[PATCH AUTOSEL 4.19 016/252] RDMA/netlink: Do not always generate an ACK for some netlink operations

[PATCH AUTOSEL 4.19 016/252] RDMA/netlink: Do not always generate an ACK for some netlink operations

[Sasha Levin]

From: Håkon Bugge <haakon.bugge@oracle.com>

[ Upstream commit a242c36951ecd24bc16086940dbe6b522205c461 ]

In rdma_nl_rcv_skb(), the local variable err is assigned the return value
of the supplied callback function, which could be one of
ib_nl_handle_resolve_resp(), ib_nl_handle_set_timeout(), or
ib_nl_handle_ip_res_resp(). These three functions all return skb->len on
success.

rdma_nl_rcv_skb() is merely a copy of netlink_rcv_skb(). The callback
functions used by the latter have the convention: "Returns 0 on success or
a negative error code".

In particular, the statement (equal for both functions):

   if (nlh->nlmsg_flags & NLM_F_ACK || err)

implies that rdma_nl_rcv_skb() always will ack a message, independent of
the NLM_F_ACK being set in nlmsg_flags or not.

The fix could be to change the above statement, but it is better to keep
the two *_rcv_skb() functions equal in this respect and instead change the
three callback functions in the rdma subsystem to the correct convention.

Fixes: 2ca546b92a02 ("IB/sa: Route SA pathrecord query through netlink")
Fixes: ae43f8286730 ("IB/core: Add IP to GID netlink offload")
Link: https://lore.kernel.org/r/20191216120436.3204814-1-haakon.bugge@oracle.com
Suggested-by: Mark Haywood <mark.haywood@oracle.com>
Signed-off-by: Håkon Bugge <haakon.bugge@oracle.com>
Tested-by: Mark Haywood <mark.haywood@oracle.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Reviewed-by: Jason Gunthorpe <jgg@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
---
 drivers/infiniband/core/addr.c     | 2 +-
 drivers/infiniband/core/sa_query.c | 4 ++--
 2 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/drivers/infiniband/core/addr.c b/drivers/infiniband/core/addr.c
index 9e5f70e7122a0..6e96a2fb97dc4 100644
--- a/drivers/infiniband/core/addr.c
+++ b/drivers/infiniband/core/addr.c
@@ -136,7 +136,7 @@ int ib_nl_handle_ip_res_resp(struct sk_buff *skb,
 	if (ib_nl_is_good_ip_resp(nlh))
 		ib_nl_process_good_ip_rsep(nlh);
 
-	return skb->len;
+	return 0;
 }
 
 static int ib_nl_ip_send_msg(struct rdma_dev_addr *dev_addr,
diff --git a/drivers/infiniband/core/sa_query.c b/drivers/infiniband/core/sa_query.c
index 8be082edf986f..9881e6fa9fe4d 100644
--- a/drivers/infiniband/core/sa_query.c
+++ b/drivers/infiniband/core/sa_query.c
@@ -1078,7 +1078,7 @@ int ib_nl_handle_set_timeout(struct sk_buff *skb,
 	}
 
 settimeout_out:
-	return skb->len;
+	return 0;
 }
 
 static inline int ib_nl_is_good_resolve_resp(const struct nlmsghdr *nlh)
@@ -1149,7 +1149,7 @@ int ib_nl_handle_resolve_resp(struct sk_buff *skb,
 	}
 
 resp_out:
-	return skb->len;
+	return 0;
 }
 
 static void free_sm_ah(struct kref *kref)
-- 
2.20.1

[PATCH AUTOSEL 4.19 126/252] RDMA/rxe: Fix error type of mmap_offset

[Sasha Levin]

From: Jiewei Ke <kejiewei.cn@gmail.com>

[ Upstream commit 6ca18d8927d468c763571f78c9a7387a69ffa020 ]

The type of mmap_offset should be u64 instead of int to match the type of
mminfo.offset. If otherwise, after we create several thousands of CQs, it
will run into overflow issues.

Link: https://lore.kernel.org/r/20191227113613.5020-1-kejiewei.cn@gmail.com
Signed-off-by: Jiewei Ke <kejiewei.cn@gmail.com>
Reviewed-by: Jason Gunthorpe <jgg@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
---
 drivers/infiniband/sw/rxe/rxe_verbs.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/infiniband/sw/rxe/rxe_verbs.h b/drivers/infiniband/sw/rxe/rxe_verbs.h
index 6a75f96b90962..b4e24362edbb0 100644
--- a/drivers/infiniband/sw/rxe/rxe_verbs.h
+++ b/drivers/infiniband/sw/rxe/rxe_verbs.h
@@ -407,7 +407,7 @@ struct rxe_dev {
 	struct list_head	pending_mmaps;
 
 	spinlock_t		mmap_offset_lock; /* guard mmap_offset */
-	int			mmap_offset;
+	u64			mmap_offset;
 
 	atomic64_t		stats_counters[RXE_NUM_OF_COUNTERS];
 
-- 
2.20.1

[PATCH AUTOSEL 4.19 134/252] mlx5: work around high stack usage with gcc

[Sasha Levin]

From: Arnd Bergmann <arnd@arndb.de>

[ Upstream commit 42ae1a5c76691928ed217c7e40269db27f5225e9 ]

In some configurations, gcc tries too hard to optimize this code:

drivers/net/ethernet/mellanox/mlx5/core/en_stats.c: In function 'mlx5e_grp_sw_update_stats':
drivers/net/ethernet/mellanox/mlx5/core/en_stats.c:302:1: error: the frame size of 1336 bytes is larger than 1024 bytes [-Werror=frame-larger-than=]

As was stated in the bug report, the reason is that gcc runs into a corner
case in the register allocator that is rather hard to fix in a good way.

As there is an easy way to work around it, just add a comment and the
barrier that stops gcc from trying to overoptimize the function.

Link: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92657
Cc: Adhemerval Zanella <adhemerval.zanella@linaro.org>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
---
 drivers/net/ethernet/mellanox/mlx5/core/en_stats.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_stats.c b/drivers/net/ethernet/mellanox/mlx5/core/en_stats.c
index 8255d797ea943..9a68dee588c1a 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_stats.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_stats.c
@@ -211,6 +211,9 @@ void mlx5e_grp_sw_update_stats(struct mlx5e_priv *priv)
 			s->tx_tls_resync_bytes	+= sq_stats->tls_resync_bytes;
 #endif
 			s->tx_cqes		+= sq_stats->cqes;
+
+			/* https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92657 */
+			barrier();
 		}
 	}
 
-- 
2.20.1

[PATCH AUTOSEL 4.19 145/252] RDMA/core: Fix locking in ib_uverbs_event_read

[Sasha Levin]

From: Jason Gunthorpe <jgg@mellanox.com>

[ Upstream commit 14e23bd6d22123f6f3b2747701fa6cd4c6d05873 ]

This should not be using ib_dev to test for disassociation, during
disassociation is_closed is set under lock and the waitq is triggered.

Instead check is_closed and be sure to re-obtain the lock to test the
value after the wait_event returns.

Fixes: 036b10635739 ("IB/uverbs: Enable device removal when there are active user space applications")
Link: https://lore.kernel.org/r/1578504126-9400-12-git-send-email-yishaih@mellanox.com
Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Reviewed-by: Håkon Bugge <haakon.bugge@oracle.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
---
 drivers/infiniband/core/uverbs_main.c | 32 ++++++++++++---------------
 1 file changed, 14 insertions(+), 18 deletions(-)

diff --git a/drivers/infiniband/core/uverbs_main.c b/drivers/infiniband/core/uverbs_main.c
index 357de3b4fdddf..5404717998b07 100644
--- a/drivers/infiniband/core/uverbs_main.c
+++ b/drivers/infiniband/core/uverbs_main.c
@@ -273,7 +273,6 @@ void ib_uverbs_release_file(struct kref *ref)
 }
 
 static ssize_t ib_uverbs_event_read(struct ib_uverbs_event_queue *ev_queue,
-				    struct ib_uverbs_file *uverbs_file,
 				    struct file *filp, char __user *buf,
 				    size_t count, loff_t *pos,
 				    size_t eventsz)
@@ -291,19 +290,16 @@ static ssize_t ib_uverbs_event_read(struct ib_uverbs_event_queue *ev_queue,
 
 		if (wait_event_interruptible(ev_queue->poll_wait,
 					     (!list_empty(&ev_queue->event_list) ||
-			/* The barriers built into wait_event_interruptible()
-			 * and wake_up() guarentee this will see the null set
-			 * without using RCU
-			 */
-					     !uverbs_file->device->ib_dev)))
+					      ev_queue->is_closed)))
 			return -ERESTARTSYS;
 
+		spin_lock_irq(&ev_queue->lock);
+
 		/* If device was disassociated and no event exists set an error */
-		if (list_empty(&ev_queue->event_list) &&
-		    !uverbs_file->device->ib_dev)
+		if (list_empty(&ev_queue->event_list) && ev_queue->is_closed) {
+			spin_unlock_irq(&ev_queue->lock);
 			return -EIO;
-
-		spin_lock_irq(&ev_queue->lock);
+		}
 	}
 
 	event = list_entry(ev_queue->event_list.next, struct ib_uverbs_event, list);
@@ -338,8 +334,7 @@ static ssize_t ib_uverbs_async_event_read(struct file *filp, char __user *buf,
 {
 	struct ib_uverbs_async_event_file *file = filp->private_data;
 
-	return ib_uverbs_event_read(&file->ev_queue, file->uverbs_file, filp,
-				    buf, count, pos,
+	return ib_uverbs_event_read(&file->ev_queue, filp, buf, count, pos,
 				    sizeof(struct ib_uverbs_async_event_desc));
 }
 
@@ -349,9 +344,8 @@ static ssize_t ib_uverbs_comp_event_read(struct file *filp, char __user *buf,
 	struct ib_uverbs_completion_event_file *comp_ev_file =
 		filp->private_data;
 
-	return ib_uverbs_event_read(&comp_ev_file->ev_queue,
-				    comp_ev_file->uobj.ufile, filp,
-				    buf, count, pos,
+	return ib_uverbs_event_read(&comp_ev_file->ev_queue, filp, buf, count,
+				    pos,
 				    sizeof(struct ib_uverbs_comp_event_desc));
 }
 
@@ -374,7 +368,9 @@ static __poll_t ib_uverbs_event_poll(struct ib_uverbs_event_queue *ev_queue,
 static __poll_t ib_uverbs_async_event_poll(struct file *filp,
 					       struct poll_table_struct *wait)
 {
-	return ib_uverbs_event_poll(filp->private_data, filp, wait);
+	struct ib_uverbs_async_event_file *file = filp->private_data;
+
+	return ib_uverbs_event_poll(&file->ev_queue, filp, wait);
 }
 
 static __poll_t ib_uverbs_comp_event_poll(struct file *filp,
@@ -388,9 +384,9 @@ static __poll_t ib_uverbs_comp_event_poll(struct file *filp,
 
 static int ib_uverbs_async_event_fasync(int fd, struct file *filp, int on)
 {
-	struct ib_uverbs_event_queue *ev_queue = filp->private_data;
+	struct ib_uverbs_async_event_file *file = filp->private_data;
 
-	return fasync_helper(fd, filp, on, &ev_queue->async_queue);
+	return fasync_helper(fd, filp, on, &file->ev_queue.async_queue);
 }
 
 static int ib_uverbs_comp_event_fasync(int fd, struct file *filp, int on)
-- 
2.20.1

[PATCH AUTOSEL 4.19 146/252] IB/hfi1: Add software counter for ctxt0 seq drop

[Sasha Levin]

From: Mike Marciniszyn <mike.marciniszyn@intel.com>

[ Upstream commit 5ffd048698ea5139743acd45e8ab388a683642b8 ]

All other code paths increment some form of drop counter.

This was missed in the original implementation.

Fixes: 82c2611daaf0 ("staging/rdma/hfi1: Handle packets with invalid RHF on context 0")
Link: https://lore.kernel.org/r/20200106134228.119356.96828.stgit@awfm-01.aw.intel.com
Reviewed-by: Kaike Wan <kaike.wan@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
---
 drivers/infiniband/hw/hfi1/chip.c   | 10 ++++++++++
 drivers/infiniband/hw/hfi1/chip.h   |  1 +
 drivers/infiniband/hw/hfi1/driver.c |  1 +
 drivers/infiniband/hw/hfi1/hfi.h    |  2 ++
 4 files changed, 14 insertions(+)

diff --git a/drivers/infiniband/hw/hfi1/chip.c b/drivers/infiniband/hw/hfi1/chip.c
index b09a4b1cf397b..1221faea75a68 100644
--- a/drivers/infiniband/hw/hfi1/chip.c
+++ b/drivers/infiniband/hw/hfi1/chip.c
@@ -1687,6 +1687,14 @@ static u64 access_sw_pio_drain(const struct cntr_entry *entry,
 	return dd->verbs_dev.n_piodrain;
 }
 
+static u64 access_sw_ctx0_seq_drop(const struct cntr_entry *entry,
+				   void *context, int vl, int mode, u64 data)
+{
+	struct hfi1_devdata *dd = context;
+
+	return dd->ctx0_seq_drop;
+}
+
 static u64 access_sw_vtx_wait(const struct cntr_entry *entry,
 			      void *context, int vl, int mode, u64 data)
 {
@@ -4247,6 +4255,8 @@ static struct cntr_entry dev_cntrs[DEV_CNTR_LAST] = {
 			    access_sw_cpu_intr),
 [C_SW_CPU_RCV_LIM] = CNTR_ELEM("RcvLimit", 0, 0, CNTR_NORMAL,
 			    access_sw_cpu_rcv_limit),
+[C_SW_CTX0_SEQ_DROP] = CNTR_ELEM("SeqDrop0", 0, 0, CNTR_NORMAL,
+			    access_sw_ctx0_seq_drop),
 [C_SW_VTX_WAIT] = CNTR_ELEM("vTxWait", 0, 0, CNTR_NORMAL,
 			    access_sw_vtx_wait),
 [C_SW_PIO_WAIT] = CNTR_ELEM("PioWait", 0, 0, CNTR_NORMAL,
diff --git a/drivers/infiniband/hw/hfi1/chip.h b/drivers/infiniband/hw/hfi1/chip.h
index 36b04d6300e54..c9a352d8a7e13 100644
--- a/drivers/infiniband/hw/hfi1/chip.h
+++ b/drivers/infiniband/hw/hfi1/chip.h
@@ -909,6 +909,7 @@ enum {
 	C_DC_PG_STS_TX_MBE_CNT,
 	C_SW_CPU_INTR,
 	C_SW_CPU_RCV_LIM,
+	C_SW_CTX0_SEQ_DROP,
 	C_SW_VTX_WAIT,
 	C_SW_PIO_WAIT,
 	C_SW_PIO_DRAIN,
diff --git a/drivers/infiniband/hw/hfi1/driver.c b/drivers/infiniband/hw/hfi1/driver.c
index d5277c23cba60..769e114567a03 100644
--- a/drivers/infiniband/hw/hfi1/driver.c
+++ b/drivers/infiniband/hw/hfi1/driver.c
@@ -734,6 +734,7 @@ static noinline int skip_rcv_packet(struct hfi1_packet *packet, int thread)
 {
 	int ret;
 
+	packet->rcd->dd->ctx0_seq_drop++;
 	/* Set up for the next packet */
 	packet->rhqoff += packet->rsize;
 	if (packet->rhqoff >= packet->maxcnt)
diff --git a/drivers/infiniband/hw/hfi1/hfi.h b/drivers/infiniband/hw/hfi1/hfi.h
index 232fc4b59a98c..59c133935e23a 100644
--- a/drivers/infiniband/hw/hfi1/hfi.h
+++ b/drivers/infiniband/hw/hfi1/hfi.h
@@ -1093,6 +1093,8 @@ struct hfi1_devdata {
 
 	char *boardname; /* human readable board info */
 
+	u64 ctx0_seq_drop;
+
 	/* reset value */
 	u64 z_int_counter;
 	u64 z_rcv_limit;
-- 
2.20.1

[PATCH AUTOSEL 4.19 157/252] RDMA/uverbs: Remove needs_kfree_rcu from uverbs_obj_type_class

[Sasha Levin]

From: Jason Gunthorpe <jgg@mellanox.com>

[ Upstream commit 8bdf9dd984c18375d1090ddeb1792511f619c5c1 ]

After device disassociation the uapi_objects are destroyed and freed,
however it is still possible that core code can be holding a kref on the
uobject. When it finally goes to uverbs_uobject_free() via the kref_put()
it can trigger a use-after-free on the uapi_object.

Since needs_kfree_rcu is a micro optimization that only benefits file
uobjects, just get rid of it. There is no harm in using kfree_rcu even if
it isn't required, and the number of involved objects is small.

Link: https://lore.kernel.org/r/20200113143306.GA28717@ziepe.ca
Signed-off-by: Michael Guralnik <michaelgur@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
---
 drivers/infiniband/core/rdma_core.c | 23 +----------------------
 include/rdma/uverbs_types.h         |  1 -
 2 files changed, 1 insertion(+), 23 deletions(-)

diff --git a/drivers/infiniband/core/rdma_core.c b/drivers/infiniband/core/rdma_core.c
index c4118bcd51035..c2c9bd72b350f 100644
--- a/drivers/infiniband/core/rdma_core.c
+++ b/drivers/infiniband/core/rdma_core.c
@@ -49,13 +49,7 @@ void uverbs_uobject_get(struct ib_uobject *uobject)
 
 static void uverbs_uobject_free(struct kref *ref)
 {
-	struct ib_uobject *uobj =
-		container_of(ref, struct ib_uobject, ref);
-
-	if (uobj->uapi_object->type_class->needs_kfree_rcu)
-		kfree_rcu(uobj, rcu);
-	else
-		kfree(uobj);
+	kfree_rcu(container_of(ref, struct ib_uobject, ref), rcu);
 }
 
 void uverbs_uobject_put(struct ib_uobject *uobject)
@@ -753,20 +747,6 @@ const struct uverbs_obj_type_class uverbs_idr_class = {
 	.lookup_put = lookup_put_idr_uobject,
 	.destroy_hw = destroy_hw_idr_uobject,
 	.remove_handle = remove_handle_idr_uobject,
-	/*
-	 * When we destroy an object, we first just lock it for WRITE and
-	 * actually DESTROY it in the finalize stage. So, the problematic
-	 * scenario is when we just started the finalize stage of the
-	 * destruction (nothing was executed yet). Now, the other thread
-	 * fetched the object for READ access, but it didn't lock it yet.
-	 * The DESTROY thread continues and starts destroying the object.
-	 * When the other thread continue - without the RCU, it would
-	 * access freed memory. However, the rcu_read_lock delays the free
-	 * until the rcu_read_lock of the READ operation quits. Since the
-	 * exclusive lock of the object is still taken by the DESTROY flow, the
-	 * READ operation will get -EBUSY and it'll just bail out.
-	 */
-	.needs_kfree_rcu = true,
 };
 EXPORT_SYMBOL(uverbs_idr_class);
 
@@ -954,7 +934,6 @@ const struct uverbs_obj_type_class uverbs_fd_class = {
 	.lookup_put = lookup_put_fd_uobject,
 	.destroy_hw = destroy_hw_fd_uobject,
 	.remove_handle = remove_handle_fd_uobject,
-	.needs_kfree_rcu = false,
 };
 EXPORT_SYMBOL(uverbs_fd_class);
 
diff --git a/include/rdma/uverbs_types.h b/include/rdma/uverbs_types.h
index acb1bfa3cc99a..f70155cc73979 100644
--- a/include/rdma/uverbs_types.h
+++ b/include/rdma/uverbs_types.h
@@ -97,7 +97,6 @@ struct uverbs_obj_type_class {
 	int __must_check (*destroy_hw)(struct ib_uobject *uobj,
 				       enum rdma_remove_reason why);
 	void (*remove_handle)(struct ib_uobject *uobj);
-	u8    needs_kfree_rcu;
 };
 
 struct uverbs_obj_type {
-- 
2.20.1

[PATCH AUTOSEL 4.19 183/252] RDMA/uverbs: Verify MR access flags

[Sasha Levin]

From: Michael Guralnik <michaelgur@mellanox.com>

[ Upstream commit ca95c1411198c2d87217c19d44571052cdc94725 ]

Verify that MR access flags that are passed from user are all supported
ones, otherwise an error is returned.

Fixes: 4fca03778351 ("IB/uverbs: Move ib_access_flags and ib_read_counters_flags to uapi")
Link: https://lore.kernel.org/r/1578506740-22188-6-git-send-email-yishaih@mellanox.com
Signed-off-by: Michael Guralnik <michaelgur@mellanox.com>
Signed-off-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
---
 include/rdma/ib_verbs.h | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/include/rdma/ib_verbs.h b/include/rdma/ib_verbs.h
index 54e4d1fd21f8f..874cd6e94093b 100644
--- a/include/rdma/ib_verbs.h
+++ b/include/rdma/ib_verbs.h
@@ -3864,6 +3864,9 @@ static inline int ib_check_mr_access(int flags)
 	    !(flags & IB_ACCESS_LOCAL_WRITE))
 		return -EINVAL;
 
+	if (flags & ~IB_ACCESS_SUPPORTED)
+		return -EINVAL;
+
 	return 0;
 }
 
-- 
2.20.1

[PATCH AUTOSEL 4.19 184/252] IB/mlx4: Fix memory leak in add_gid error flow

[Sasha Levin]

From: Jack Morgenstein <jackm@dev.mellanox.co.il>

[ Upstream commit eaad647e5cc27f7b46a27f3b85b14c4c8a64bffa ]

In procedure mlx4_ib_add_gid(), if the driver is unable to update the FW
gid table, there is a memory leak in the driver's copy of the gid table:
the gid entry's context buffer is not freed.

If such an error occurs, free the entry's context buffer, and mark the
entry as available (by setting its context pointer to NULL).

Fixes: e26be1bfef81 ("IB/mlx4: Implement ib_device callbacks")
Link: https://lore.kernel.org/r/20200115085050.73746-1-leon@kernel.org
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Reviewed-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
---
 drivers/infiniband/hw/mlx4/main.c | 20 ++++++++++++++++----
 1 file changed, 16 insertions(+), 4 deletions(-)

diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c
index 9386bb57b3d71..a19d3ad14dc37 100644
--- a/drivers/infiniband/hw/mlx4/main.c
+++ b/drivers/infiniband/hw/mlx4/main.c
@@ -246,6 +246,13 @@ static int mlx4_ib_update_gids(struct gid_entry *gids,
 	return mlx4_ib_update_gids_v1(gids, ibdev, port_num);
 }
 
+static void free_gid_entry(struct gid_entry *entry)
+{
+	memset(&entry->gid, 0, sizeof(entry->gid));
+	kfree(entry->ctx);
+	entry->ctx = NULL;
+}
+
 static int mlx4_ib_add_gid(const struct ib_gid_attr *attr, void **context)
 {
 	struct mlx4_ib_dev *ibdev = to_mdev(attr->device);
@@ -306,6 +313,8 @@ static int mlx4_ib_add_gid(const struct ib_gid_attr *attr, void **context)
 				     GFP_ATOMIC);
 		if (!gids) {
 			ret = -ENOMEM;
+			*context = NULL;
+			free_gid_entry(&port_gid_table->gids[free]);
 		} else {
 			for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
 				memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
@@ -317,6 +326,12 @@ static int mlx4_ib_add_gid(const struct ib_gid_attr *attr, void **context)
 
 	if (!ret && hw_update) {
 		ret = mlx4_ib_update_gids(gids, ibdev, attr->port_num);
+		if (ret) {
+			spin_lock_bh(&iboe->lock);
+			*context = NULL;
+			free_gid_entry(&port_gid_table->gids[free]);
+			spin_unlock_bh(&iboe->lock);
+		}
 		kfree(gids);
 	}
 
@@ -346,10 +361,7 @@ static int mlx4_ib_del_gid(const struct ib_gid_attr *attr, void **context)
 		if (!ctx->refcount) {
 			unsigned int real_index = ctx->real_index;
 
-			memset(&port_gid_table->gids[real_index].gid, 0,
-			       sizeof(port_gid_table->gids[real_index].gid));
-			kfree(port_gid_table->gids[real_index].ctx);
-			port_gid_table->gids[real_index].ctx = NULL;
+			free_gid_entry(&port_gid_table->gids[real_index]);
 			hw_update = 1;
 		}
 	}
-- 
2.20.1