affinity.lock);
-	return ret ? ret : strnlen(buf, PAGE_SIZE);
-}
-
-int hfi1_get_sdma_affinity(struct hfi1_devdata *dd, char *buf)
-{
-	struct hfi1_affinity_node *entry;
-
-	mutex_lock(&node_affinity.lock);
-	entry = node_affinity_lookup(dd->node);
-
-	if (!entry) {
-		mutex_unlock(&node_affinity.lock);
-		return -EINVAL;
-	}
-
-	cpumap_print_to_pagebuf(true, buf, &entry->def_intr.mask);
-	mutex_unlock(&node_affinity.lock);
-	return strnlen(buf, PAGE_SIZE);
-}
diff --git a/drivers/infiniband/hw/hfi1/affinity.h b/drivers/infiniband/hw/hfi1/affinity.h
index b89ea3c..42e6331 100644
--- a/drivers/infiniband/hw/hfi1/affinity.h
+++ b/drivers/infiniband/hw/hfi1/affinity.h
@@ -102,10 +102,6 @@ int hfi1_get_proc_affinity(int);
 /* Release a CPU used by a user process. */
 void hfi1_put_proc_affinity(int);
 
-int hfi1_get_sdma_affinity(struct hfi1_devdata *dd, char *buf);
-int hfi1_set_sdma_affinity(struct hfi1_devdata *dd, const char *buf,
-			   size_t count);
-
 struct hfi1_affinity_node {
 	int node;
 	struct cpu_mask_set def_intr;
diff --git a/drivers/infiniband/hw/hfi1/sysfs.c b/drivers/infiniband/hw/hfi1/sysfs.c
index edba224..919a547 100644
--- a/drivers/infiniband/hw/hfi1/sysfs.c
+++ b/drivers/infiniband/hw/hfi1/sysfs.c
@@ -49,7 +49,6 @@
 #include "hfi.h"
 #include "mad.h"
 #include "trace.h"
-#include "affinity.h"
 
 /*
  * Start of per-port congestion control structures and support code
@@ -623,27 +622,6 @@ static ssize_t show_tempsense(struct device *device,
 	return ret;
 }
 
-static ssize_t show_sdma_affinity(struct device *device,
-				  struct device_attribute *attr, char *buf)
-{
-	struct hfi1_ibdev *dev =
-		container_of(device, struct hfi1_ibdev, rdi.ibdev.dev);
-	struct hfi1_devdata *dd = dd_from_dev(dev);
-
-	return hfi1_get_sdma_affinity(dd, buf);
-}
-
-static ssize_t store_sdma_affinity(struct device *device,
-				   struct device_attribute *attr,
-				   const char *buf, size_t count)
-{
-	struct hfi1_ibdev *dev =
-		container_of(device, struct hfi1_ibdev, rdi.ibdev.dev);
-	struct hfi1_devdata *dd = dd_from_dev(dev);
-
-	return hfi1_set_sdma_affinity(dd, buf, count);
-}
-
 /*
  * end of per-unit (or driver, in some cases, but replicated
  * per unit) functions
@@ -658,8 +636,6 @@ static DEVICE_ATTR(serial, S_IRUGO, show_serial, NULL);
 static DEVICE_ATTR(boardversion, S_IRUGO, show_boardversion, NULL);
 static DEVICE_ATTR(tempsense, S_IRUGO, show_tempsense, NULL);
 static DEVICE_ATTR(chip_reset, S_IWUSR, NULL, store_chip_reset);
-static DEVICE_ATTR(sdma_affinity, S_IWUSR | S_IRUGO, show_sdma_affinity,
-		   store_sdma_affinity);
 
 static struct device_attribute *hfi1_attributes[] = {
 	&dev_attr_hw_rev,
@@ -670,7 +646,6 @@ static struct device_attribute *hfi1_attributes[] = {
 	&dev_attr_boardversion,
 	&dev_attr_tempsense,
 	&dev_attr_chip_reset,
-	&dev_attr_sdma_affinity,
 };
 
 int hfi1_create_port_files(struct ib_device *ibdev, u8 port_num,

--
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 6/6] IB/hfi1: Fix integrity check flags default values

[Dennis Dalessandro]

From: Jakub Pawlak <jakub.pawlak-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

Prevent setting up integrity check flags when module is loaded
with NO_INTEGRITY capability.

Reviewed-by: Dean Luick <dean.luick-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Jakub Pawlak <jakub.pawlak-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/infiniband/hw/hfi1/hfi.h  |   40 +++++++++++++++++++++++++------------
 drivers/infiniband/hw/hfi1/pio.c  |   13 +++---------
 drivers/infiniband/hw/hfi1/sdma.c |   19 ++----------------
 3 files changed, 32 insertions(+), 40 deletions(-)

diff --git a/drivers/infiniband/hw/hfi1/hfi.h b/drivers/infiniband/hw/hfi1/hfi.h
index 7eef11b..3c06d20 100644
--- a/drivers/infiniband/hw/hfi1/hfi.h
+++ b/drivers/infiniband/hw/hfi1/hfi.h
@@ -1848,7 +1848,13 @@ extern struct mutex hfi1_mutex;
 static inline u64 hfi1_pkt_default_send_ctxt_mask(struct hfi1_devdata *dd,
 						  u16 ctxt_type)
 {
-	u64 base_sc_integrity =
+	u64 base_sc_integrity;
+
+	/* No integrity checks if HFI1_CAP_NO_INTEGRITY is set */
+	if (HFI1_CAP_IS_KSET(NO_INTEGRITY))
+		return 0;
+
+	base_sc_integrity =
 	SEND_CTXT_CHECK_ENABLE_DISALLOW_BYPASS_BAD_PKT_LEN_SMASK
 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_PBC_STATIC_RATE_CONTROL_SMASK
 	| SEND_CTXT_CHECK_ENABLE_DISALLOW_TOO_LONG_BYPASS_PACKETS_SMASK
@@ -1863,7 +1869,6 @@ static inline u64 hfi1_pkt_default_send_ctxt_mask(struct hfi1_devdata *dd,
 	| SEND_CTXT_CHECK_ENABLE_CHECK_VL_MAPPING_SMASK
 	| SEND_CTXT_CHECK_ENABLE_CHECK_OPCODE_SMASK
 	| SEND_CTXT_CHECK_ENABLE_CHECK_SLID_SMASK
-	| SEND_CTXT_CHECK_ENABLE_CHECK_JOB_KEY_SMASK
 	| SEND_CTXT_CHECK_ENABLE_CHECK_VL_SMASK
 	| SEND_CTXT_CHECK_ENABLE_CHECK_ENABLE_SMASK;
 
@@ -1872,18 +1877,23 @@ static inline u64 hfi1_pkt_default_send_ctxt_mask(struct hfi1_devdata *dd,
 	else
 		base_sc_integrity |= HFI1_PKT_KERNEL_SC_INTEGRITY;
 
-	if (is_ax(dd))
-		/* turn off send-side job key checks - A0 */
-		return base_sc_integrity &
-		       ~SEND_CTXT_CHECK_ENABLE_CHECK_JOB_KEY_SMASK;
+	/* turn on send-side job key checks if !A0 */
+	if (!is_ax(dd))
+		base_sc_integrity |= SEND_CTXT_CHECK_ENABLE_CHECK_JOB_KEY_SMASK;
+
 	return base_sc_integrity;
 }
 
 static inline u64 hfi1_pkt_base_sdma_integrity(struct hfi1_devdata *dd)
 {
-	u64 base_sdma_integrity =
+	u64 base_sdma_integrity;
+
+	/* No integrity checks if HFI1_CAP_NO_INTEGRITY is set */
+	if (HFI1_CAP_IS_KSET(NO_INTEGRITY))
+		return 0;
+
+	base_sdma_integrity =
 	SEND_DMA_CHECK_ENABLE_DISALLOW_BYPASS_BAD_PKT_LEN_SMASK
-	| SEND_DMA_CHECK_ENABLE_DISALLOW_PBC_STATIC_RATE_CONTROL_SMASK
 	| SEND_DMA_CHECK_ENABLE_DISALLOW_TOO_LONG_BYPASS_PACKETS_SMASK
 	| SEND_DMA_CHECK_ENABLE_DISALLOW_TOO_LONG_IB_PACKETS_SMASK
 	| SEND_DMA_CHECK_ENABLE_DISALLOW_BAD_PKT_LEN_SMASK
@@ -1895,14 +1905,18 @@ static inline u64 hfi1_pkt_base_sdma_integrity(struct hfi1_devdata *dd)
 	| SEND_DMA_CHECK_ENABLE_CHECK_VL_MAPPING_SMASK
 	| SEND_DMA_CHECK_ENABLE_CHECK_OPCODE_SMASK
 	| SEND_DMA_CHECK_ENABLE_CHECK_SLID_SMASK
-	| SEND_DMA_CHECK_ENABLE_CHECK_JOB_KEY_SMASK
 	| SEND_DMA_CHECK_ENABLE_CHECK_VL_SMASK
 	| SEND_DMA_CHECK_ENABLE_CHECK_ENABLE_SMASK;
 
-	if (is_ax(dd))
-		/* turn off send-side job key checks - A0 */
-		return base_sdma_integrity &
-		       ~SEND_DMA_CHECK_ENABLE_CHECK_JOB_KEY_SMASK;
+	if (!HFI1_CAP_IS_KSET(STATIC_RATE_CTRL))
+		base_sdma_integrity |=
+		SEND_DMA_CHECK_ENABLE_DISALLOW_PBC_STATIC_RATE_CONTROL_SMASK;
+
+	/* turn on send-side job key checks if !A0 */
+	if (!is_ax(dd))
+		base_sdma_integrity |=
+			SEND_DMA_CHECK_ENABLE_CHECK_JOB_KEY_SMASK;
+
 	return base_sdma_integrity;
 }
 
diff --git a/drivers/infiniband/hw/hfi1/pio.c b/drivers/infiniband/hw/hfi1/pio.c
index 385e4dc..e60fe3e 100644
--- a/drivers/infiniband/hw/hfi1/pio.c
+++ b/drivers/infiniband/hw/hfi1/pio.c
@@ -668,19 +668,12 @@ void sc_set_cr_threshold(struct send_context *sc, u32 new_threshold)
 void set_pio_integrity(struct send_context *sc)
 {
 	struct hfi1_devdata *dd = sc->dd;
-	u64 reg = 0;
 	u32 hw_context = sc->hw_context;
 	int type = sc->type;
 
-	/*
-	 * No integrity checks if HFI1_CAP_NO_INTEGRITY is set, or if
-	 * we're snooping.
-	 */
-	if (likely(!HFI1_CAP_IS_KSET(NO_INTEGRITY)) &&
-	    dd->hfi1_snoop.mode_flag != HFI1_PORT_SNOOP_MODE)
-		reg = hfi1_pkt_default_send_ctxt_mask(dd, type);
-
-	write_kctxt_csr(dd, hw_context, SC(CHECK_ENABLE), reg);
+	write_kctxt_csr(dd, hw_context,
+			SC(CHECK_ENABLE),
+			hfi1_pkt_default_send_ctxt_mask(dd, type));
 }
 
 static u32 get_buffers_allocated(struct send_context *sc)
diff --git a/drivers/infiniband/hw/hfi1/sdma.c b/drivers/infiniband/hw/hfi1/sdma.c
index 7102a07..1d81cac 100644
--- a/drivers/infiniband/hw/hfi1/sdma.c
+++ b/drivers/infiniband/hw/hfi1/sdma.c
@@ -2009,11 +2009,6 @@ static void sdma_hw_start_up(struct sdma_engine *sde)
 	write_sde_csr(sde, SD(ENG_ERR_CLEAR), reg);
 }
 
-#define CLEAR_STATIC_RATE_CONTROL_SMASK(r) \
-(r &= ~SEND_DMA_CHECK_ENABLE_DISALLOW_PBC_STATIC_RATE_CONTROL_SMASK)
-
-#define SET_STATIC_RATE_CONTROL_SMASK(r) \
-(r |= SEND_DMA_CHECK_ENABLE_DISALLOW_PBC_STATIC_RATE_CONTROL_SMASK)
 /*
  * set_sdma_integrity
  *
@@ -2022,19 +2017,9 @@ static void sdma_hw_start_up(struct sdma_engine *sde)
 static void set_sdma_integrity(struct sdma_engine *sde)
 {
 	struct hfi1_devdata *dd = sde->dd;
-	u64 reg;
-
-	if (unlikely(HFI1_CAP_IS_KSET(NO_INTEGRITY)))
-		return;
-
-	reg = hfi1_pkt_base_sdma_integrity(dd);
-
-	if (HFI1_CAP_IS_KSET(STATIC_RATE_CTRL))
-		CLEAR_STATIC_RATE_CONTROL_SMASK(reg);
-	else
-		SET_STATIC_RATE_CONTROL_SMASK(reg);
 
-	write_sde_csr(sde, SD(CHECK_ENABLE), reg);
+	write_sde_csr(sde, SD(CHECK_ENABLE),
+		      hfi1_pkt_base_sdma_integrity(dd));
 }
 
 static void init_sdma_regs(

--
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